初创公司尽调
尽调报告 AI infrastructure / sovereign cloud late-stage private 2026-06-25

Neysa

面向印度主权 AI 云基础设施,押注 GPU 容量扩张,并有财务赞助方支持的规模资本

Neysa 切入主权 AI 基础设施的抓手可信,也拿到了赞助方支持的规模化资本;但收入质量、利用率、利润率和债务结构缺少公开证据,目前约 $1.4B 的价格还撑不住。

封面要素

已宣布融资方案 01
1.2 USD billions [CV001]
公开估值锚点 02
1.4 USD billions [CV002]
截至目前已披露股权融资 03
650 USD millions [CI007]
融资公告时现网 GPU 04
1200 GPUs [CV005]
目标 GPU 部署 05
20000 GPUs-plus [CV005]
2026 年 2 月报道披露员工数 06
110 employees [CV007]

公司概况

Neysa 是一家总部位于孟买的 AI 加速云公司,由 Sharad Sanghi 和 Anindya Das 于 2023 年创立。公司把 Velocis 定位为印度主权 AI 云栈,面向受监管企业、初创公司、研究用户和公共部门工作负载,把 GPU 基础设施、编排、可观测性、推理和 AI 安全打包在一起。公开证据显示公司已有真实客户验证,并在 2026 年 2 月拿到 Blackstone 领投的大额融资;但支撑估值的经营记录仍不透明,因为收入、利润率、利用率和债务条款都未公开披露。

官网
neysa.ai
成立时间
2023-01-01
创始人
Sharad Sanghi, Anindya Das
创立地点
Mumbai, India
总部
Mumbai, India
产品
主权 AI 云基础设施,通过 Velocis 栈以 GPU-as-a-Service、AI Platform-as-a-Service 和推理服务销售,托管部署选项覆盖 VM、裸金属、Kubernetes、可观测性和 AI 安全。
客户
印度金融服务、科技、医疗、公共部门、研究机构和 AI 原生初创公司的工作负载;这些客户需要本地数据处理、更低延迟的 GPU 访问或托管 AI 基础设施。
商业模式
GPU 计算靠用量和承诺容量变现,并在印度托管基础设施上叠加平台、推理、可观测性和托管 AI 服务。
阶段
late-stage private
融资情况
$20M 种子轮、$30M Series A,以及 2026 年 2 月最高 $600M 股权融资加计划 $600M 债务融资;独立追踪机构称,剔除拟议债务部分后,迄今已披露股权融资约 $650M。
[CO001, CO003, CO005, CO006, CO007, CO009, CO010, CO011]

执行摘要

主要优势

  • 创始人与市场匹配度在印度 AI 基础设施里格外强,把 Netmagic 时期的数据中心执行经验与当下主权云需求接了起来。
  • 公开客户案例在成本、在线率和延迟上给出了具体价值,尤其适合受监管、运营强度高的工作负载。
  • 印度推动主权算力之际,Blackstone 背书融资明显增强了 Neysa 锁定 GPU 供给、建设本地 AI 产能的能力。

主要风险

  • 公开披露没有 ARR、毛利率、利用率、债务定价或优先权结构条款,$1.4B 估值很难承销。
  • Neysa 的经济性取决于能否从约 1,200 张在线 GPU 扩到 20,000 张以上,同时不被供给、电力或上架率短板卡住。
  • 超大规模云厂商和印度本土强对手会挤压价格;企业和政府采购周期又长且集中。

未决问题

  • 经审计收入、积压订单、毛利率、烧钱速度或现金 runway 数字均未公开。
  • 债务分层条款、抵押包、契约和再融资假设仍未披露。
  • 关于客户集中度、留存、合同期限和续约率的公开证据很浅。
  • 员工数、在线 GPU 库存和设施精确落地时间的公开数据相互冲突,需要管理层对齐解释。

目录

Chapter 01

01公司概览

1.1 身份、布局和产品模式

Neysa 把自己定位为 AI Acceleration Cloud 提供商,而不是应用层 AI 初创公司。公司公开材料称,业务是一层专为企业、初创公司、研究机构和公共部门组织打造的基础设施:客户想要本地 GPU 容量、编排和安全能力,又不想把多个供应商拼在一起。法律实体记录显示公司可能在 2022 年 12 月注册,但公司和多数媒体一致把 2023 年视为运营创立年份,因此本报告采用这一最可辩护的公开简称。 公司总部位于孟买,并公开列出孟买、班加罗尔和钦奈办公室。当前计算站点证据比办公室证据更窄:Sacra 明确提到孟买和 Bangalore 数据中心的预布线容量,公开报道则支持与 NTT Data 和 Telangana 政府规划 Hyderabad 集群。因此,本章把孟买视为已确认总部,把班加罗尔和钦奈视为已确认办公室,把 Hyderabad 视为已宣布的规模化项目,而非已经投运的容量。 产品层面,Velocis 是 Neysa 商业模式的核心操作系统。官方材料把它呈现为全栈环境,覆盖 GPU-as-a-Service、AI Platform-as-a-Service、推理端点、编排、可观测性和 AI/ML 安全。定价和产品页显示,公司已经在销售多种 GPU 系列和部署模式;这很关键,因为 Neysa 卖的不只是原始芯片,也在卖控制权、本地驻留和可预测经济性。Moneycontrol 将 Neysa 描述为印度本土 AWS/Azure 式替代方案,这与其定位方向一致,但 Neysa 的真实差异化比超大规模云厂商等价物更窄,也更偏基础设施。[CO001, CO002, CO003, CO004, CO005, CO006]

Neysa — 关键 KPI 快照
指标数值 / 状态日期置信度尽调缺口
创立 / 法律时间运营创立:2023;法律注册:2022-12-162023 / 2022-12-16索取公司注册证书和创立时间线备忘录,以核对法律与运营起点
总部Art Guild House, Phoenix Marketcity Kurla, Mumbai(总部地址)2026-06-25
办公室Mumbai、Bengaluru、Chennai2026-06-25确认哪些城市承载算力,哪些仅有商业 / 工程团队
估值2026 年 2 月公开报道约 ~$1.4B2026-02索取最终 term sheet,以确认企业价值与 post-money 口径
已融资股权种子轮、Series A 和 2026 年 2 月轮合计至少 $650M 股权2026-02核对 cap-table 数据库与公司公告,确认准确累计金额
计划债务Blackstone 轮附带最高 $600M 债务融资2026-02核验债务文件、期限、covenant,以及完整额度是否已关闭
已上线 GPU 数公开数字冲突:~1,200(TechCrunch)vs ~2,000(SiliconANGLE)2026-02索取逐站点的已安装、可计费和已利用 GPU 库存
计划 GPU 规模长期在印度部署 20,000+ 块 GPU;Hyderabad 项目另被报道为 25,000 块 GPU2026-02 / 2025-04澄清 25,000 块 Hyderabad 容量是取代还是超过 20,000 块公司计划
员工数110(TechCrunch 2026 年 2 月)至 122(Tracxn 2026 年 5 月)2026-02 至 2026-05索取当前花名册和按城市划分的招聘计划
客户 / 证据点报道称有 20+ 客户和试点;公开证据覆盖 IISc、Innoviti、TIFIN、ITQ、HDFC Bank、PhonePe、Juspay、Fractal 和 AI-native 初创公司2025-2026索取已签约 logo、收入集中度,以及生产 vs 试点拆分
收入 / ARR2026-06-25没有经验证的公开收入披露;索取 FY2025-FY2026 财务报表

收入为空表示本章没有验证到公开数字。已上线 GPU 和员工数两行有意保留来源冲突,而不是选择一个缺乏支持的精确数字。

[CO001, CO002, CO003, CO004, CO021, CO031]
FO002: Neysa — 公司快照逻辑

印度本土 AI 需求、资本、基础设施足迹和客户分群如何串起 Neysa 故事。

[CO003, CO004, CO006, CO009, CO030, CO033]

1.2 创始人、领导层和治理覆盖

公开创始团队很清晰:Sharad Sanghi 是联合创始人兼 CEO,Anindya Das 是联合创始人兼 CTO。对于印度基础设施初创公司,两人的创始人—市场匹配度异常强,因为二人都来自 Netmagic 和 NTT 生态,而该生态曾帮助定义印度数据中心和托管云容量。公司、投资人和媒体材料在解释为什么 Neysa 值得被托付资本密集型主权计算建设时,都大量依赖这段过往运营履历。 治理披露比创始人披露更薄。首页和 VCCircle 材料确认硅谷老将 BV Jagadeesh 担任董事长,Neysa 又在 2026 年 6 月宣布前 Wipro 和 YES Bank CIO Anup Purohit 担任战略顾问。这证明公司有外部指导,但没有给出完整董事会地图或 Blackstone 交易后的控制框架。Tracxn 的公开董事会片段过于不完整,不能作为治理记录依赖;本章也未能核实 Blackstone 取得多数控制权后,委员会构成、保留事项或少数股东保护条款。 尽调含义是,Neysa 看起来仍由创始人主导,并拥有可信的外部背书,但关键人物依赖也很明显。Sanghi 是资本形成、基础设施关系和对外叙事的中心;Das 则支撑技术可信度。这一阶段这可以是优势,但也意味着后期投资人应要求明确的治理文件,然后再假设交易后的控制环境已完全制度化。[CO010, CO011, CO012, CO013, CO014]

领导层与创始人表
人物角色背景创始人-市场匹配 / 职能覆盖关键人物依赖
Sharad Sanghi联合创始人兼 CEONetmagic 创始人;后来负责 NTT data-center 业务印度基础设施建设的主要融资人和外部运营者高 — 融资、生态访问和公司叙事都以其为核心
Anindya Das联合创始人兼 CTO前 Netmagic / NTT 基础设施负责人掌握云、网络和平台设计的技术可信度高 — 技术架构和基础设施执行集中在这里
BV Jagadeesh董事长Silicon Valley 基础设施企业家;公司网站称其为 NetScaler 创始 CEO增加外部基础设施模式识别能力和可信度中 — 外部监督有价值,但公开职责范围披露很少
Anup Purohit战略顾问前 Wipro、YES Bank、RBL Bank 等机构 CIO增强对企业买家的理解和受监管行业 go-to-market 匹配中 — 顾问角色帮助企业销售动作,但不控制日常运营

公开来源清楚识别了创始人、董事长和一名战略顾问,但没有披露完整的 post-Blackstone 董事会名单或委员会结构。

[CO010, CO011, CO012, CO013, CO014]

1.3 融资历史、估值和规模信号

Neysa 的资本故事推进得异常快。公司在 2024 年初完成 $20 million 种子轮,2024 年 10 月完成 $30 million Series A,随后在 2026 年 2 月宣布 Blackstone 领投交易,组合为最高 $600 million 股权融资,并计划再取得 $600 million 债务融资。公开数据库和新闻来源大体收敛到 2026 年 2 月轮次估值约 $1.4 billion,但各来源对 post-money 和 enterprise value 的口径并不总是一致。 投资人名单很关键,因为它混合了传统风险资本、战略基础设施可信度和后期资本。种子轮和 Series A 投资方包括 Z47、Nexus Venture Partners 和 NTTVC,Tracxn 还列出 Blume Ventures 和 Anchorage 等投资方。2026 年 2 月融资加入 Blackstone、Teachers’ Venture Growth、TVS Capital 和 360 ONE,公开报道称 Blackstone 获得多数股权。股权结构的演进说明,Neysa 在不到三年内从风险资本支持的产品成型,走向基础设施规模融资。 规模指标方向上强劲,但还没有完全对齐。官方和独立报道都认同 20,000+ GPU 扩张计划,但当前现网 GPU 数量存在差异:TechCrunch 报道约 1,200 张 GPU,SiliconANGLE 报道约 2,000 张。员工数也类似,TechCrunch 称 2026 年 2 月有 110 名员工,Tracxn 到 2026 年 5 月列出 122 名。这些差异并非致命,但足以说明后续章节不应在未经管理层确认的情况下传播单一精确规模数字。[CO015, CO016, CO017, CO018, CO019, CO020]

利益相关方或投资者图谱
利益相关方角色控制 / 经济重要性公开证据尽调要求
Blackstone2026 年 2 月领投方$600M 股权承诺后成为多数股东;未来治理的核心角色公司 PR、TechCrunch、VCCircle、ET索取股东协议、保留事项和董事会控制条款
Teachers’ Venture Growth / Ontario Teachers 关联投资方成长轮共同投资方参与 2026 年 2 月轮;机构背书公司 PR、TechCrunch、Tracxn确认出资规模和治理权利
TVS Capital成长轮共同投资方参与 2026 年 2 月轮;增加印度本土 PE 资本支持公司 PR、ET、Tracxn澄清持股比例和 follow-on 权利
360 ONE成长轮共同投资方参与 2026 年 2 月轮;印度本土财富 / 资管资本公司 PR、ET、Tracxn澄清持股规模和投资期限
Nexus Venture Partners多轮投资方从种子轮到 2026 年 2 月轮均参与;持续信念信号Series A 文章、VCCircle、Tracxn确认 pro-rata 参与和董事会观察权
Z47种子轮和 Series A 支持方与创始人网络和印度 venture 生态绑定的早期信念投资方种子轮 PR、Series A PR、Tracxn确认 Z47 在 Blackstone 轮后是否保留有意义持股
NTTVC种子轮和 Series A 支持方来自 data-center / telecom 生态的战略资本信号Series A 文章、种子轮 PR、Tracxn澄清商业合作权,以及 NTT Data 关系是否超出 venture capital
创始人 / 管理层运营领导层即使控制权转向 Blackstone 后,仍是执行核心公司材料、VCCircle、Tracxn索取创始人 vesting、二级流动性和留任方案细节

这张图谱聚焦已公开披露、具备资本、控制或执行重要性的利益相关方;具体 post-round 持股比例未从 primary cap table 验证。

[CO015, CO016, CO017, CO018, CO019, CO020]
FO003: Neysa — 规模和就绪度 KPI

截至运行日,公开可见的规模、部署就绪度和风险指标。

本图强调战略就绪度和风险,而不是重复完整快照表。

[CO021, CO020, CO033, CO036, CO037, CO027]

1.4 里程碑、客户验证和负面信号

Neysa 的公开里程碑密度足够高,使其看起来像一个运营中的平台,而不只是一个融资故事。公司从 2024 年种子轮出发,2024 年 7 月发布 Velocis,2024 年 10 月声称已有付费客户,2024 年 12 月宣布与 Data Science Wizards 的保险云合作,2025 年 4 月浮现 NTT Data-Telangana Hyderabad 计算集群计划,2026 年 5 月与 Pipeshift 推出印度境内推理,2026 年 6 月又加入 Anup Purohit 作为战略顾问。这一节奏支持一种判断:公司正在同时搭建基础设施和市场进入能力。 客户验证强于典型早期基础设施公司的公开足迹。公司案例研究展示了 Indian Institute of Science、Innoviti、TIFIN 和 ITQ 的部署,更广泛的网站材料提到 HDFC Bank、PhonePe、Juspay、Fractal、Navana.ai、Smallest.ai、Graylabs.ai 和 Arrowhead AI。跨这些例子看,Neysa 最可复制的角度不是通用云托管,而是主权且对性能敏感的 AI 工作负载;在这些场景里,数据驻留、延迟或成本透明度会让超大规模云厂商显得不那么有吸引力。 主要负面信号是经济性的,而不是法律层面的。Sacra 明确把超大规模云厂商价格竞争、监管变化和计划债务融资引入的杠杆列为重大风险。Fortune India 又补充了一条有用警示:Blackstone 交易前,Neysa 已把首批 $50 million 中约 $44 million 投入 GPU 基础设施,说明该商业模式能快速吞掉资本,因此高度依赖利用率纪律。简言之,公司有实质动能,但业务现在必须证明资本强度、本地需求增长和主权计算差异化能够转化为持久经济性。[CO025, CO026, CO027, CO028, CO029, CO030]

里程碑表
日期事件类型金额 / 状态参与方含义
2022-12-16Neysa Networks Private Limited 注册成立创立创始人;Maharashtra 法律实体提供 2023 年运营创立叙事之下的法律起点
2023-01Sanghi 和 Das 主导 Neysa 筹备阶段开始创立公司启动 / 成立Sharad Sanghi;Anindya Das标志主权 AI 基础设施论点的运营起点
2024-03 / 2024-04种子轮宣布融资$20MZ47、Nexus、NTTVC为初始平台和基础设施建设提供资金
2024-07Velocis 发布产品旗舰平台发布Neysa把公司从概念转成可销售的产品环境
2024-10-22Series A 宣布;提及付费客户融资$30MNTTVC、Z47、Nexus显示早期商业牵引和重复投资方支持
2024-12-04与 DSW 合作推出 Insurance AI Cloud合作已发布Neysa;Data Science Wizards将垂直解决方案策略扩展到保险
2025-04-18Hyderabad AI 数据中心 MOU 公开扩张Rs 10,500 crore / 400MW / 25,000 GPUs(报道)Neysa;NTT Data;Telangana 政府显示锚定印度规模主权计算容量的雄心
2026-02-16Blackstone 领投融资宣布融资$600M 股权 + 计划 $600M 债务;~$1.4B 估值(报道)Blackstone;Teachers’ Venture Growth;TVS Capital;360 ONE;Nexus 等投资方让 Neysa 进入基础设施规模资本化和多数控制区间
2026-05-27Pipeshift 合作推出印度境内推理产品生产基础设施已上线Neysa;Pipeshift将平台从训练和编排扩展到主权推理
2026-06-10Anup Purohit 加入担任战略顾问治理顾问已任命Neysa;Anup Purohit增加企业治理和受监管行业可信度
2026-06杠杆和竞争仍是持续负面主题负面未解决Sacra 行业分析债务、价格竞争和利用率风险现在与产品动能同样重要

日期采用公告或报道事件时间。最后一条负面行被纳入,是因为公开记录显示,即使没有离散的法律或运营危机,经济风险仍在持续。

[CO001, CO002, CO015, CO017, CO019, CO021]
FO001: Neysa — 战略建设时间线

从法人设立到主权推理上线,再到 Blackstone 推动扩张的战略拐点。

[CO002, CO014, CO015, CO017, CO019, CO021]

1.5 图表

Chapter 02

02市场分析

2.1 市场边界和市场结构

Neysa 并不争夺印度全部云支出。它的相关市场包括 GPU-as-a-service、AI 训练和推理集群、AI 平台基础设施、主权 / 公共计算池,以及在印度运行这些工作负载所需的托管、供电、制冷和编排层。它不包括通用 SaaS、横向企业软件、标准 CPU 云,以及大多数与高密度计算无关的非 AI 数据中心建设。这个边界很重要,因为广义公有云或数据中心总量会夸大 Neysa 的真实机会,除非它们被收窄到重视本地 GPU、数据驻留、性能调优或专属支持的工作负载。印度当前形成三层市场结构:IndiaAI 主导的公共主权计算层;Microsoft、AWS、Google 和 Oracle 等全球超大规模云厂商,销售区域云足迹加主权就绪控制;以及 Neysa、Yotta 等本地新型云 / 数据中心运营商,打包专用 GPU、国内托管和印度特定服务水平。战略替代模式也很清楚:买家可以留在通用超大规模云厂商上,自建私有集群,或使用承诺更低延迟、更强驻留姿态和更可预测 GPU 经济性的印度 AI 云。[CM001, CM005, CM006, CM008, CM011, CM015]

市场定义表
细分 / 类别纳入支出排除支出买方 / 付款方为什么对 Neysa 重要
主权 / 公共 AI 计算IndiaAI 计算容量、补贴 GPU 池、公共利益模型开发通用 e-governance 软件、非 AI IT 服务Government of India、邦级机构、公共 R&D为本土 GPU 提供商创造政策支持的需求底线
受监管企业 AI 基础设施专用 GPU 集群、推理端点、面向 BFSI/医疗/公共服务的安全 AI 托管与 AI 或 residency 无关的通用公有云消费CIO/CTO、CISO、受监管业务单元预算本地托管、数据保证和支持可以支撑 neo-cloud 采用
AI-native 初创公司和 GCC 负载训练、微调、推理、MLOps、AI 平台运营商品化 dev/test CPU 云和通用 SaaS 工具初创公司创始人、GCC 工程负责人、产品预算灵活 GPU 容量和支持采用最快的细分
Hyperscaler / AI-lab 溢出与本地推理溢出集群、本地推理节点、主权就绪部署印度以外全球区域容量Hyperscaler、frontier lab、平台团队如果印度需求必须本地服务,可能成为渠道 / 客户细分
底层数字基础设施AI-ready 数据中心外壳、电力、散热、网络、编排传统低密度企业 colocation数据中心运营商、基础设施投资者必要的供给层,但比 Neysa 的直接可服务市场更宽

边界表把 AI-specific 基础设施与更广泛的云和数据中心支出分开,避免把相邻 TAM 误认为 Neysa 的直接市场。

[CM001, CM005, CM006, CM011, CM015, CM018]
FM004: 采用漏斗 / 价值链地图

Neysa 位于五步价值链中段:政策和本地化、数据中心就绪度、云 / GPU 层、部署工具,再到行业专属 AI 工作负载。

[CM011, CM015, CM027, CM036, CM040, CM049]

2.2 规模测算视角和 Neysa 可服务市场

公开来源没有给出一个干净、直接测量的印度 AI 云或 GPU 基础设施 TAM,因此本章采用多个视角。最宽的相邻市场是 IDC 对印度公有云到 2029 年达到 $30.4B 的预测,但其中包含许多 Neysa 永远不会触达的工作负载。更收窄的基础设施视角来自 Arizton 对印度数据中心市场的估算:2025 年 $9.79B,2031 年增至 $21.03B;Cushman 和 JLL 关于多吉瓦建设管线的证据也支撑这一视角,但这些数字仍包含通用容量。最有决策价值的底线是主权 / 公共计算:S&P 称,到 2025 年中,IndiaAI 招标已授出 34,371 张 GPU,补贴价格低至每 GPU 小时 $1.36;官方治理指南称,截至 2026 年 2 月,38,000+ 张 GPU 已接入;ETGovernment 称还在追加 20,000 张 GPU,使总数达到约 58,000。把这些已披露公共计算足迹年化,在私人企业需求之前,可得到约 $0.41B-$0.69B 的证据支撑底线。因此,Neysa 的 SAM 位于这个公共底线与更宽的云 / 数据中心相邻市场之间,集中在需要国内 GPU 集群而非通用云捆绑包的受监管企业、政府项目、GCC、初创公司和 AI 实验室。[CM003, CM004, CM020, CM021, CM023, CM024]

TAM / SAM / SOM 或规模测算视角表
视角 / 发布方年份地理范围数值CAGR / 增长方法置信度局限
IDC,经 CRN Asia 引用2026印度公有云(广义相邻市场)到 2029 年 $30.4B22.6% 年增长贸易媒体报道引用的广义公有云预测远宽于 AI/GPU 基础设施;包含许多 Neysa 不会服务的负载
Arizton2026印度数据中心市场$9.79B (2025) 至 $21.03B (2031)13.59% CAGR横跨 IT、电力、散热和建设的投资市场预测基础设施 capex 视角,不是 AI-cloud 收入或 GPU-specific 支出
Cushman & Wakefield2026印度数据中心容量1.6 GW 运营中;3.1 GW 在建 / 规划中管线扩张中运营和管线容量的全球市场比较容量视角,不是支出;包含通用云和企业数据中心
JLL2024印度 AI-ready 容量增量H2 2024 至 2026 年新增 604 MW;7.3M sq ft;$3.8B capex未来增量,不是 CAGR与 AI 集群相关的需求 / 管线分析仅限近期增量;不是完整市场估算
S&P Global 资料2025/2026印度主权 / 公共计算底线2025 年授予 34,371 块 GPU;到 2026 年 2 月上线 38,000+ 块;宣布扩容后约 ~58,000 块2025-2026 快速跃升招标授予、治理指南和已宣布新增容量GPU 数量只是看供给 / 采购的视角,不是 AI 云总收入口径
推导的公共算力年化下限2025-2026印度主权 / 公共算力按 $1.36/GPU-hour 计算,年化等值为 $0.41B-$0.69BN/A将 S&P 披露的 IndiaAI 底价套用到已授予、已上线或已宣布的 GPU 数量假设按补贴价格满负荷使用;不包括私营企业和高端专用容量

公开资料没有干净拆出印度 AI 云 / GPU TAM。因此,本章保留多种不可直接比较、但对决策有用的视角,并清楚标注每一种口径衡量的内容。

[CM003, CM004, CM023, CM024, CM026, CM028]
FM001: Neysa 在印度的可服务市场层级

可用视角从广义云邻近市场逐步收紧到 AI 专属本土算力:广义公有云、数据中心基础设施、主权算力底线,再到 Neysa 可直接服务的子集。

本图刻意混合邻近支出和容量代理,因为没有公开来源能清晰剥离印度 AI 云 / GPU TAM。层级标签直接说明计量基础。

[CM023, CM037, CM045, CM048]
FM002: 印度主权 GPU 足迹区间(千个 GPUs)

主权 / 公共算力的可支撑区间,来自已授予 GPUs、已接入 GPUs 和近期已公布扩张路径。

所有数值均以千个 GPUs 表示,指主权 / 公共算力足迹,而非印度商业 GPU 总容量。

[CM004, CM015, CM020, CM045, CM046, CM047]

2.3 买家、主权动态和采用路径

这个市场的预算归属很分散。政府需求来自 IndiaAI 本身、行业部委、公共机构和州级项目,它们需要补贴算力、模型开发和公益 AI 应用。银行、医疗和公共服务等受监管企业越来越重视本地数据托管、可审计性,以及把敏感模型训练工作流留在印度境内的能力。GCC 和 AI 原生初创公司构成第三类买家:它们未必拥有超大规模预算,但往往是专用 GPU 容量、推理端点和部署支持最快的采用者。超大规模云厂商和全球 AI 实验室在需要国内推理容量、溢出集群或主权就绪足迹时,也可能成为客户或渠道伙伴。因此,采用路径不是简单的“买家到云区域”;它常常从政策或合规触发开始,走向本地托管要求,再进入公共主权计算,或进入能提供专用 GPU 加动手集成的新型云运营商。Microsoft 明确向印度组织营销主权公有云和主权私有云产品,Neysa 和 Yotta 则主打印度优先的性能、合规和低摩擦部署。这一动态解释了为什么市场能同时容纳超大规模云厂商和本地专门厂商,而不是赢家通吃。[CM002, CM006, CM012, CM016, CM019, CM022]

细分市场 / 买方图谱
细分市场买方用户付费方工作流预算负责人采用触发因素
政府 / 公共算力IndiaAI Mission、部委、公共机构研究人员、创业公司、公共部门 AI 团队联邦 / 邦公共预算共享 GPU 访问、模型开发、公共利益 AIMeitY / 部委项目负责人以补贴方式使用本土算力,并获得本土模型支持
受监管 BFSI / 支付银行、金融科技公司、支付运营商AI/ML、反欺诈、客服、风控团队企业 IT 和业务单元预算推理、微调、安全训练、模型托管CIO/CTO/CISO 加受监管业务线RBI 数据本地化、可审计性、更低延迟、成本可预期
医疗 / 公共服务医院、健康科技公司、公共平台临床、运营、分析、公民服务团队企业或公共服务预算模型微调、推理、文档和语音工作负载CIO / 数字化转型负责人敏感数据、本土托管、政策对齐
GCC 和 AI 原生创业公司工程负责人、创始人、平台团队开发者、ML 工程师、产品团队R&D / 平台预算突发训练、快速实验、推理端点CTO / 工程副总裁需要快速部署、专属支持和印度本地性能
超大规模云厂商 / 全球实验室 / 大型企业云平台团队、AI 实验室、企业平台团队区域基础设施团队、应用负责人全球基础设施预算溢出集群、本地推理、主权就绪部署区域云 / 平台负责人印度用户密度、合规和境内延迟要求

在这个市场里,买方、用户和付费方并不相同;政策或合规触发因素往往决定客户是进入主权算力池,还是签私有新型云合同。

[CM002, CM006, CM012, CM016, CM019, CM022]
FM003: 买方 / 分群地图

采用路径从政策和合规触发出发,流向主权 / 公共算力池或印度专用 AI 云合同,再进入行业工作负载。

[CM006, CM012, CM016, CM019, CM022, CM039]

2.4 增长驱动和约束

需求正在上升,因为政策、资本和使用终于开始对齐。IndiaAI 降低了初创公司和研究者的进入成本;超大规模云厂商在增加主权就绪容量;Blackstone 对 Neysa 的融资和 Yotta 的扩张显示机构资本相信印度需要本土 AI 计算;数据中心开发商也在孟买、Hyderabad、钦奈、Pune、Delhi NCR,以及 Vizag 等新兴 AI 枢纽扩张。关键约束在于,已宣布容量不等于已经投运、能够产生收入的 AI 基础设施。Cushman、JLL 和 CRN 都描述了一个市场:执行能力、电力可得性、土地获取和热设计正变得与需求同样重要。S&P 还补充了更硬的物理瓶颈:数据中心电力需求可能从 2024 年的 13 TWh 升至 2030 年的 57 TWh;能源已占运营支出约 65%;水资源压力威胁主要城市集群;未来五年可能需要 15-30 GW 额外可再生能源容量。GPU 供应和利用率可见度也偏弱。对 Neysa 来说,机会是真实的,但转化速度取决于它能否比通用超大规模云厂商或印度 GPU 云竞争对手更快解决电力、制冷和部署瓶颈。[CM011, CM017, CM020, CM024, CM027, CM028]

增长驱动因素与约束表
驱动因素 / 约束方向时间含义尽调问题
IndiaAI Mission 和主权算力政策增长驱动2024-2026为本土算力提供商创造明确公共需求、补贴访问和合法性核实实际利用率、预留逻辑,以及政策容量转化为付费工作负载的比例
超大规模云厂商主权就绪建设增长驱动2025-2029证明印度是战略性 AI 区域,也让 AI 工作负载境内托管变得常态化跟踪超大规模云厂商定价是否压缩新型云的差异化空间
Blackstone/Yotta 资本承诺增长驱动2026 年起释放机构资金看好本土 AI 基础设施的信号,并加速供给建设确认硬件交付排期、融资条款和利用率假设
受监管行业的数据驻留和审计需求增长驱动当前推动 BFSI、医疗和政府买方转向本地 AI 部署模式识别哪些行业可在控制措施下使用超大规模云厂商,哪些必须使用专用本土集群
电力可用性和电网执行约束当前至 2030即使需求可见,也可能拖慢 AI 就绪园区,并推高运营成本按城市和园区绘制电力可用性,而不只看已宣布兆瓦数
用水和冷却强度约束当前至 2030高密度 AI 机架会在已经承压的城市市场带来用水和散热风险确认目标园区的冷却设计、水源和回收安排
GPU 供给和硬件交付周期约束当前限制已宣布需求转化为可用容量的速度验证分配优先级、供应商组合和进口 / 物流韧性
利用率不透明和合同价值不透明约束当前公开 GPU 数量无法说明付费利用率、预留组合或收入获取在 NDA 下索取管线、占用率和合同组合数据,以收紧 SAM/SOM 测算

上行情景不只取决于抽象的 AI 需求,更取决于资本、政策和硬件能否落成有电力、冷却和客户利用率支撑的运营集群。

[CM027, CM028, CM029, CM031, CM032, CM033]

2.5 图表

Chapter 03

03竞争格局

3.1 竞争格局概览

Neysa 进入的不是一片空白的印度 GPU 云市场。最接近的直接替代者是印度本土、印度托管的提供商,它们承诺不同组合的主权计算、AI 基础设施和受监管工作负载适配:Yotta 的 Shakti Cloud、E2E Networks、Tata Communications Vayu,以及 Sify CloudInfinit+AI 等相邻企业替代方案。它们外围还有第二圈超大规模云厂商——AWS、Azure 和 Google Cloud——这些厂商在印度仍然相关,因为它们已经掌握大型企业预算、广泛开发者生态和相邻托管服务。第三圈是 CoreWeave 等全球 AI 原生“新型云”,它们不是印度主权提供商,但代表印度提供商越来越需要对标的运营和经济性基准。 对 Neysa 来说,更重要的竞争问题不是“谁有 GPU?”,而是“谁能提供印度境内计算、采购简便性、托管 AI 工作流和企业信任的最佳组合?”按这个标准,市场已经分层。Yotta 强调原始主权规模和旗舰合作;E2E 强调自助服务和透明小时经济性;Tata 强调合规、网络覆盖和企业交付;超大规模云厂商强调广度;CoreWeave 则展示全球前沿规模下的 AI 原生基础设施形态。这意味着 Neysa 的战略竞争是多线作战:它必须在集成和安全上胜过本土对手,同时避免被超大规模云厂商锁定和 IndiaAI 补贴渠道扩张挤出。[CP001, CP006, CP011, CP020, CP022, CP024]

竞争对手概况表
竞争对手类别规模 / 证明点定价 / 打包方式最适合买方相对 Neysa 的战略短板
Neysa本土主权 AI 云官方定价页披露库存,包括 392 个 H100 SXM、200 个 H200 SXM,以及 L40S、L4 和 MI300销售主导的预留式打包;公司称定价全包,覆盖算力、存储、出站流量、K8s、笔记本环境、MLOps 和推理受监管企业、生产级 AI 团队、开放权重模型开发者原始资源规模看起来小于 Yotta 和超大规模云厂商;官方对比称,托管专有模型 API 不是重点
Yotta Shakti Cloud本土主权 AI 云官方网站称拥有 8,000+ 个 H100 GPU、NVIDIA Cloud Partner 身份和印度托管主权云已发布月度工作站和集群价格;平台访问单独销售政府、公共部门、企业和大型主权 AI 训练工作负载公开证据显示,其一体化 MLOps 和 AI 专用安全弱于 Neysa;部分商业层看起来是分开的
E2E NetworksGPU 优先的印度云MeitY 入围 GPU 云,发布 B200/H200/H100/L4 价格,并执行 IndiaAI 订单小时、月度和年度标价,定位自助采购创业公司、研究人员、突发工作负载、成本敏感型 AI 团队打包安全、可观测性或全栈企业控制的证据较少
Tata Communications Vayu企业主权 AI 云印度驻留主权云姿态,加上 GPU 云、连接能力和企业客户覆盖报价主导的 GPU 打包,宣称按需付费且无意外出站流量费大型企业、政府、受监管行业、既有 Tata 客户群公开定价透明度低于 E2E 或 Yotta,产品主导自助服务的公开证据也更少
AWS超大规模云厂商印度区域和本地区域;P5 H100 以及 P5e/P5en H200 系列由计算器和合同驱动,而不是印度专属主权 AI 套餐已在 AWS 服务上标准化的企业生态广度强,但印度优先的主权打包较弱
Azure超大规模云厂商ND 系列 GPU 加深厚的 Microsoft 企业版图按需付费 / 企业协议式采购以 Microsoft 为中心的企业和混合云买方数据驻留模型存在按服务划分的例外,不如本土“全在印度”营销口径干净
Google Cloud超大规模云厂商全球 43 个区域、130 个可用区,提供 AI 导向基础设施和按区域定价工具全球 SKU / 区域价格表,而不是主权 AI 套餐需要 Google AI 生态、全球路由和性价比工具的买方印度主权打包或简化受监管行业采购的证据较少
Sify CloudInfinit+AI本土相邻企业服务商基于印度数据中心足迹和企业 ICT 客户基础推出 GPUaaS按需付费 GPUaaS,定位训练、推理、分析、渲染和仿真既有 Sify 企业客户追加 GPU 容量一体化 MLOps / 安全深度和详细标价的公开证据较少
CoreWeave全球 AI 原生新型云对标专为 AI 打造的云,加上 SEC 文件披露的 AI 基础设施定位企业 / 合同式 AI 原生云经济模型前沿模型实验室和全球 AI 原生团队未看到印度主权托管、IndiaAI 渠道访问或本地公共部门姿态的已审阅证据

各行混合了官方厂商披露、技术文档、政府来源、申报文件和新闻。“战略短板”是分析归纳,不是直接引语,应理解为相对 Neysa 公开定位的差距。

[CP001, CP005, CP007, CP011, CP012, CP020]
FP001: 竞争定位图 — 主权契合度 vs. 托管 AI 栈深度

本土供应商集中在高主权契合度一侧,超大规模云厂商集中在高生态深度一侧;在主权阵营内,Neysa 公开切口显示出异常高的托管栈深度。

坐标轴是有证据支撑的序数评分,不是实测基准。X 轴估计在印度对主权 / 受监管工作负载的适配度;Y 轴估计公开可见的托管 AI 技术栈深度和生态完整度。

[CP001, CP007, CP011, CP020, CP023, CP024]

3.2 国内主权和印度托管替代方案

国内市场已按买家类型分化。Neysa 推出一体化全栈 AI 云,Yotta 把自己呈现为主权超大规模 GPU 资产,E2E 则是 GPU 优先、自助式平台,并有高度可见的标价。Tata Communications Vayu 更接近企业转型捆绑包——GPU 云加连接、主权云和托管服务层——Sify 则是相邻的企业 ICT 既有厂商,正在增加 GPU-as-a-service。IndiaAI 的官方分配系统很重要,因为它把公共政策转化为渠道力量:如果提供商已经入围、获得补贴并出现在国家计算门户上,就不必每个工作负载都靠直销赢下。 实际结果是,在买家更看重数据驻留、印度计费、采购简便性和部署支持,而不是最宽云目录的地方,Neysa 的国内竞争最激烈。Yotta 在规模和旗舰国家 AI 信号重要的地方看起来最强。E2E 对初创公司、研究团队和成本敏感的突发工作负载最有吸引力,因为其定价异常透明。Tata 在受监管行业信任、网络和账户覆盖重要的地方最强。Sify 作为既有企业供应商具备可信度,但就整合 MLOps 深度和公开 GPU 价格披露而言,已审阅公开证据比 Neysa、Yotta 或 E2E 更薄。[CP002, CP004, CP007, CP009, CP012, CP017]

定价和采购对比
服务商已发布价格 / 定价模式商业形态透明度 / 额外收费信号对 Neysa 的含义
Neysa预留式全包费率;公开页面展示 SKU 库存,但没有简单的一行式价目表销售主导的预留访问,覆盖裸金属、VM 和托管 Kubernetes公司称一档费率覆盖算力、存储、出站流量、K8s、Jupyter、MLflow、W&B 和推理端点支撑企业买方的预算可预期性,但公开价格透明度低于 E2E 或 Yotta 的价目页
YottaH100 AI Lab 工作站标价从 ₹27,000/month 到 ₹1,504,000/month;平台访问 ₹70,000/month月度工作站、VM、裸金属、集群和支持目录比只报价的企业云更透明,但平台访问是额外行项目适合已承诺投入的主权 AI 项目;Neysa 用一体化平台经济性反击
E2E NetworksB200 每小时 ₹624;H200 每小时 ₹300;H100 每小时 ₹249;L4 每小时 ₹49小时、月度或年度套餐,自助采购在已审阅的印度导向竞争对手中,公开价目表最明确;税费和存储 / 服务另计是 Neysa 获取创业公司和开发者客户时的强基准,尤其当团队偏好按需付费时
Tata Communications Vayu按需付费和承诺使用姿态;提供 H100/H200/L40S,但未公开价目表企业主导采购,绑定主权云和网络集成官方页面强调成本可预期、无意外出站流量费,但具体 GPU 标价仍在销售背后帮助 Tata 切入大客户,但公开横向基准比 E2E 或 Yotta 更难做
AWS通过文档和计算器按区域、实例和承诺期定价按需付费、预留和企业合同已审阅页面描述实例系列和区域,而非简单的印度 AI 套餐价目表超大规模云厂商广度很强,但只想要印度托管 GPU 栈的团队会觉得采购相对复杂
Azure跨区域和全球选项,按服务和协议定价企业协议和按需付费模式已审阅的数据驻留和 GPU 页面优先讲治理与能力,而不是简单公开 GPU 价目表最适合既有 Microsoft 体系,而不是想要简单主权 AI 套餐的买方
Google CloudSKU 和区域定价表,加选择器工具全球定价模型绑定工作负载、区域和折扣工具层面透明,但没有包装成印度主权 AI 商业套餐适合做全球优化的买方;对印度专属采购叙事的适配较弱
CoreWeave合同式企业定价;已审阅页面没有公开简单标价AI 原生平台围绕性能和运营效率销售比起静态标价,更强调运营透明度是 AI 原生经济模型的重要基准,但今天在印度境内不是最直接的采购替代品

保留来源中,只有 Yotta 和 E2E 提供容易复用的公开标价。Neysa、Tata 和超大规模云厂商要么描述定价理念、容量或工具,要么没有在已审阅页面给出干净的印度专属 GPU 横向价目表。

[CP004, CP009, CP012, CP013, CP025, CP028]

3.3 超大规模云厂商和全球新型云类比

即便 AWS、Azure 和 Google Cloud 不是最纯粹的主权云答案,它们在印度仍具战略相关性。AWS 拥有印度区域和本地区域,以及 H100/H200 支持的 P5 系列。Azure 拥有 ND 系列 GPU 基础设施和深厚企业渗透,但其自身数据驻留文档明确显示,部分 AI 和管理元数据可能因服务选择离开所选地理区域。Google Cloud 强调全球区域、性价比工具和具备数据驻留能力的区域,但已审阅页面仍把 AI 基础设施呈现为全球云目录的一部分,而不是专为印度主权 AI 打造的捆绑包。换言之,超大规模云厂商靠生态引力竞争,而不是靠印度优先包装。 CoreWeave 很重要,因为它是 AI 原生云专业化的全球基准。其官网和 S-1 描述了一朵专为 AI 打造的云,围绕裸金属、编排、可观测性,以及为模型训练和推理优化的存储 / 网络展开。这比 AWS、Azure 或 GCP 的通用云模式更接近印度主权 AI 云所追求的运营模式。但 CoreWeave 是类比对象,而不是直接的印度主权竞争对手:保留来源没有显示其参与 IndiaAI、提出印度特定数据驻留主张,或进行本地公共部门定位。对 Neysa 来说,CoreWeave 更像是产品架构和经济性的基准,而不是印度境内短期市场进入阻碍。[CP020, CP021, CP023, CP024, CP025, CP031]

能力、合规与上市路径矩阵
购买标准NeysaYottaE2ETata VayuAWSAzureGCPCoreWeave
印度主权托管信息部分部分部分无公开印度证据
已发布 GPU 标价部分部分部分部分部分
一体化 MLOps / AI 工作流叙事部分部分部分
公开强调 AI 专用安全层无公开证据无公开证据无公开证据通过更广生态部分覆盖通过更广生态部分覆盖通过更广生态部分覆盖部分平台安全
裸金属或专用大集群姿态部分部分
印度政府 / 受监管行业渠道证据无公开印度证据
企业分销和相邻服务广度部分部分很高很高很高
自助式开发者采购部分部分

矩阵单元格是基于官方产品页、文档和新闻综合出的证据支持型序位判断。“部分”通常表示能力存在,但没有被定位为核心差异点,或在保留来源中受例外条件限制。

[CP002, CP010, CP014, CP023, CP026, CP027]
FP002: 按竞争者类型划分的功能宽度与渠道力量图

Neysa 相比同业最强的公开差异,在于同时打出主权姿态、集成 MLOps 和 AI 安全叙事;超大规模云厂商胜在生态宽度,CoreWeave 胜在 AI 原生运营模式。

表格把多项属性压缩成是 / 部分 / 否或低 / 中 / 高判断,依据保留的公开证据,而非未披露客户访谈。

[CP002, CP010, CP012, CP023, CP027, CP029]

3.4 转换成本、防御性和竞争风险

Neysa 的防御性更可能来自工作流集成,而不是单纯 GPU 访问。通过 IndiaAI 入围、Yotta 规模化建设、E2E 透明自助目录和 Tata 主权企业平台,国内市场供应越来越充足。如果 GPU 供应扩大、补贴降低转换壁垒,简单的“印度托管计算”会更容易被复制。Neysa 更强的故事在于,买家可以在一份合同里一起采购计算、MLOps 和 AI 安全。对希望减少供应商、获得更清晰成本控制的受监管企业买家,这一点最重要。 即便如此,转换成本也有双向作用。已经投入 Azure ML 或更广泛 AWS 和 Google 体系的超大规模云客户,可能更愿意把模型开发留在现有合同、身份系统和平台工具内。国内云之间相对更容易多云并用,因为多家公司围绕 Kubernetes、集群式 GPU 访问或标准存储接口营销自己。但预留容量、支持流程、实验追踪习惯、安全控制和采购审批,在团队开始训练模型或把模型投入生产服务后,仍会形成真实粘性。因此,Neysa 面临的主要竞争风险不只是某个单一对手,而是一个拥挤市场:每种替代方案都占据买家决策栈的一部分——规模、价格透明度、主权信任、企业分销或生态广度。[CP036, CP037, CP038, CP039, CP040, CP041]

护城河耐久性和竞争风险登记表
护城河主张 / 威胁受益方严重性证据支持的理由对 Neysa 的含义
一体化 AI 栈加安全是真实切入口NeysaNeysa 是已审阅本土服务商中唯一公开把算力、MLOps 和 AI 安全作为一份合同来销售的公司如果买家更看重少用供应商、部署能满足监管,而不是只追最低裸 GPU 价格,这一路径就站得住
主权 GPU 裸供给越来越商品化Yotta、E2E、Tata、IndiaAI 云服务商IndiaAI 扩容叠加 Yotta 和 E2E 规模化建设,扩大了印度本土算力供给Neysa 不能只靠「印度托管 GPU」构筑护城河
自助式小时计价把初创公司拉向 E2EE2E Networks在已审阅的印度本土厂商中,E2E 的公开小时价目表最清晰Neysa 或许能赢下企业打包方案,但如果不简化入门计价,开发者主导的份额会流失
企业采购和连接能力偏向 TataTata CommunicationsTata 把主权云、GPU 云和网络集成一起卖进受监管客户Neysa 必须证明,产品集成的价值能压过 Tata 的客户控制力和信任资产
超大规模云厂商的生态锁定仍然很强AWS / Azure / GCP更广的云合同、平台工具和相邻服务,抬高了既有客户的迁出成本Neysa 需要讲清迁移路径,而不只是把 GPU 账单压低
Azure 地理例外削弱了部分 AI 工作负载的「纯主权」叙事印度本土主权厂商Azure 文档列出若干情形:元数据或模型处理可能发生在所选地理区域之外在受监管用例里,这给 Neysa、Tata、Yotta 和 E2E 打开了切入点
IndiaAI 补贴渠道扩大了竞争对手的分发面Yotta、E2E、NxtGen、Tata、Jio 等官方分配和分轮入围机制,在纯直销之外创造了需求流即使 Neysa 产品很强,渠道入口也能加速竞争对手被采用
全球 AI 原生云设定了性能预期CoreWeave 及全球同类CoreWeave 在编排、有效吞吐和运营模式上定义了 AI 原生标杆即便印度本土 GTM 保护 Neysa 的本地位置,产品架构也必须保持竞争力

严重性是分析判断。「高」表示即使 Neysa 的核心技术可靠,该威胁也能实质改变买家选择或 GTM 效率。

[CP036, CP037, CP038, CP040, CP041, CP042]
FP003: 护城河与就绪度 KPI

公开数据更支持 Yotta 的主权规模、E2E 的价格透明度,以及 Tata 的企业成本 / 主权叙事;Neysa 披露的库存有意义,但小于最大的主权建设项目。

条目混合了库存、公开价格点和项目规模指标。单位未归一化,因为目的在于判断竞争就绪度,而不是做单一算术比较。

[CP005, CP007, CP009, CP012, CP016, CP018]

3.5 图表

Chapter 04

04财务

4.1 收入模式、定价和变现层

Neysa 的公开界面指向一种比简单现货 GPU 租赁更宽的收入模式,但根基仍是出售计算容量。公司公开列出三种核心产品形态——托管 VM 实例、裸金属 GPU 服务器和托管 Kubernetes 集群——再叠加公有、私有和混合部署选项。这在财务上很关键,因为价格歧视逻辑清晰可见:按需工作负载把实验和突发需求变现,预留或承诺容量则用折扣换取更好的前瞻可见度。同一价格页还去掉了几类常见云附加费,包括 ingress、egress 和推理交易费,这支撑了 Neysa 不仅按原始峰值性能竞争,也按账单可预测性竞争的叙事。 更有意思的问题是,Neysa 能否在商品化计算之上挂上更高质量收入。答案是“很可能可以,但公开投资人仍无法量化”。Velocis 打包编排、AI Studio、可观测性、推理端点,以及 Aegis 等安全功能;Pipeshift 合作又把产品扩展到专用托管推理环境。这些功能可能提高转换成本,并打开平台或服务变现空间。不过,Neysa 没有披露多少收入来自经常性预留容量,多少来自专业服务,也没有说明托管推理是按高溢价软件层计费,还是只是转嫁计算成本的包装。公开层面,公司像一家 B2B GPU 云和主权 AI 平台,拥有多条变现杠杆;私下层面,缺失的收入结构数据是第一个重大承保障碍。[CI009, CI010, CI011, CI012, CI013, CI014]

收入流表
收入流机制单位当前公开状态收入质量尽调问题
按需 GPU 实例在共享云上按小时消耗 VM 或容器每 GPU-hour 的 USDL4/L40S/H100/H200 公开标价已上线转化快,但可能波动最大,也最受利用率影响需要按 SKU 拆分的实际综合费率和占用率
预留容量1 至 36 个月承诺方案,附带折扣月度或期限承诺已公开承诺框架和折扣说法如果有最低承诺和续约行为支撑,质量更高需要已签 ARR/MRR、取消权和续约率
裸金属专用集群单租户 8-GPU 节点及更大的专用环境月度节点费或集群费公开裸金属 SKU 和客户证据可见ACV 可能更高、流失更低,但资本开支重需要最低合同期限,以及按集群类型拆分的毛利率
托管推理端点通过 Velocis 和 Pipeshift 提供单租户或专用推理 API合同费或用量费产品已公开存在,但独立定价未披露可能比裸算力更黏,也更像软件收入需要定价指标、挂载率和毛利拆分
平台 / MLOps / 安全工具AI Studio、编排、可观测性和 Aegis 控制订阅、打包或服务能力已公开,变现形式未公开如果单独收费,可能提升留存和综合毛利需要单独收入科目或挂载率披露
解决方案工程 / 引导上线部署协助、调优和工作负载优化项目费或服务费案例研究能看出支持强度,但看不到定价如果不是经常性收入,质量最低;只有作为先落地再扩张的入口才有价值需要服务收入占比和毛利率

仅使用公开产品证据;实际收入组合、折扣和挂载率仍未披露。

[CI009, CI011, CI012, CI013, CI015, CI016]
定价 / 变现表
SKU / 模型公开标价承诺信号部署形态含义
L4 24GB$1.17 / hour36 个月预留为 $428.37 / 月托管 VM / 容器入门级推理或实验通道
L40S 48GB$1.95 / hour36 个月预留为 $713.96 / 月托管 VM / 容器中档训练或推理价格点
H100 SXM 80GB$4.39 / hour36 个月预留为 $1,779.96 / 月托管 VM / 容器有公开标价锚点的高性能训练 SKU
H200 SXM 141GB$4.73 / hour36 个月预留为 $1,866.78 / 月托管 VM / 容器展示了最新公开高端 GPU 标价
8x L40S 裸金属$4,306.62 / month偏承诺的月度节点单租户裸金属显示定价从纯公用事业式收费,转向专用容量合同
8x H100 裸金属$12,433.64 / month偏承诺的月度节点单租户裸金属更高 ACV 形态,适合有承诺的企业工作负载
8x H200 裸金属$13,822.86 / month偏承诺的月度节点单租户裸金属显示高端硬件靠更大的专用节点变现

标价不等于实际成交价;企业折扣、最低承诺和打包软件带来的上浮仍未知。

[CI010, CI011, CI012, CI013]
FI001: 收入模型桥

公开证据显示,Neysa 的变现路径是分层推进:先吃 GPU 消耗,再转向预留容量和托管 AI 服务。

流程是概念性但有来源支撑;Neysa 未按产品线披露实际构成占比或收入确认政策。

[CI009, CI011, CI015, CI016, CI017, CI048]

4.2 单位经济性代理和服务交付成本逻辑

Neysa 不公布毛利率、贡献利润率、获客成本或回本期。因此,最好的公开解读来自客户验证,以及硬件供给所暗示的成本架构。TIFIN、Innoviti 和 ITQ 三个客户案例都用同一种语言描述 Neysa 的价值:比超大规模云厂商或通用云支出更低、延迟更好,并为受监管或高容量工作负载提供更强控制。节省幅度从 40% 到 65%,运营指标包括 99.95% uptime、低于 2 秒的 P99 延迟、每秒 2,500 tokens、低于 30 秒的处理循环,以及快速生产部署。这些不是经审计财务指标,但确实表明 Neysa 正在赢得一些工作负载;在这些场景里,专用或预留基础设施比按 token 计价的 API 使用更能摊销固定成本。 但前提是占用率必须保持高位。Neysa 的价格表间接暴露了成本基础:H100 和 H200 集群、重 NVMe 节点、高带宽互连,以及支持密集的托管环境。CoreWeave 的 S-1 提供了一个有用的公开可比,解释这套经济逻辑。它显示,即使规模化的新型云,如果利用率、集中度或融资假设落空,也可能在增长中亏钱;文件还明确称,由于系统低效,现实世界 AI 基础设施往往只能交付理论峰值输出的 35% 到 45%。对 Neysa 来说,含义很直接:预留集群、私有部署和长期受监管工作负载不是可有可无的追加销售,而是把快速折旧硬件转化为可行利润率路径的经济机制。[CI014, CI020, CI021, CI022, CI023, CI024]

单位经济表
指标公开数值 / 代理指标置信度重要性尽调问题
公司 TCO 主张比超大规模云厂商低 40% 至 60%界定了 Neysa 的定价承诺,但没有独立审计需要同口径基准测试方法
TIFIN 支出节省GPU 云支出比超大规模云厂商低 65%暗示受监管企业的预留工作负载可以高效变现需要工作负载范围,以及前后计算量
Innoviti TCO 节省TCO 比通用云低 60%支撑数据敏感型生产推理的经济性需要合同期限和准确工作负载画像
ITQ TCO 节省TCO 比通用云低 40%说明在极高 token 量下,经济性也能跑通需要完整成本桥,包括支持和模型调优
延迟 / 吞吐代理指标ITQ 的 P99 低于 2s,约 2,500 tokens/s性能好有助于守住实际成交价和客户留存需要这些基准测试期间的利用率
可靠性代理指标TIFIN 的可用性为 99.95%宕机会直接伤害已承诺企业集群的变现需要 SLA 定义和服务积分结构
公开利用率空值 / 未披露对资本开支重的 GPU 云来说,利用率是核心毛利驱动项需要按 SKU 拆分的占用率,以及预留与现货组合
公开毛利率空值 / 未披露没有毛利率,就无法可靠建模回本或偿债能力需要产品线毛利率和折旧政策
公开烧钱 / 现金续航空值 / 未披露对下一轮融资时点和契约韧性至关重要需要月度烧钱、现金,以及未提取债务 / 股权额度

客户案例只是方向性代理指标,不是经审计的单位经济明细;空值是有意保留的证据缺口。

[CI014, CI020, CI021, CI022, CI023, CI024]
FI002: 单位经济桥

公开毛利叙事不主要取决于标价,而取决于能否让专用基础设施被承诺工作负载高强度占用。

Neysa 披露了客户结果代理指标,但未披露毛利率或占用率,因此该桥展示因果逻辑,而不是量化的公司特定利润率。

[CI020, CI021, CI022, CI023, CI024, CI036]

4.3 资本结构、资本开支强度和债务含义

Neysa 的 2026 年融资方案是最清晰的公开信号:这首先是一家基础设施公司,其次才是一家软件公司。已宣布结构为最高 $600 million 股权加拟议 $600 million 债务,相对于公司年龄和此前融资基础都很大,并且明确绑定到扩张至超过 20,000 张 GPU。即便没有管理层对精确资本开支的指引,一个简单启发式也能显示下一阶段多么吃资本。如果最终只有计划债务用于硬件建设,这个资本池已意味着每张目标 GPU 约 $30,000;如果全套资金都投入建设,在计入数据中心改造、网络、软件或营运资本之前,总资金池升至每张目标 GPU 约 $60,000。这就是债务结构与标题估值同样重要的原因。 公开层面,债务是故事中最不透明的部分。NewsBytes 称,印度 GPU 抵押贷款可按 GPU 价值融资 40% 到 70%,利率最高 14%;只有当贷款人相信利用率和合同期限会保持高位时,这才可行。外部负面研究进一步凸显风险。Compute Forecast 认为,短缺时期的 GPU 债务往往按已经压缩的租赁假设承销;CNBC 则强调,当贷款人围绕较短寿命计算资产为长期设施融资时,会产生“GPU debt treadmill”。CoreWeave 的文件显示,即使对规模化运营商,债务堆栈也可能变得很大。Neysa 最终可能比风险叙事更保守,但在披露期限、抵押品和合同收入覆盖之前,投资人承保的是一场带有部分隐藏债务工具的主权 AI 扩张。[CI001, CI002, CI003, CI004, CI005, CI007]

资本充足性表
指标公开信号含义置信度尽调问题
已承诺股权最高 $600M相比此前融资,股权垫很厚,但仍绑定重资产扩建确认资金交割时间和分阶段提款安排
计划债务计划另增 $600M,取决于文件签署相比 2024 年纯风险投资资本结构,债务会实质改变风险画像需要贷款方名单、期限、票息和契约组合
目标机队>20,000 GPUs 计划在印度部署显示固定资产基数和对应折旧会大幅上台阶需要按 GPU 代际和站点拆分的部署节奏
当前机队代理指标公开报道约 1,200 块上线 GPU;其他引用暗示约 2,000 块已安装基数只是近似值,因此扩张倍数仍然模糊索取按日期和 SKU 拆分的机队清单
每块目标 GPU 的示意资本池$30k 至 $60k说明占用率和合同期限为什么比表面估值更重要需要真实 capex 预算,包括非 GPU 基础设施
印度 GPU 债务市场代理指标40% 至 70% LTV,利率最高 14%暗示如果利用率落后,债务服务成本会变贵需要 Neysa 特定的 LTV、债务成本和摊还安排
现金 / 烧钱 / 现金续航未公开披露仅凭公开证据无法判断现金续航索取月度现金余额、烧钱和预测现金续航
收入增长目标TechCrunch 称收入目标是明年增长超过三倍这是正向需求信号,但没有起始基数,无法支撑投资决策需要月度收入基数和支持该目标的已签积压订单

各行混合了已披露事实和明确启发式估算;除规模外,所有债务结构字段仍未披露。

[CI001, CI002, CI005, CI031, CI032, CI033]
FI004: 资本强度 / 现金流图

Neysa 的建设把风险集中在硬件资本开支、非 GPU 基础设施,以及未来压在两者之上的债务服务。

矩阵把已披露的战略动作转成现金流桶;只有债务和 GPU 扩张头条是直接公开的。

[CI002, CI004, CI030, CI032, CI037, CI038]

4.4 披露缺口和承保结论

Neysa 的公开披露问题不在于缺乏需求证据,而在于几乎所有决策级财务证据都缺失。收入是最好的例子。Tofler 和 Tracxn 都给出收入区间,但一个称 ₹10 crore 到 ₹25 crore,另一个称 ₹10 crore 到 ₹50 crore;二者都不能替代经审计报表、经常性收入结构或 backlog。现金、烧钱速度、毛利率、利用率、客户集中度和债务服务覆盖也一样:本章可以推断商业模式,但无法仅凭公开来源完整承保公司。这也是财务分析必须严守 null 的原因。已审阅证据没有可信基础来编造 ARR、runway、毛利率或净留存。 因此,底线判断是混合但自洽的。收入质量看起来好于纯现货市场,因为 Neysa 似乎在向重视主权、延迟和支持的受监管企业销售专用容量、私有部署和托管推理。这应让最终收入基底比商品化突发计算更有粘性。与此同时,资本开支强度显然很高,新增债务层提高了对利用率、定价权或硬件淘汰判断出错的成本。换句话说:商业模式在战略上可行,也赶上商业窗口,但公开记录仍支撑的是一个依赖融资的基础设施命题,而不是一个已透明盈利的软件命题。任何严肃尽调都应从把本章的 null 转成董事会级数字开始。[CI040, CI041, CI042, CI043, CI044, CI046]

公开财务缺口表
缺失指标重要性当前公开代理指标精确尽调路径
收入 / ARR / 经常性组合估值、债务规模和收入质量判断都需要它只有宽泛第三方收入区间获取经审计报表,并按产品线搭建月度收入桥
毛利率 / 贡献毛利率决定客户节省成本的说法是否仍能留下有吸引力的单位经济无公开披露索取计算、用电、支持和折旧的 COGS 拆分
现金、烧钱和现金续航决定下一轮融资时点和契约韧性无公开披露索取当前现金、预测烧钱和未提取资本额度
利用率 / 预留组合 / 续约占用率是 capex 回本和偿债覆盖的核心驱动项案例研究暗示生产使用,但没有组合层面的指标索取按 SKU 拆分的占用率、预留与按需组合,以及续约队列
债务期限、票息、抵押、契约新债务批次是下行风险中最大的摆动因素只有规模公开索取已签债务条款清单和董事会融资备忘录
客户集中度 / 积压订单长期合同决定扩建由需求托底,还是带有投机性只有公共部门引用和客户标识索取头部客户集中度、积压订单和合同期限数据

空值是有意保留的:在公开证据缺失处,本章不编造 ARR、烧钱、利润率或积压订单。

[CI040, CI041, CI042, CI043, CI047, CI048]
FI003: 财务估计区间

公开数据只能支撑有边界的区间和启发式估计,不能支撑精确财务承保。

区间混合了第三方收入区间、客户验证差异、债务市场代理和简单的每目标 GPU 融资启发式;任何一项都不应视作经审计指引。

[CI032, CI033, CI041, CI042, CI043]

4.5 图表

Chapter 05

05产品与技术

5.1 按客户工作流理解的产品表面

Neysa 把 Velocis 营销为全栈 AI 加速云,而不是单一 GPU 租赁页面。首页和产品页持续呈现三条买家旅程:配置主权 GPU 容量,在预集成环境中构建或微调模型,并在已附带治理和可观测性的情况下发布生产推理。这个定位很重要,因为公开材料没有停在原始计算上。它们把集成 MLOps、模型注册表、实验追踪、推理端点、集中式仪表盘和安全控制都列为商业表面的一部分。平台取向也明确是 open-source-first,强调 Jupyter、PyTorch、Hugging Face、MLflow、Kubeflow、Git 和容器工作流,而不是封闭专有模型栈。定价证据强化了同一点。Neysa 发布了分数和完整 H100 配置,但把这些条目与可预测 TCO、公有或私有部署选项和托管支持主张配在一起。首页和案例研究中的客户引语进一步说明,实际待办任务是在印度境内把 AI 原型推进到生产,同时不放弃数据驻留、延迟控制或成本可预测性。[CE001, CE002, CE003, CE018, CE020, CE021]

产品模块 / 资产矩阵
模块 / 资产主要用户状态 / 成熟度有证据支撑的能力差异化尽调缺口
GPUaaS 计算池基础设施和 ML 团队已上线 / 公开定价按需和承诺 H100 容量,包括分片和 full-SXM 选项主权、印度优先定位,且公开挂牌价格透明按 SKU、区域、裸金属与 VM 拆分的准确上线装机基数未披露
aiPaaS 生命周期控制平面ML 工程师和平台团队已上线 / 正在积极推广集成环境覆盖训练、微调、模型注册、实验跟踪、监控和 CI/CD 集成试图消除工具链蔓延,而不是让买家自己拼 MLOps相比营销深度,公开 API 和控制台文档仍然偏薄
托管推理端点应用和产品团队已上线 / 正在扩展为开源和定制模型提供专用及托管推理,带自动扩缩容和安全层把实验桥接到生产部署,且运维负担更低公开 SLA 细节和基准披露仍有限
安全和治理层安全、平台和合规团队已上线 / 有外部认证RBAC、SSO、审计日志、加密、BYOK、KMS 集成和零信任访问主权 AI 基础设施,叠合买家可见的治理主张SOC 2 范围和控制映射并未完全公开
客户部署证据BFSI、零售、研究和语音 AI 买家已上线 / 存在客户证据TIFIN、Innoviti、IISc、Nurix、Navana 和 Smallest 均被引用为工作负载类比或客户证据覆盖受监管 BFSI、零售现场运营、学术训练和实时语音 AI引用数量仍不多,且大多由公司筛选
Marketplace 生态ISV 和解决方案伙伴路线图 / 即将推出集成进 Velocis 的精选 AI 原生应用、智能体和 SaaS 工具可能把 Neysa 从基础设施供应商推向更宽的生态平台正式可用日期、合作伙伴名单和变现模型均未公开

各行把已明确交付的界面和路线图占位符分开;marketplace 状态明确标为即将推出,而不是按正式可用处理。

[CE001, CE002, CE003, CE009, CE020, CE025]
工作流 / 用例表
用户任务当前工作流压力Neysa 解决方案界面可衡量收益已知限制
在印度境内训练或微调企业模型团队需要主权 GPU 资源,不能等进口算力,也不想自建集群专用或分片 GPU、预置 Jupyter 与框架环境,并支持私有或混合部署选项训练和微调可以更快拿到资源,数据也留在本地处理公开材料没有披露具体区域覆盖或排队时间统计
为时延敏感应用发布生产推理共享 API 会带来时延尖峰、冷启动和跨境路由Velocis 推理端点与 Pipeshift 合作提供单租户、兼容 OpenAI 的端点Nurix 称 TTFT 提升 3x,Arrowhead 一天内上线这些结果绑定具体客户,尚未归一成公开基准套件
运行受监管 BFSI AI 工作负载境外托管推理和波动的单位经济性会冲击合规与业务模型TIFIN 在印度境内用 Velocis 做训练、实验和推理案例研究称 GPU 云支出降低 65%,可用性达 99.95%证据来自公司撰写的案例研究,不是独立审计
自动化多模态零售支持核验共享黑盒 API 给不到足够的栈可见性,也难以保证时延确定性Innoviti 将 Qwen 3.0 VL 工作负载迁到 Neysa 专用推理基础设施案例研究称 TCO 降低 60%,时延低于 30 秒,自动核验准确率达 96%没有发布原始基准测试方法
用高显存算力支撑研究级训练学术实验室在共享集群上常遇到排队争用,显存余量也不够IISc 用专用裸金属 GPU 节点做合成数据生成、训练和评估据称,团队训练了 3300 万组草图-图像对和两个开放权重模型,且没有缩小设计规模公开案例研究没有说明具体 GPU 数量或总训练时长

收益来自已发布案例研究或合作伙伴说法,应视为方向性证据点,而不是可普遍迁移的基准。

[CE002, CE018, CE019, CE027, CE031, CE035]
FE002: 客户工作流 / 运营流程

公开的 Neysa 工作流从主权和工作负载选择开始,随后进入配置、构建或微调、治理和生产推理。

[CE002, CE003, CE007, CE018, CE027, CE031]

5.2 架构、编排和硬件栈

Neysa 架构页上有最具体的产品证据。Velocis 描述了一条从数据摄取到推理的端到端 ML 生命周期,并点名具体控制平面层:AI 集群管理、AI 调度器、资源管理器,以及以裸金属、虚拟机或容器交付 GPU 的支持。存储横跨对象、块和 NFS 模式;集成表面覆盖 GitHub 或 GitLab、Docker、MLflow、Kubeflow、Airflow、SIEM 工具、VPC 连接和企业 IAM。公开可观测性主张也具体到足以进入尽调:Neysa 称仪表盘追踪 GPU 利用率、磁盘利用率和 NVMe 分配,并可按需提供自定义指标。硬件披露没有软件披露全面,但仍有意义。Neysa 发布 H100 分数和完整 SXM 供给,并在产品和案例研究材料中另行提到 H100 和 H200 可用;NVIDIA 文档则支持这些 SKU 为什么对需要高内存带宽、NVLink 互连和低延迟横向扩展行为的训练和推理工作负载重要。因此,架构故事是自洽的:Neysa 试图拥有的不只是计算配置,还包括位于 GPU 容量和客户生产 AI 工作流之间的编排与操作层。[CE004, CE005, CE006, CE007, CE008, CE009]

技术 / 运营架构表
层级 / 组件作用有证据支撑的实现细节关键依赖风险
AI 集群管理为 AI 工作负载组织算力池架构页明确把它列为 aiPaaS 生命周期的一部分可用 GPU 容量和调度器质量公开细节没有披露放置逻辑或租户隔离机制
AI 调度器将工作负载匹配到算力资源被列为控制平面组件,并关联到可扩展训练和推理流水线准确的资源元数据和工作负载感知能力没有公开基准显示负载下的调度效率
资源管理器分配并跟踪基础设施资源在公开架构中与 AI 调度器和集群管理并列遥测、配额控制和管理员策略配额和策略自动化深度没有公开文档
算力交付模式以裸金属、VM 或容器形式提供 GPU架构页明确列出三种交付模式GPU 供给、虚拟化栈和运维工具没有公开矩阵说明各模式可用哪些 GPU SKU
存储层持久化数据、检查点和模型工件明确列出对象、块和 NFS 支持;案例研究还提到 NVMe底层存储网络和吞吐设计公开材料没有发布存储性能基准
集成界面将 Velocis 接入现有开发和企业工具身份:SSO/SAML/LDAP/RBAC;Dev/MLOps:GitHub 或 GitLab、Docker、MLflow、Kubeflow、Airflow;云:VPC 和混合支持客户 IAM、容器和 CI/CD 体系公开文档没有展示已测试集成的广度或连接器成熟度
可观测性和日志监控基础设施和模型运行仪表盘声称覆盖 GPU 利用率、磁盘利用率、NVMe 分配和自定义指标可靠的遥测采集和 UI 深度没有发布日志和指标的公开截图或 schema 级文档
推理和伙伴层将训练好的模型转成生产 API 和工作流推理端点、自动扩缩容概念和 Pipeshift 集成把栈延伸到训练之外合作伙伴软件、安全控制和网络合作伙伴带来的性能收益未必能泛化到所有工作负载

本表区分公开点名的组件和推断行为;缺少基准与 API 文档仍是最大的架构尽调缺口。

[CE004, CE005, CE006, CE007, CE008, CE010]
FE001: 产品架构图

Velocis 把主权基础设施、灵活算力交付、编排、开发者工具、推理服务和治理叠成一个运营栈。

该技术栈使用 Neysa 命名的产品组件和公开集成列表;由于 Neysa 尚未发布深入技术参考架构,它仍是分析师重构。

[CE001, CE003, CE004, CE005, CE006, CE008]

5.3 信任、主权和部署验证

安全、合规和本地化是 Neysa 差异化故事的核心;对早期基础设施供应商而言,公开证据好于平均水平。Neysa 记录了 zero-trust 访问、项目和资产级 RBAC、SSO 与 IAM 集成、可导出审计轨迹、静态和传输中加密,以及客户管理密钥支持。它还声称拥有 ISO/IEC 27001:2022 认证和 SOC 2 合规,CSA STAR 注册表也独立列出 Neysa Velocis 的 Level 1 自评和 Level 2 认证记录。本地化论证同样明确。Neysa 的活动、融资和合作伙伴材料反复称,工作负载、prompts、模型权重和企业数据可以留在印度数据中心内,并把这点描述为对 BFSI、医疗、政府和公共服务用例尤其重要。客户验证与这一叙事一致。TIFIN 提到印度边界内数据处理和 99.95% uptime,用于受监管金融工作负载;Innoviti 强调确定性的低于 30 秒延迟、白盒控制和对支付数据敏感的运营;IISc 描述用于开放权重研究的高内存裸金属训练;首页引用和合作伙伴材料则指向 Nurix、Navana 和 Smallest 等语音 AI 构建者。合起来看,信任故事是主权运营加生产支持,而不是事后叠上的合规勾选项。[CE012, CE013, CE014, CE015, CE016, CE017]

信任 / 质量 / 合规表
控制 / 认证状态范围重要性缺口
零信任访问模型声称已上线覆盖从配置到训练、调优和服务的全流程支撑租户隔离和受监管工作负载没有公开架构材料详细解释执行边界
RBAC 加 SSO/IAM 集成声称已上线权限可按项目、角色画像或资产分配;SSO 和 IAM 被列为集成项企业管理员可将 Velocis 映射进现有身份控制没有公开管理员指南或权限模式
审计日志声称已上线每个操作、访问事件和部署触发都会记录并可导出对治理、取证和合规团队很重要留存窗口、导出格式和告警控制未披露
加密和 BYOK/KMS声称已上线数据和模型工件在静态和传输中加密,并支持客户管理密钥对敏感模型权重和受监管数据至关重要公开文档没有说明支持哪些 KMS 后端或密钥轮换工作流
ISO/IEC 27001:2022 和 SOC 2Neysa 声称具备安全页面列出两项;面向客户的信任表述较宽泛传递最低限度的企业安全成熟度信号公开来源没有披露 SOC 2 报告范围或例外项
CSA STAR 列名和认证独立列名CSA 注册表显示 Neysa Velocis 的 Level 1 自评和 Level 2 认证条目将部分信任主张从 Neysa 营销文案中外部化CSA 列名不能替代买方对控制运行的完整尽调

控制项在功能或注册表层面有证据支撑,但买方级信任材料仍部分受限,或无法公开取得。

[CE012, CE013, CE014, CE015, CE016, CE017]
FE004: 产品成熟度 / 能力图

GPUaaS、aiPaaS 和安全是当下最明确的公开界面;市场生态和完整机群透明度,在公开证据中仍明显不成熟。

[CE020, CE025, CE026, CE028, CE031, CE035]

5.4 路线图信号、扩张路径和技术风险

Neysa 的路线图信号在与融资、合作伙伴发布和明确产品占位符交汇处最强。架构页称路线图设计用于吸收新的 GPU SKU、模型格式、agents、微调和向量数据库。主产品页把 marketplace ecosystem 标注为即将推出,而非普遍可用;这很重要,因为它区分了已交付模块和愿景模块。Pipeshift 发布把产品延伸到面向开源模型的单租户、OpenAI-compatible 推理,并暗示近期市场进入重点是对延迟敏感的企业工作负载,例如语音 AI、企业搜索、copilots 和推理系统。Blackstone 和 TechCrunch 又加入基础设施扩张层:资金被指定用于计算、网络、存储、编排、可观测性和安全软件,并明确提出在印度长期超过 20,000 张 GPU 的目标。同一批来源也暴露了主要技术风险。精确现网 GPU 组合仍披露不足,marketplace 时点含糊,公开 SLA 文档偏薄,外部新闻报道仍强调芯片供应、数据中心建设和电力可得性等行业约束。换言之,Neysa 的路线图可信且资本支持越来越强,但一些最重要的承保细节仍藏在幕后。[CE025, CE026, CE028, CE029, CE030, CE034]

路线图 / 发布 / 开发阶段表
日期 / 阶段功能 / 里程碑状态含义来源
2025-12 产品叙事刷新博客把 Velocis 介绍为 AI 加速云,定位包括 H100/H200、裸金属、PaaS 和推理已上线 / 已发布公开定位从泛云语言转向模块化 AI 栈叙事Neysa 博客
2025-12 至 2026-01 信任信号Neysa Velocis 列入 CSA STAR Level 1 和 Level 2 记录已上线 / 独立列名提升安全敏感买方的采购可信度CSA STAR 注册表
2026-02 扩张融资Blackstone 领投融资,支撑 GPU 规模迈向 20,000+,并建设软件近期 / 资本已承诺为路线图主张增加资产负债表支撑Blackstone 和 TechCrunch
2026-05 实时推理扩展Pipeshift 合作为印度境内开源模型增加单租户、兼容 OpenAI 的推理已上线 / 已宣布强化 Neysa 在语音、copilot 和企业自动化上的叙事Neysa 新闻稿、ExpressComputer、Pipeshift
2026-06 客户证据扩展IISc、Innoviti 和 TIFIN 案例研究加深了研究、零售和 BFSI 的垂直证据已上线 / 已发布暗示产品正从定位走向可重复部署模式Neysa 案例研究
路线图占位项核心产品页把 Marketplace 生态标为即将推出路线图 / 尚未 GA潜在生态上行存在,但收入时点和伙伴广度仍不清楚Neysa Velocis 页面

前瞻项目仅限公开里程碑和明确占位项;缺少正式公开产品路线图仍是真实尽调约束。

[CE021, CE025, CE026, CE027, CE028, CE040]
FE003: 关键依赖图

Neysa 扩大规模依赖 GPU 供应、印度设施和电力建设、合作伙伴软件、外部资本,以及主权敏感客户的持续需求。

依赖节点把明确公开披露与紧邻的运营要求合并;Neysa 尚未发布正式供应商图谱。

[CE018, CE023, CE027, CE042, CE043, CE049]
Chapter 06

06客户

6.1 客户分群和购买动作

Neysa 的公开定位很宽,但并不随机。其行业页面明确把需求分成受监管金融机构、保险公司、数字商业团队、制造商、研究机构和 AI 原生初创公司,并由一个共同购买命题串起:客户想跑 GPU 密集型 AI 工作负载,但不想把多个超大规模云服务拼在一起,也不想把数据带出印度。受监管板块最清晰。BFSI 和保险页面强调 RBI、IRDAI、隐私、可审计性和模型治理要求;主权 AI 博客则把本地司法辖区和本地采购基础设施描述为采购要求,而不是品牌选择。零售和制造买家更多围绕运营瓶颈展开,而非监管:高推理成本、集成复杂、从试点到生产延迟,以及边缘或工厂车间部署需求。初创公司处在光谱另一端:卖点是即时 GPU 访问、按量定价、无等待名单和低摩擦实验。这种组合说明 Neysa 试图同时服务高价值受监管企业和行动更快的 AI 原生构建者,但公开证据显示,在主权、延迟和支持比全球广度更重要的印度优先行业里,其可引用性最强。定价和合作伙伴项目语言也显示,Neysa 预期采用会先从实验或定向工作负载开始,再走向更广的企业标准化。[CU001, CU002, CU003, CU004, CU005, CU006]

客户细分表
细分买方 / 用户 / 付款方代表性用例公开证据战略价值关键缺口
BFSI / 财富 / 支付银行、NBFC、金融科技公司、财富管理机构欺诈、风险评分、客户智能、投资旅程TIFIN 案例;TOI 点名 Perfios 和 Juspay;GreyLabs 为 BFSI 语音分析提供伙伴路径高:受监管工作负载和主权数据定位契合 Neysa 核心论点没有按金融子细分披露公开收入结构、合同期限或续约数据
保险保险公司和理赔 / 承保团队理赔自动化、文档 AI、定价、续保率、合规仅有保险行业页;未发现点名保险公司部署中:受监管买方匹配度强缺少点名生产客户证据
零售 / 电商 / 支付运营零售产品、数据和运营团队;Innoviti 作为中间付款方推荐、定价、流失预测、现场支持验证Innoviti 案例,以及 Reliance Retail、Shoppers Stop、DMart 等零售商名称高:可见生产指标和大型运营足迹Innoviti 可能是中间平台,而不是 Neysa 的直接零售商 ARR
旅行分销ITQ / Travelport 生态运营方在推理规模下解读航空改签和取消规则ITQ 案例研究中:展示了特定领域工作流中的高量推理除 Neysa 材料外,没有独立客户侧佐证
研究 / 教育公共研究实验室、大学、学生开发者模型训练、AI 实验室、多模态研究IISc 案例和 O3SLM 项目页中:有力证明算力用途和印度研究相关性研究使用不一定等同于可重复企业收入
AI 原生创业公司 / 前沿建设者创业创始人、ML 团队、伙伴主导的创业客户训练、调优、推理 API、快速资源配置创业公司页面;Pipeshift 与 Nurix 和 Arrowhead AI 的部署;WEKA 提到从创业公司到企业的范围中高:把 TAM 扩展到受监管企业之外公开支出水平和客户标识留存未披露

细分结合了 Neysa 自有垂直页面、案例研究和独立报道;许多行更能证明解决方案适配,而非披露的收入贡献。

[CU001, CU002, CU003, CU004, CU005, CU006]
客户增长 / 采用轨迹表
指标数值 / 状态日期来源置信度含义缺失分母
已上线 GPU 机群约 1,200 块 GPU 已上线2026-02TechCrunch / Moneycontrol展示当前已安装基础,而不只是未来目标没有按客户或地区拆分
目标 GPU 部署随时间部署 20,000+ 块 GPU2026-02TechCrunch / Moneycontrol / Entrackr显示为客户需求激进扩建容量没有披露已签需求来支撑目标
日客户工作负载代理指标客户每天处理超过 1 亿 tokens2026WEKA 客户故事暗示多个客户存在真实生产使用没有按客户或工作负载类型拆分
TIFIN 结果GPU 支出降低 65%,可用性 99.95%2026 案例研究Neysa TIFIN 案例受监管金融领域最强的公开成本与可靠性证据单一案例研究;未披露合同规模
Innoviti 结果TCO 降低 60%;每天 7,000+ 条日志;自动核验准确率 96%;时延低于 30 秒2026 案例研究Neysa + Innoviti 案例研究大规模零售 / 支付运营中的具体生产部署结果绑定一个工作流,不是整个客户账户
ITQ 使用规模每月数千亿 tokens2026 案例研究Neysa ITQ 案例展示旅行领域的企业级推理规模未披露支出或席位数

本表混合了已安装基础代理指标、容量计划和案例研究结果;许多数值来自公司声称或伙伴报告,缺少客户级分母。

[CU016, CU017, CU022, CU023, CU025, CU036]
采购摩擦与渠道影响表
主题证据客群含义尽调问题
数据本地化 / 主权控制BFSI 和保险页面强调对齐 RBI / IRDAI、可审计性和区域内数据处理受监管企业采购审批取决于合规和架构审查,不只是模型准确率审阅架构图、密钥管理模型和面向监管的文档
试点转生产BFSI、保险、零售和制造页面都提到从 PoC 或试点转生产存在延迟跨垂直行业预算转化可能滞后于技术热情索取平均转化时间、试点成功率和付费生产率
成本透明度定价页面提供按小时计费、预留折扣,并承诺没有 egress / API 意外费用初创公司和企业工作负载负责人低摩擦试验有助于拿下交易将样本工作负载的实际账单与超大规模云厂商替代方案对比
伙伴主导分销伙伴页面承诺经销商、SI/MSP、ISV、推荐费、收入分成和联合销售渠道生态渠道可能实质影响客户发现和购买 Neysa 的方式索取按直销、伙伴来源和联合销售动作拆分的 sourced pipeline
工作负载优化负担Economic Times 称,一家 Neysa 客户仍因工作负载未优化而被 token 成本和延迟困住企业 AI 采用者客户成功可能需要的不只是原始 GPU 访问,还要架构帮助索取托管优化操作手册、FinOps 工具和参考工作负载基准

这张图表把营销和媒体证据转成具体采购摩擦主题;多数条目描述采购机制,而不是已赢单结果。

[CU009, CU010, CU011, CU041, CU043, CU045]
FU001: 客户旅程图

展示 Neysa 如何把受监管或成本敏感需求推进到生产采用和扩张。

阶段划分综合了公开案例、定价表述和负面评价证据,并非公司披露的官方漏斗。

[CU007, CU008, CU021, CU032, CU041, CU042]
FU002: 采用 / 部署漏斗

展示 Neysa 的公开客户证据如何从宽泛目标客群,收窄到少数可引用的部署案例。

计数为本章作者对保留公开证据的统计,并非公司报告的漏斗指标。

[CU019, CU028, CU030, CU038, CU040, CU044]

6.2 具名客户验证和使用模式

最有力的公开客户证据来自能讲清楚具体生产负载的案例,而不是单纯摆 logo。TIFIN 是最完整的证据点:Neysa 称 TIFIN 服务印度最大的共同基金和财富管理机构,在另一家本土 neocloud 出现可靠性问题后迁移了生产负载,相比 hyperscaler 节省 65% GPU 支出,并达到 99.95% 正常运行时间。Innoviti 是最清晰的规模化运营参考:其支付网络覆盖 2,000 个城市的 50,000 多家商户,并包括 Reliance Retail、Shoppers Stop、DMart 等企业零售商;Neysa 与 Innoviti 都把这次部署描述为从概念验证走向生产 AI 推理环境,TCO 降低 60%,每天处理 7,000 多条工单日志,自动验证准确率达 96%,延迟低于 30 秒。ITQ 把故事延伸到旅游分销场景:航空公司政策解读每月消耗数千亿 token,据称通用云的经济性已经跑不通。Neysa 还把一名未具名的全球 medtech 用户和 IISc Bangalore 列为证据,说明平台触达了受监管的医疗创新和研究密集型负载。独立报道扩大了可见客户面:The Times of India 点名 Juspay、Swiggy、Perfios 为关键客户;WEKA 称 Neysa 服务的客户从初创公司到大型企业不等,并支持每天超过 1 亿 token 的处理量。合在一起看,证据支持真实采用,但仍依赖相对少量的公开参考,以及客户、合作伙伴和公司各自叙述混杂的材料。[CU012, CU013, CU014, CU015, CU016, CU017]

点名客户证据表
客户 / 用户细分部署 / 用例生产 / 试点结果 / 证据质量限制 / 缺口
TIFIN财富 / 金融科技面向印度财富和共同基金用例的生产 AI 工作负载生产点名高管引用;GPU 支出降低 65%;可用性 99.95%;存在客户侧 TIFIN 印度足迹未披露合同期限、ARR 或续约
Innoviti支付 / 零售运营覆盖支付终端和服务日志的 AI 驱动现场运营智能生产点名客户侧案例佐证从 PoC 进入生产;TCO 降低 60%,商户网络 50,000+证据限于具体工作流,并通过 Innoviti 路由,而非直接终端商户合同
ITQ Technologies旅行分销用于解读航空改签 / 取消规则的推理生产点名特定领域工作流达到企业级体量;数千家代理接入 Travelport 库存仅保留 Neysa 侧案例研究;未发现 ITQ 对 Neysa 关系的独立确认
IISc Bangalore Visual Computing Lab 实验室研究 / 教育为 O3SLM 草图-语言模型工作提供算力生产研究工作负载点名实验室、具体模型,并有 AAAI 2026 项目页佐证研究证据不能直接证明可重复商业支出
全球 MedTech 领导者(未具名)医疗健康 / 医疗科技面向机器人、诊断和细胞疗法创新的 AI 基础设施类生产,但客户未具名用例和团队规模具体;受监管垂直适配度强客户身份、合同规模和续约状态未披露
Juspay / Swiggy / Perfios支付 / 商业 / 金融科技The Times of India 将其列为关键客户阶段不清楚独立媒体把 logo 集扩展到官方案例研究之外没有按客户标识披露公开工作负载、阶段、结果或留存细节

公开客户证据混合了直接点名部署、客户侧佐证和独立媒体点名;未具名医疗科技以及未具名公共部门 / 前沿实验室需求,仍无法精确到行级。

[CU012, CU013, CU016, CU017, CU019, CU021]
留存 / 重复使用 / 满意度表
指标 / 代理指标数值 / 状态细分置信度尽调问题
净收入留存(NRR)未公开披露所有细分索取按企业与创业客户客群拆分的过去 12 个月 NRR
毛收入留存(GRR)未公开披露所有细分按客群索取 GRR、流失 ARR 和总 logo 留存率
具名客户的续约 / 合同期限未公开披露TIFIN, Innoviti, ITQ, MedTech索取合同开始日期、最低期限和续约机制
复用代理指标:TIFIN迁至 Neysa 后完成生产迁移,商业化部署路径也更快BFSI确认上线后部署是否在 GPU 数量、产品或业务单元上扩张
复用代理指标:Innoviti从 PoC 转入生产,每天处理日志并做实时推理零售 / 支付运营确认上线后处理量或覆盖地域是否扩大
公开客户满意度 / 评价保留证据中没有形成可靠的 Neysa 专属公开评价集所有客群索取客户推荐和支持 SLA,不要依赖稀疏的公开评价信号

留存质量只能从持续生产使用表述和工作流规模推断,不能从已披露的订阅或客户分群指标验证;这是一个实质性尽调缺口。

[CU018, CU021, CU040]
FU003: 客户证据矩阵

按部署成熟度、成果具体性、留存可见度和证据质量,对比公开证据。

矩阵评级是作者基于保留公开证据的具体程度作出的判断,并非供应商提供的评分。

[CU016, CU017, CU021, CU022, CU024, CU028]

6.3 持久性、采购摩擦与集中度风险

持久性是 Neysa 公开客户故事里最薄弱的一环。已审阅材料没有披露 NRR、GRR、流失率、合同期限、最低承诺、续约率或头部客户收入占比,因此无法从公开信息直接判断这些可见参考是否黏性足够,还是只是近期新增。现有代理指标积极但不完整:TIFIN 描述了迁移后的生产部署和更快的商业 rollout;Innoviti 与 ITQ 都把 Neysa 定位为让规模化生产可行的基础设施;与 Pipeshift、GreyLabs 的合作动作暗示方案型分销正在扩大。但同一组证据也暴露了摩擦。Neysa 自己的细分页面反复提到,从试点或 PoC 进入生产的延迟、数据本地化审查、GPU 成本压力和集成开销,都是预算真正放大前必须跨过去的障碍。Economic Times 还指出,企业客户即便采用新模型后,仍在处理 token 成本和延迟问题;ClusterMAX 的负面评测则是最清晰的外部警报:SemiAnalysis 给 Neysa 评为 Bronze,并指出其安全性、易用性、入门、调度和监控相较国际竞争对手存在弱项。因此,集中度风险很实质。公开点名的证据集中在少数印度中心的垂直场景——财富管理、支付和零售现场运营、旅游分销、研究,以及一个未具名 medtech 客户;独立媒体又补充了一些具名客户,但没有披露收入权重。这意味着 Neysa 的伙伴计划和新的伙伴主导产品可能对多元化很重要,但当前公开证据还没有说明增长中有多少来自重复扩张,又有多少来自新 logo 获取。[CU032, CU033, CU034, CU035, CU040, CU041]

扩张与集中风险表
扩张驱动 / 集中风险类型影响尽调路径
BFSI、保险、医疗和政府等受监管行业的主权需求扩张驱动高正面:契合 Neysa 的印度本地化价值主张按行业测试 pipeline 转化率和平均交易规模
与 Pipeshift 和 GreyLabs 联合推出的伙伴主导产品扩张驱动中正面:可借解决方案打包和联合销售拓宽漏斗量化伙伴来源 pipeline、胜率和挂载率
公开具名证据集中在少数垂直行业和案例 logo集中风险高:如果支出集中,可见推荐可能夸大客户广度索取前 10 大客户、收入占比,以及按 logo 划分的部署阶段
未具名 medtech 与未披露的公共部门 / 前沿实验室需求集中风险中:暗示 pipeline 有广度,但公开可审计性弱获取按垂直行业划分的匿名客户分群数量和 ARR
试点转生产摩擦和合规签批周期集中风险中至高:拖慢预算爬坡,也可能延迟续约或扩张衡量从技术演示到付费生产的平均实施时间
ClusterMAX 暴露的平台质量问题反向集中 / 执行风险高:安全、上线和监控缺口会削弱企业客户背书能力询问第三方审计、RBAC 路线图、正常运行历史,以及 2025 年后的平台修复

风险重点在客户背书集中和转化摩擦,而非资产负债表敞口;公开证据没有披露收入集中度比例。

[CU034, CU040, CU042, CU043, CU044, CU045]

6.4 展示材料

Chapter 07

07风险

7.1 资本强度和利用率是一阶风险,因为其他上行假设都建立在这两点之上

公开记录现在把 Neysa 描绘得更像一个基础设施资产负债表故事,而不是软件初创公司。Blackstone 2026 年 2 月的交易不是小规模增长延伸;它把最高 $600M 新股权与计划中的 $600M 债务配在一起,并把融资直接绑定到在印度部署超过 20,000 块 GPU 的计划。TechCrunch 同时报道称,Neysa 约有 1,200 块 GPU 已上线,目标是在高级客户对话转化后一年内把容量扩大到三倍以上。这给公司留下的运营窗口很窄。定价页显示 committed-use 折扣很激进,包括多年 H100 预留的大幅节省;公司还宣传不对出口流量或推理交易额外收费。这些条款能帮助抢份额,但也让上座率、预留量捕获和机群利用率成为利润率修复的决定因素。如果部署延迟、债务文件收紧,或客户停留在试点而不是转向持续预留使用,下行会先打到现金流,再打到估值。[CR001, CR002, CR003, CR004, CR005, CR006]

缓释措施与终止标准表
风险可监控触发项阈值 / 事件行动含义
资本强度跑在利用率前面上线 GPU 部署、预留用量采用或付费客户增长落后于基础设施承诺集群增长与已签约 / 已预留消耗之间出现任何跨季度缺口下调估值预期,重新承销现金 runway,并在增加敞口前要求更新债务和占用率数据
GPU 供应集中管理层披露交付滑坡或延长时间线,同时维持 capex 承诺以 Nvidia 为主的采购出现实质延迟,且没有芯片多元化抵消假设收入爬坡更慢,获客摩擦更高
电力和站点准备度承压MW 扩张、冷却或 RTC 电力安排落后于 AI 机架密度要求站点启用延迟,或出现本地电网 / 冷却约束证据降低对正常运行时间声明的信心,并增加电力采购尽调
安全 / 合规失败关键 CERT-In 类问题、重大客户事件,或公开泄露 / 合规争议任何严重事件如果没有快速客户沟通和有记录的补救在控制证据改善前,将其视为受监管工作负载采用的投资逻辑破裂点
客户集中 / 证据缺口收入增长快于披露节奏,但公开证据仍局限于少数灯塔 logo下一轮刷新周期内,具名生产部署或背书质量没有拓宽对利用率和留存假设施加集中度折价
政策或 IndiaAI 依赖政策支持需求走弱,或竞争对手拿走可见补贴工作负载IndiaAI 相关上线实质减少,或战略入围资格相关性下降将 thesis 重构为纯商业需求,并降低终局倍数假设

这些终止标准把公开证据转成可观察阈值;它们不是预测,而是界定资本、利用率或控制假设何时应被视为失效。

[CR001, CR003, CR007, CR014, CR021, CR024]
FR002: 风险传导图

展示 Neysa 的基础设施风险如何传导到收入、融资和估值。

[CR001, CR002, CR003, CR004, CR006, CR014]

7.2 GPU 供应、电力可得性和 hyperscaler 竞争相互叠加,而不是各自独立

对一家年轻公司而言,Neysa 的供给侧暴露异常集中。Business Standard 引述管理层称,约 95% 机群基于 Nvidia,仅有部分 AMD 库存,并且还在讨论更广泛的芯片多元化。这很重要,因为同一个 2026 年新闻周期里,Neysa 一边明确谈到供应链韧性,一边又在与 AWS、Azure、Google 的产品竞争;这些产品已经宣传大规模 H100 或 H200 集群、深度互连和成熟 AI 生态。竞争问题不只是标价问题。Neysa 必须拿到芯片、完成部署、供电、散热并保持满载,同时买家会继续把它与 GPU 供应、软件广度和融资能力都大得多的供应商比较。行业证据进一步加压:印度 data-centre 评论现在指向运营用电需求从 2025 年约 1 GW 跳升到 FY32 的 13 GW;AI-ready 机架可能需要 80–120 kW,大型负载还需要大量储能和冗余供电规划。供应滑坡、电力瓶颈,或付费利用率爬坡慢于预期,都会彼此强化。[CR007, CR008, CR009, CR010, CR011, CR012]

伙伴 / 依赖风险登记表
依赖交易对手角色集中度失效情景严重性缓释措施剩余敞口
加速器供应Nvidia(目前只有有限 AMD 多元化)主要 GPU 供应层交付延迟、配额受限或价格压力拖慢计划中的 20,000 块 GPU 建设关键管理层提到计划推进芯片多元化,并持续与替代供应商沟通高——95% Nvidia 组合仍意味着短期依赖高度集中
基础设施融资Blackstone 领投股权,加上计划中的债务贷款方为集群、存储、网络和数据中心扩张提供资本利用率追上前,债务条款收紧,或未来股权融资价格更差关键大规模初始融资和经验丰富的基础设施支持方改善融资可得性高——该模式仍高度依赖资产负债表,对融资敏感
政策驱动需求IndiaAI Mission 和公共部门采购渠道需求创造、资格准入和补贴算力可见度中高Mission 需求放缓、转向其他入围提供商,或补贴条款变化入围资格和主权定位让 Neysa 仍可参与当前项目中高——政策顺风是共享的,并非独占
推理和应用伙伴Pipeshift 和更广泛的开放权重生态工作负载上线和生产推理栈伙伴表现、路线图延误或模型目录缺口削弱 Neysa 的价值主张中高OpenAI 兼容 API 和开放权重定位降低单一工具锁定中——伙伴生态有帮助,但仍处早期
竞争基准AWS、Azure、Google Cloud 和印度 GPU 云同行定价、软件广度和供应规模参照客户选择既有厂商或多云部署,限制 Neysa 提价或推动长期预留的能力关键本地数据驻留叙事、集成 MLOps 和支持差异化高——超大规模云厂商和本土对手可以在单个部署上砸更多钱或做出更大规模

依赖按运营层分组,而不是穷尽法律实体名单,因为 Neysa 没有公开披露所有贷款方、托管机房伙伴或采购交易对手。

[CR001, CR002, CR007, CR008, CR009, CR010]
FR003: 依赖图

Neysa 要让主权 AI 基础设施保持竞争力,主要依赖的外部层。

[CR007, CR008, CR009, CR010, CR011, CR014]

7.3 政策、安全与主权是护城河的一部分,但也带来更重的合规负担

Neysa 最强的差异化主张是,印度企业可以把敏感 AI 负载、prompt 和推理流量留在印度境内,并获得比外国托管技术栈更强的运营控制。这个定位在商业上有用,但也意味着公司主动承担了更密集的合规面。隐私政策列出共享责任模型:Neysa 负责基础设施安全,客户租户仍负责自己的数据隐私和访问控制。产品页面宣称具备 RBAC、加密、BYOK、审计日志、ISO/IEC 27001:2022 认证和 SOC2 合规;监控岗位招聘则显示,事件管理、补丁、根因分析和 SLA 升级已经是实际运营职能,而不是未来愿景。外部义务也在上升。CERT-In 2026 年 6 月指南明确覆盖云服务提供商,并要求严重漏洞立即披露、加速修复。DPDP Act 和 IndiaAI Mission 又给数据处理、本地部署和公共部门就绪度增加了预期。这些都是有价值的需求顺风,但也提高了失败成本,因为安全事故或合规失误会最先打到受监管客户。[CR018, CR019, CR020, CR021, CR022, CR023]

监管 / 法律风险登记表
规则 / 义务司法辖区 / 暴露面状态可能性严重性缓释措施剩余敞口尽调路径
AI / 云提供商的 DPDP Act 义务印度数据处理和跨境传输暴露面现行法律,分阶段合规中高本地部署姿态、隐私通知、共同责任框架和企业控制高——受监管工作负载会放大任何隐私或传输失误索取 DPO 工作流、泄露通知 playbook,以及把第 8、10、16 条落到操作层的客户附录
CERT-In 漏洞与披露义务印度的云服务、API、软件和托管基础设施2026 年 6 月指引集已生效公开资料描述了安全团队、事件监控、补丁流程和产品级可审计性高——严重事件会面临快速披露和补救预期获取漏洞管理节奏、关键补丁 SLA 达成情况和客户通知模板的证据
IndiaAI Mission / 公共部门依赖有补贴的主权算力生态,以及由入围资格驱动的需求当前政策顺风,不是永久合同护城河IndiaAI 入围、境内基础设施和主权定位有助于符合资格中高——政策支持由多家提供商共享,也可能改变采购组合索取 IndiaAI 驱动需求与纯商业需求的收入拆分,以及按客群划分的续约行为
隐私政策共同责任客户租户、基础设施安全和跨境传输披露活跃的合同 / 法律暴露面中高政策区分 Neysa 管理的基础设施控制和客户管理的租户数据责任中高——共同控制边界含糊,事件后可能引发争议或甩锅风险审阅 DPA、客户安全附录和租户默认配置文件
安全认证营销与可检查证据产品页面公开声称 ISO 27001 和 SOC2公开页面提供了部分证据中高认证声明、RBAC、加密和审计日志均有公开表述中高——公开网站没有给出完整审计包、正常运行历史或控制例外索取证书编号、最新认证日期、控制范围和面向客户的保证材料包

各行按公开来源可见的重大法律和合规敞口排序;Neysa 未公布合同排期、事件文件或完整审计材料,因此这份登记表并不完整。

[CR018, CR019, CR020, CR022, CR023, CR024]
运营 / 质量 / 安全风险登记表
失效模式可能性严重性缓释成熟度剩余敞口未解决缺口
GPU 集群到货速度慢于收入承诺或预留销售所需中高关键低至中——融资规模强,但供应韧性仍是管理层的现实问题高——集群填不满,债务、折旧和运营开销就缺少覆盖没有公开供应合同、交付节奏或集群利用率披露
电力、冷却或电网准备度落后于 AI 密度建设关键低至中——行业规划存在,但印度全境电力瓶颈仍是结构性问题高——AI 机架同时加剧正常运行和成本压力没有公开的 Neysa 站点级 PUE、MW pipeline 或已签约 RTC 电力细节
安全事件或重大漏洞引发客户不信任和监管关注中——安全工具和流程声明已公开,但第三方证明有限高——受监管客户可能会对任何失误快速反应没有公开事件历史、平均解决时间或审计问题趋势
生产规模推理工作负载中,延迟或可靠性承诺未兑现中——案例研究显示低于 30 秒或 TTFT 声明,也有事件运营岗位配置中高——参考客户承担关键任务,对性能退化的容忍度可能更低没有公开 SLA 达成情况或正常运行时间序列
利用率成熟前,预留经济性和全包定价压缩利润率低至中——标价透明,折扣可能帮助赢得客户高——无意外成本定位可能把基础设施成本内化吸收没有公开毛利率、电力转嫁或客户分群利用率数据

这张表混合了公司自述缓释措施和行业层面的运营事实;Neysa 未公布 KPI 证据时,未解决缺口列直接列出尽调问题,不做猜测。

[CR003, CR004, CR005, CR013, CR014, CR015]

7.4 公开客户证据可信,但范围仍窄,集中度和执行风险依然偏高

现在已有真实的公开部署证据,但还不足以抵消集中度担忧。Neysa 网站在研究、支付、语音 AI 和企业自动化等场景中点名少数客户和用例;更早的融资报道则提到,付费客户覆盖 AI-native 初创公司、媒体、软件供应商、公共部门用户和其他企业类别。记录最充分的负载也最苛刻。IISc 案例研究描述了 3,300 万对数据生成流水线,以及在专用 bare metal 上反复运行 7B 和 13B 训练。Innoviti 描述了 50,000 多家商户、每天 7,000 多条服务工单日志、支付数据敏感性,以及确定性低于 30 秒推理的硬性要求。这些都是有意义的证据点,但也说明早期参考客户集中在高风险负载里;一次宕机、迁移延迟或成本超支都可能放大声誉损害。Neysa 招聘页面上的监控岗位强化了同一个判断:可靠运营、补丁和事件响应已经是核心执行肌肉,不是后台锦上添花。[CR028, CR029, CR030, CR031, CR032, CR033]

人员 / 执行风险登记表
角色 / 职能依赖或缺口可能性严重性缓释措施尽调路径
创始人主导商业和基础设施领导公开可信度仍高度集中在 Sharad Sanghi 和 Anindya Das 身上Netmagic 履历和强投资人背书有助于建立企业信任索取组织架构图、授权委派图,以及销售、基础设施和安全团队的梯队深度
可靠性和事件运营生产工作负载要求 24/7 监控、升级、补丁和 RCA 纪律中高专职监控角色、ITIL 式事件流程和工具预期已经明确审阅事件待处理清单、升级矩阵和按职能划分的值班覆盖
现场和客户成功工程高接触企业和公共部门部署可能比通用云模式需要更多陪跑Neysa 把动手支持和快速响应作为差异化卖点询问服务人员数量、部署时间线和客户 / 支持人员比例
安全、审计和合规运营客户和监管审查升温后,公开控制声明需要持续产出证据安全页面强调可审计性、加密和策略执行获取最新认证范围、整改待处理清单和第三方测试节奏
财务和容量规划债务支持的 GPU 采购会放大预测错误或利用率延迟的成本关键经验丰富的投资人和基础设施履历可能强化纪律索取按季度拆分的 capex 计划、债务偿付时间表、利用率预测和下行情景控制

这里的执行风险不在于 AI 需求是否存在,而在于 Neysa 能否让运营、财务和可靠性职能跟上基础设施承诺的速度。

[CR001, CR002, CR014, CR020, CR021, CR030]

7.5 治理和估值下行如何传导,取决于 Neysa 能否把基础设施规模转成可重复现金生成

公开材料讲出了连贯的增长故事,但仍没有给出成熟上市云运营商会提供的披露深度。这很关键,因为风险传导路径很清楚。如果芯片交付放慢、IndiaAI 相关需求走弱、企业转化继续呈 bursty 而非预留型,或者正常运行时间和安全证据更多由营销主导而非独立审计,那么利用率就会低于计划。放在 capital-light SaaS 公司里,这主要伤害倍数;放在叠加债务的 AI 基础设施建设里,它还会削弱现金转化、融资灵活性,以及与供应商和锚定客户谈判的筹码。行业背景也说明同一点。BusinessLine 和 ET Energy 都强调,data-centre 经济性回收靠后、对电力敏感、审批重,并暴露于利用率爬坡。Neysa 的上行真实存在,但公开记录也表明,如果规模晚于 capex 和估值到来,投资逻辑会很快断裂。[CR006, CR038, CR039, CR040, CR041, CR042]

FR001: 风险热力图

按公开证据呈现的影响和发生可能性,定位 Neysa 的剩余风险。

[CR001, CR003, CR014, CR020, CR024, CR028]

7.6 展示材料

Chapter 08

08估值

8.1 融资与公开价格背景

Neysa 当前估值背景很容易写成标题,却很难承销。融资是真实的:官方与独立报道都指向一个 $1.2B 融资包,分为最高 $600M 股权和计划中额外 $600M 债务,Blackstone 预计将成为多数股东。独立报道最常引用的标题估值约为 $1.4B,但公开记录把它描述为企业价值,并未披露优先权结构、债务定价或 covenant 包;这些才决定少数股权的真实经济价格。这个区别很重要。由 sponsor 主导的多数轮叠加债务层,可以改善 GPU 采购、电力获取和客户引介,但也可能嵌入下行保护,让普通股价值没有标题数字看起来那么慷慨。公开运营披露也很薄。Neysa 约有 1,200 块 GPU 已上线,希望扩至超过 20,000 块,员工约 110 人,并称明年收入应增长三倍以上;但公司没有公开披露收入基数、毛利率、利用率或客户集中度,无法证明当前价格是保守的,还是已经计入未来。[CV001, CV002, CV003, CV004, CV005, CV006]

FV002: 估值敏感性

展示在不同可比倍数下,Neysa 需要多少年化收入才能支撑 $1.4B 估值。

数值为 $1.4B 估值除以公开可见可比倍数后隐含的收入门槛,单位为百万美元;这些是敏感性锚点,不是预测。

[CV018, CV023, CV026, CV035, CV037, CV038]

8.2 可比基准与情景区间

可比公司更能说明 Neysa 还必须证明什么,而不是说明什么价格已经合理。CoreWeave 是披露规模的最接近公开 neo-cloud 参照,2026 年 6 月交易价格约为 Q1 年化收入的 6.6x,但这是在其构建了近 $100B backlog 和超过 1 GW active power 之后。DigitalOcean 更偏推理和 managed-cloud,交易价格接近 2026 年指引收入的 14x 或 ARR 的 16x,同时利润率为正,并披露 AI ARR 中超过 70% 来自推理服务和核心云,而非 bare metal。Together AI 的私募基准约为年化收入的 7.5x,更接近 CoreWeave,而不是热度更高的公开市场异类 Nebius;Nebius 2026 年 6 月市值隐含超过 Q1 年化收入的 40x。Nebius 能维持这个溢价,是因为它披露了超高增长、公开市场流动性和 1.2 GW 新工厂容量。Neysa 没有披露这些经济指标。在当前 $1.4B 标记下,如果按 15x 倍数,Neysa 需要约 $93M 收入;按 10x 需要 $140M;按 CoreWeave 式 6.6x 需要 $212M。没有披露收入,就不能凭公开证据说估值便宜。因此,牛市、基准和熊市情景都需要围绕明确的门槛经济性来框定,而不是只围绕叙事。[CV017, CV018, CV019, CV020, CV021, CV022]

牛市 / 基准 / 熊市场景表
情景收入 / 利用率假设倍数假设隐含估值区间概率信号 / 关键风险
牛市年化收入约 $150M-$220M,利用率强,软件 / 推理附加收入可见,并有长期签约需求12x-14x 收入$1.8B-$3.0B需要主权需求在超大规模云厂商缩小差距前转化为持久合同
基准年化收入约 $90M-$140M,算力加服务经济性混合,利用率中等9x-12x 收入$0.8B-$1.7B如果 Neysa 做出规模但仍是小众主权供应商,这是最可能情景
熊市年化收入低于 $60M,利用率弱,且相当多经济性被债务服务或折扣吃掉5x-8x 收入$0.3B-$0.7B由供应延误、需求可见度薄弱或价格竞争触发
当前公开估值公开证据未披露当前收入基数对应 $93M 收入约 15x,$140M 收入约 10x,或 $212M 收入约 6.6x当前 $1.4B问题不在价格本身,而在缺少已披露收入锚定价格

情景区间是基于公开可比公司倍数的明确假设,不应误读为管理层预测。

[CV036, CV037, CV038, CV039, CV040, CV042]
可比估值表
可比对象状态公开指标估值 / 倍数相关性局限
CoreWeave上市 AI 云2026 年 Q1 收入 $2.078B;待履约订单 $99.4B市值 $55.03B;约 6.6x 年化 Q1 收入GPU 优先基础设施最接近的规模化上市新云参照杠杆和集中度风险较重;比 Neysa 更成熟
Nebius上市 AI 云2026 年 Q1 收入 $399M市值 $65.92B;约 41.3x 年化 Q1 收入展示公开市场愿为超高增长 AI 云动能支付的价格异常溢价;不是常规承销基准
DigitalOcean上市托管云FY2025 收入 $901M;ARR $970M;AI ARR $120M市值 $15.5B;约 14.2x 指引 2026 年收入可参照软件附加、推理占比较高的云模式更广泛的 SMB 云业务,利润率为正,运营历史更长
Oracle OCI上市在位云业务FY2026 OCI 收入 $18.1B;RPO $638B全公司市值 $453.76B;分部无法直接拆分体现在位巨头给 AI 基础设施带来的规模和融资弹性混有数据库、SaaS 和其他业务;不是创业公司可比对象
Together AI私有 AI 基础设施 / API 平台Sacra 估算:约 $1B 年化收入正洽谈约 $7.5B 融资前估值;约 7.5x 年化收入收入可见 AI 基础设施的私有市场参照估算来自第三方分析,非经审计披露
Lambda私有 GPU 云2025 年报道披露 25,000+ GPUs 和 5,000+ 客户2025 年 2 月轮次 $2.5B;随后 2025 年 11 月 >$1.5B Series E显示专业 GPU 云的规模和反复资本需求收入未公开披露,无法直接计算倍数
Crusoe私有一体化 AI 基础设施2025 年前三季度订单额增长 5x;1.2 GW 园区阶段已上线> $10B Series E 估值一体化电力加云溢价参照电力主导模式比 Neysa 更垂直整合

可比集合刻意混合上市新云、托管云类比对象和私有 AI 基础设施建设者,因为 Neysa 尚未披露足够经济性,无法支撑更窄的纯粹同业匹配。

[CV017, CV018, CV022, CV023, CV025, CV026]
FV003: 估值 / 回报区间

在熊、基准、牛三种情景下,用明确的利用率和倍数假设给出示意性估值区间。

所有数值均为示意性股权价值,单位为百万美元。熊市情景假设利用率偏弱且倍数压缩;基准情景假设规模有意义但偏小众;牛市情景假设合约需求强劲,且软件附加销售跑通。

[CV040, CV042, CV043, CV048]

8.3 投资论点 vs. 反论点

Neysa 的正面案例在概念上自洽。印度计算缺口很大,政策方向支持,受监管行业的企业买家确实似乎重视本土基础设施、更低延迟和比通用云菜单更好的支持。公司也似乎明白,单纯出租 GPU 不够:其公开定位始终把 GPU 基础设施与编排、可观测性和 AI 安全打包。如果这个组合真的落到生产环境,Neysa 可以拿到比纯容量中介更好的倍数。反论点比管理层话术承认的更强。Hyperscaler 仍是真正的竞争基准,因为它们可以在更广的企业合同里交叉补贴 GPU 定价。公开记录还显示,AI-cloud 扩张越来越是信用和利用率游戏,而不只是需求故事。Data Center Knowledge 报道称,容量提供商现在最看重信用质量、终端客户可见度和长期利用率确定性。Neysa 在这些证据点上仍处早期。公司自己的材料承认 Blackwell 尚未部署,千 GPU 单次训练仍是建设目标。简言之:论点可以成立,但前提是利用率、软件 attach 和客户黏性出现得比债务和竞争压缩经济性更快。[CV008, CV009, CV010, CV011, CV012, CV013]

正方 / 反方论点表
维度正方论点反方论点什么会改变判断
主权 AI 需求印度特有的数据驻留、延迟和公共部门需求,给本土玩家留出切口政策叙事可能跑在真实变现前面;若在位巨头更激进本地化,切口会变窄展示已签约的受监管行业负载,以及稳定续约行为
全栈差异化集成可观测性、MLOps 和安全能力,相比单纯 GPU 租赁可抬高切换成本软件附加收入可能仍太小,抵不过算力商品化披露附加率、利润率,以及托管服务收入占比
赞助方优势Blackstone 可帮助采购、融资和客户引荐PE 控股加债务,也可能意味着优先权保护和更严格回报门槛分享条款清单、债务定价和任何投资人保护安排
客户证据公开客户证言显示,本地性、支持和合规确实影响真实买家客户证言不等于已披露 ARR、利用率或集中度提供 cohort 数据、预留容量承诺和集中度指标
可比公司支撑CoreWeave、Together AI 和 DigitalOcean 证明,真正的云资产可按远高于通用基础设施的倍数交易这些可比公司披露了 Neysa 未披露的规模和收入,Nebius 又属于异常高值披露足够运营数据,让 Neysa 能可信落入可比区间
竞争只要全球待履约订单让美国新云厂商暂时专注别处,印度专精可能有价值超大规模云厂商才是真正对手,因为它们能交叉补贴 GPU 定价并打包合同证明相对 AWS/Azure/GCP 有持久的 TCO、合规或服务优势

各行把公司质量和价格质量拆开,让推荐结论明确保持估值敏感。

[CV012, CV013, CV016, CV027, CV040, CV041]
FV001: 推荐逻辑

梳理从主权需求顺风、产品差异化,到隐藏经济性风险,再到最终建议的链条。

[CV008, CV013, CV037, CV043, CV048]
FV004: 投资 KPI

以 IC 风格记分卡评估 Neysa 在当前公开价格下是否值得投资,而不只是评估公司质量。

评分仅基于公开证据,由作者判断。收入可见度和资本结构透明度是当前最大的扣分项。

[CV011, CV012, CV013, CV015, CV016, CV048]

8.4 建议、触发因素与最终尽调

建议是继续研究,不是因为 Neysa 缺乏战略相关性,而是因为公开档案尚未证明当前价格可投。有足够证据相信公司在追逐真实机会:融资已经完成,capex 计划很大,创始人可信,政策和市场结构支持印度的主权 AI-cloud 细分生态。但没有足够证据相信 $1.4B 是便宜价格。缺失模块恰恰是约束价格纪律的变量:当前 ARR 或 run-rate revenue、毛利率、已签约利用率、头部客户集中度、债务成本,以及多数 sponsor 轮是否包含会压低普通股上行的优先权保护。实际的上调路径很直接。如果私下尽调显示,收入或合同利用率能让 Neysa 落在约 10x-15x 区间内,且结构不苛刻,评级可以转向 track 或 buy。如果条款强硬、利用率曲线偏弱,或 hyperscaler 定价在规模到来前侵蚀主权溢价,下行路径就是拉伸估值之后出现修正或稀释性 recap。因此,本章的核心信息是价格敏感性:公司可能有吸引力,但公开证据尚未证明价格合理。[CV015, CV016, CV037, CV039, CV048, CV049]

投资建议摘要表
维度评估证据基础决策含义
建议继续研究融资是真实的,但公开资料没有披露收入和投资条款经济性不要只靠公开证据承销 $1.4B 估值标记
信心公开证据足以形成价格敏感判断,但核心估值变量仍未公开私有尽调补齐缺口前,立场保持暂定
风险评级债务层、控股赞助方结构、利用率风险和超大规模云厂商竞争会同时影响判断若继续推进,先假设需要下行保护
估值立场偏高可见可比区间要求 Neysa 披露的收入显著高于目前公开水平把本轮当作待检验的基准,而非估值背书
上调触发条件估值回到约 10x-15x 收入区间,且条款干净需要私有证据证明收入或已签约利用率,并接受优先权 / 契约条件只有尽调显示真实安全边际,才积极重启
下调触发条件债务条款强硬,或利用率证据薄弱如果经济性主要靠叙事支撑,股权逻辑会迅速恶化规模出现前若定价风险继续叠加,转向规避

本表把公开记录转化为投资立场;推荐值是作者判断,不是管理层指引。

[CV002, CV015, CV016, CV037, CV048, CV049]
论点破裂与终止触发条件表
触发因素阈值 / 事件如何传导到论点行动含义
债务经济性不及预期债务定价、契约或摊还吃掉过多现金流规模扩张从战略资产变成股权负担除非价格重置,否则从继续研究转向规避
利用率爬坡失败针对 20,000-GPU 计划,没有签约需求或健康利用率证明Capex 不再像在修护城河,而像闲置库存要求已签约利用率,否则不推进
超大规模云厂商 TCO 差距收窄AWS/Azure/GCP 定价或打包消除 Neysa 的成本 / 合规优势压缩利润率,并削弱主权细分市场按低倍数基础设施重新承销,而非软件附加云
集中度过高一两个客户或行业主导需求估值容易受流失、内建替代或政策变化冲击要求集中度折价和更紧的敞口上限
采购或电力滑坡GPU 配额、数据中心容量或支持基础设施延迟到位债务或赞助方回报时钟继续走,收入兑现却被推迟执行赶上前,把预期价值推向熊市情景

终止触发条件聚焦那些会同时损害收入兑现和倍数选择的变量。

[CV020, CV021, CV043, CV044, CV045, CV046]
最终尽调问题表
主题缺失证据为什么重要负责人 / 尽调路径
收入和利润率当前 ARR / 收入运行率、毛利率、推理与裸金属业务组合缺少这些,就无法可信套用可比区间要求月度 cohort 收入、毛利和利用率桥接表
客户集中度Top-10 客户、预留容量承诺、续约条款决定收入是耐久还是脆弱审查客户 cohort 数据和合同摘要
债务包计划中 $600M 债务的定价、担保、摊还、契约和交叉违约条款债务会改变同一公开估值下的经济价值获取贷款方条款清单和下行情景模型
优先权结构清算优先权、ratchet 调整权、反稀释、董事会权利和同意权需要借此把企业价值转换为股权价值审查最终股权文件和股权结构表分配瀑布
采购和电力GPU 配额函、交付时间表、托管 / 电力承诺检验 20,000-GPU 目标能否按时落地检查 OEM、托管和电力协议
后续退出证据赞助方资本重组、战略收购或公开市场路径证据退出路线影响可接受入场价格和持有期收入和结构验证后,再搭建退出地图

这些问题按其对估值判断的直接影响排序,而不是按一般运营好奇心排序。

[CV015, CV016, CV039, CV049, CV051]

8.5 展示材料

免责声明

本报告仅综合公开信息用于尽调分流,不构成投资建议。私营公司的经济性、利用率和融资条款在公开记录中仍不完整,应直接向管理层和领投方核实。

证据索引

结论
编号陈述可信度来源
CO001 Neysa publicly dates its founding to 2023. SO006, SO019, SO028
CO002 Tracxn identifies the legal entity as Neysa Networks Private Limited and lists its incorporation date as 2022-12-16. SO028
CO003 Neysa’s headquarters and registered address are at Art Guild House, Phoenix Marketcity Kurla, Kurla West, Mumbai. SO003, SO028
CO004 Neysa publicly lists offices in Mumbai, Bengaluru, and Chennai. SO003, SO021
CO005 Neysa describes itself as an AI Acceleration Cloud provider focused on enterprise AI infrastructure rather than consumer AI applications. SO001, SO002, SO006
CO006 Velocis is Neysa’s flagship platform and combines infrastructure, inference, orchestration, observability, optimization, and AI/ML security into one stack. SO006, SO008, SO019
CO007 Public profiles describe Neysa’s core products as GPU-as-a-Service, AI Platform-as-a-Service, and Inference-as-a-Service. SO017, SO028
CO008 Neysa’s GPU catalog and pricing pages show L4, L40S, H100, H200, and AMD MI300X instances across managed VM, bare-metal, and Kubernetes deployment models. SO017, SO018
CO009 Neysa explicitly frames its platform as sovereign compute built and operated within India. SO006, SO008, SO023
CO010 Sharad Sanghi is Neysa’s co-founder and CEO. SO005, SO019, SO021
CO011 Anindya Das is Neysa’s co-founder and CTO. SO005, SO019, SO025
CO012 BV Jagadeesh is publicly presented as Neysa’s chairman. SO001, SO025
CO013 Sanghi and Das are presented as long-time infrastructure operators whose prior work together ran through Netmagic and the NTT ecosystem. SO005, SO020, SO026
CO014 Neysa appointed former Wipro and YES Bank CIO Anup Purohit as strategic advisor in June 2026. SO019
CO015 Neysa raised a $20 million seed round in March-April 2024. SO005, SO025, SO028
CO016 The seed investors publicly associated with Neysa were Z47/Matrix Partners India, Nexus Venture Partners, and NTTVC. SO005, SO025, SO028
CO017 Neysa raised a $30 million Series A on 2024-10-22. SO004, SO028
CO018 Neysa’s Series A was co-led by NTTVC, Z47, and Nexus Venture Partners. SO004, SO025, SO028
CO019 Blackstone and co-investors agreed to invest up to $600 million of primary equity in Neysa in February 2026. SO006, SO021, SO027
CO020 Neysa said the Blackstone transaction would support an additional planned $600 million debt financing. SO006, SO021, SO027
CO021 Public market-data and press sources pegged Neysa’s valuation around $1.4 billion in the February 2026 round. SO024, SO027, SO028, SO029
CO022 The Blackstone-led financing gave Blackstone a majority stake in Neysa. SO021, SO025
CO023 The February 2026 investor group also included Teachers’ Venture Growth, TVS Capital, 360 ONE, and Nexus Venture Partners. SO006, SO021, SO028
CO024 Earlier public cap-table disclosures and profiles also named Z47, NTTVC, Blume Ventures, and Anchorage Capital/Group among Neysa backers. SO027, SO028
CO025 By October 2024 Neysa said it had paying customers across AI-first digital natives, media and entertainment, service providers, software vendors, and the public sector. SO004
CO026 Forbes India reported that Neysa was working with over 20 customers and pilots across India and global markets. SO026
CO027 Neysa’s case studies show production use in research, retail payments, wealth management, and airline-travel automation workloads. SO010, SO011, SO012, SO013
CO028 Neysa’s public materials reference deployments or direct experience with HDFC Bank, PhonePe, Juspay, and Fractal Analytics. SO001, SO003
CO029 Blackstone’s press release says Neysa’s customers span financial services, technology, healthcare, and public services. SO006
CO030 Neysa positions itself as serving enterprises, startups, and public-sector organizations. SO006, SO008, SO019
CO031 TechCrunch reported that Neysa had about 1,200 GPUs live in February 2026. SO021
CO032 SiliconANGLE reported that Neysa’s platform was powered by about 2,000 GPUs in February 2026. SO022
CO033 Neysa and multiple news reports say the company plans to deploy more than 20,000 GPUs in India over time. SO006, SO021, SO025
CO034 Outsource Accelerator reported that NTT Data, Neysa, and Telangana signed an April 2025 MOU for a 400MW Hyderabad AI data-center cluster. SO030
CO035 The Hyderabad project was described as targeting 25,000 GPUs and Rs 10,500 crore of investment. SO022, SO030
CO036 Sacra says Neysa offers pre-wired capacity in Mumbai and Bangalore data centers. SO029
CO037 Publicly available evidence supports Chennai as an office location, but this chapter did not verify Chennai as a current Neysa data-center node. SO003, SO021
CO038 Neysa says Velocis launched in July 2024 and was generally available by October 2024. SO004
CO039 Neysa and Data Science Wizards announced an insurance-cloud partnership in December 2024 to target Indian insurers. SO007
CO040 Neysa and Pipeshift launched India-based real-time inference infrastructure in May 2026. SO008
CO041 The Pipeshift partnership says prompts, inference, and enterprise data remain within India. SO008
CO042 TechCrunch reported that Neysa employed 110 people across Mumbai, Bengaluru, and Chennai in February 2026. SO021
CO043 Tracxn listed Neysa’s latest employee count at 122 as of 2026-05-01. SO028
CO044 Tracxn also showed 97 employees as of August 2025, implying rapid hiring through the Blackstone period. SO028
CO045 Exact current headcount remains unresolved because public point estimates differ between 110 and 122 employees. SO021, SO028
CO046 Fortune India reported that Neysa had deployed about $44 million of its first $50 million into GPU infrastructure before the Blackstone round. SO026
CO047 Fortune India also reported that some of India’s largest private banks were already using Neysa before the February 2026 megaraise. SO026
CO048 Moneycontrol described Neysa as a domestic alternative to AWS and Azure for local AI workloads. SO023
CO049 Sacra highlighted hyperscaler price competition and regulatory shifts as material risks to Neysa’s economics. SO029
CO050 Sacra also said planned debt financing creates leverage and execution risk if utilization or pricing disappoints. SO029
CO051 Neysa tied the Blackstone round to IndiaAI-mission and India AI Impact Summit narratives around domestic compute buildout. SO006, SO009
CO052 No public revenue or ARR figure was verified for Neysa in this chapter.
CM001 India approved the IndiaAI Mission with a budget outlay of Rs.10,371.92 crore and an initial public AI compute infrastructure target of 10,000 or more GPUs built through public-private partnership. SM001, SM019
CM002 The IndiaAI Mission is designed as a full ecosystem program covering compute, indigenous models, datasets, application development, skills, startup financing, and safe-and-trusted AI rather than as a compute-only subsidy. SM001, SM003
CM003 The India AI Governance Guidelines state that more than 38,000 GPUs had been onboarded through a subsidized national compute facility by February 2026. SM003, SM015
CM004 ETGovernment reported that India planned to add another 20,000 GPUs to the 38,000-plus already provisioned under the IndiaAI Mission, implying sovereign compute capacity of roughly 58,000 GPUs. SM015
CM005 India's sovereign-AI strategy is explicitly framed around democratised compute access, indigenous model development, and institutional control rather than dependence on foreign platforms alone. SM001, SM003, SM016
CM006 The Reserve Bank of India requires payment-system data to be stored only in India, creating a concrete localization driver for domestic AI and cloud infrastructure in financial workloads. SM004, SM019
CM007 S&P Global argues that India's data-center and AI-infrastructure demand has been strengthened by the RBI localization rule, the 2023 Digital Personal Data Protection Act, and the 2024 launch of the IndiaAI Mission. SM019
CM008 Oracle already operates two OCI public-cloud regions in India—Mumbai and Hyderabad—and explicitly treats India as a two-region country for commercial-cloud business continuity. SM007
CM009 Google Cloud markets AI-optimized infrastructure, global regions, and SLA-backed data residency for foundational and agentic workloads, which defines the baseline local providers must compete against. SM006
CM010 AWS frames cloud competition around broad regional service breadth that includes EC2, EKS, Lambda, Redshift, and SageMaker, reinforcing how wide the hyperscaler feature set is relative to local GPU specialists. SM005
CM011 Financial Express reported that Microsoft committed $17.5 billion to India between 2026 and 2029, on top of the $3 billion announced earlier, to expand cloud and AI infrastructure and sovereign digital capabilities. SM021, SM022
CM012 Microsoft's India South Central region in Hyderabad is expected to go live in mid-2026 as the company's largest cloud region in India, alongside expansion of existing regions in Chennai, Hyderabad, and Pune. SM021, SM022
CM013 CRN Asia reported that AWS has committed $8.3 billion to its Mumbai region, indicating that hyperscalers are still investing heavily in India even as local AI clouds scale up. SM021
CM014 CRN Asia reported that Google is building a $15 billion AI hub in Visakhapatnam with AdaniConneX and Nxtra by Airtel, extending India's AI infrastructure buildout beyond the traditional metro markets. SM021
CM015 Yotta's Shakti Cloud says it is India's sovereign AI cloud and advertises the country's largest NVIDIA deployment with 8,000-plus H100 GPUs and future B200 capacity. SM008
CM016 Shakti Cloud says its Microsoft-aligned sovereign infrastructure is designed to satisfy the DPDP Act and Indian data-residency norms for enterprise AI workloads. SM004, SM008
CM017 Neysa's pricing page lists H100 SXM instances from $4.39 per hour and H200 SXM from $4.73 per hour and claims up to 70% lower total cost of ownership versus general-purpose hyperscalers. SM009
CM018 Neysa positions its AI cloud as India-built, GPU-first, and deployable across private, hybrid, and public cloud models rather than as a generic public-cloud substitute. SM009, SM010
CM019 Neysa's homepage uses customer examples to argue that local infrastructure matters when buyers need lower latency, India-specific model performance, predictable cost, and data residency that hyperscalers do not fully solve. SM010
CM020 Blackstone, Neysa, and multiple news reports say Neysa plans to deploy more than 20,000 GPUs in India using a $1.2 billion financing package. SM011, SM012, SM013, SM014, SM023, SM024
CM021 TechCrunch reported that Blackstone estimates India currently has fewer than 60,000 GPUs deployed and expects that figure to scale to more than 2 million in the coming years. SM012, SM013
CM022 TechCrunch says demand for domestic AI compute in India is being driven by government programs, regulated sectors that need local data, and AI developers or labs seeking lower-latency in-country capacity. SM012, SM023
CM023 Arizton values the India data-center market at $9.79 billion in 2025 and projects it to reach $21.03 billion by 2031 at a 13.59% CAGR. SM020
CM024 JLL projects that India will add 604 MW of data-center capacity between H2 2024 and 2026, requiring 7.3 million square feet of space and $3.8 billion of capital investment. SM017
CM025 JLL identifies Navi Mumbai as a pre-leasing hotspot with potential demand of about 800 MW, illustrating how AI-cluster demand is concentrating around power-rich campuses rather than evenly across India. SM017
CM026 Cushman & Wakefield says India has 1.6 GW of operational data-center capacity and 3.1 GW under construction or planned, with Mumbai set to exceed 1 GW of operational capacity by end-2026 and Hyderabad ranking ninth globally among secondary markets. SM018
CM027 Cushman says power availability, execution capability, land access, and regulatory readiness are now central competitive variables in AI-led data-center expansion. SM018
CM028 S&P Global estimates India's datacenter electricity demand at about 13 TWh in 2024 and projects it to reach 57 TWh by 2030, alongside more than 5 GW of additional IT load capacity. SM019
CM029 S&P expects India to become the second-largest datacenter electricity-demand market in Asia-Pacific within two years, surpassing Japan and Australia. SM019
CM030 S&P says Maharashtra, Telangana, and Karnataka account for about 70% of India's operating data-center capacity, showing that supply remains geographically concentrated. SM019
CM031 S&P says energy costs represent about 65% of datacenter operating expense in India, making power economics central to neo-cloud pricing and margins. SM019
CM032 S&P says water availability is a growing risk in Mumbai, Bengaluru, and Chennai and cites Uptime Institute data that a 1 MW data-center load can need about 25.5 million liters of cooling water per year. SM019
CM033 S&P estimates that India may need 15 GW to 30 GW of additional renewable capacity over the next five years to satisfy projected datacenter electricity demand. SM019
CM034 EY argues that sovereign AI in India requires domestic infrastructure, local data control, talent, legal frameworks, and cybersecurity because dependence on foreign platforms creates resilience and security risks. SM016
CM035 The IndiaAI Mission makes startups, researchers, and public-interest AI applications explicit target users of domestic compute, meaning early market demand is not limited to large enterprises. SM001, SM003
CM036 CRN Asia reports that AI startups, GCCs, and mid-sized enterprises are active buyers of AI infrastructure and that these deployments are smaller than hyperscale campuses but more accessible to mid-market execution partners. SM021
CM037 CRN Asia cites IDC in projecting that India's public-cloud services market will reach $30.4 billion by 2029, which is a useful adjacency but materially broader than Neysa's direct AI-infrastructure market. SM021
CM038 Moneycontrol describes Neysa as a domestic alternative to AWS and Azure for GPU-as-a-service and local data hosting, especially for regulated and high-performance workloads. SM023
CM039 Shakti Cloud claims multiple in-country regions, RBI-, ISO-, and SOC-certified data centers, and low-latency nationwide access for regulated AI workloads. SM008
CM040 Financial Express reports that Microsoft launched sovereign public-cloud and sovereign private-cloud offerings for Indian organizations alongside the new infrastructure commitment. SM022
CM041 The IndiaAI portal shows the public program is building an ecosystem around initiatives, datasets, standards, research, startups, and companies rather than only around raw compute allocation. SM002
CM042 Moneycontrol says many Indian firms still rely on foreign cloud providers for AI workloads and therefore face privacy-compliance, latency, and cost concerns when infrastructure sits outside India. SM023
CM043 CRN Asia says buyers in banking, healthcare, and government are placing data-residency and sovereignty requirements on infrastructure decisions, affecting site choice, architecture, and partner selection. SM021
CM044 No public source in the current evidence set cleanly isolates India AI-cloud or GPU-infrastructure revenue by buyer segment, so SAM must be framed through adjacent cloud/data-center markets plus sovereign-compute and operator-capacity proxies. SM017, SM018, SM020, SM021
CM045 Applying S&P's disclosed IndiaAI floor price of $1.36 per GPU hour to 34,371 awarded GPUs implies an annualized public-compute spend equivalent of about $0.41 billion at full utilization. SM019
CM046 Applying the same $1.36 per GPU hour rate to the 38,000-plus GPUs cited in the 2026 governance guidelines implies an annualized public-compute floor of about $0.45 billion at full utilization. SM003, SM019
CM047 Applying the same $1.36 per GPU hour rate to the announced 58,000 sovereign GPUs implies an annualized public-compute floor of about $0.69 billion at full utilization. SM015, SM019
CM048 Neysa's planned 20,000-plus GPUs and Yotta's advertised 8,000-plus H100 GPUs imply that just two local operators already represent more than 28,000 GPUs of identifiable domestic AI-cloud supply. SM008, SM011, SM024
CM049 CRN Asia reports that roughly 30 large data-center projects were announced across India between March 2025 and April 2026, adding about 3.5 GW of planned capacity, with Andhra Pradesh and Telangana accounting for more than 2 GW. SM021
CM050 Taken together, Microsoft's sovereign-cloud launch, RBI localization rules, and local neo-cloud product positioning show that sovereignty in India is now a procurement design principle, not just a branding theme. SM004, SM008, SM022, SM023
CM051 The IndiaAI portal confirms that the public ecosystem includes datasets, standards, initiatives, startups, and companies, supporting the view that demand creation is being orchestrated at the ecosystem layer as well as the infrastructure layer. SM002
CM052 AWS treats regional infrastructure as the delivery unit for a broad menu of core cloud services, including SageMaker, which underscores how much of Neysa's differentiation must come from specialization rather than feature breadth. SM005
CM053 Google Cloud explicitly markets its regions for AI-powered and agentic workloads with data-residency benefits, confirming that generic cloud incumbents are also competing for the sovereign and AI-sensitive use cases Neysa targets. SM006
CM054 Yotta Labs operates a dedicated GPU-cloud portal, showing that Indian AI-cloud competition is moving toward productized self-service access rather than remaining a purely custom-enterprise sales motion. SM025
CP001 Neysa publicly positions itself as a full-stack AI cloud that combines compute, MLOps, and AI security for enterprise production AI and regulated sectors. SP001
CP002 Neysa says Velocis offers H100 SXM, H100 NVL, H200 SXM, L40S, L4, and AMD MI300 capacity that can be consumed as bare metal, VMs, or managed Kubernetes. SP001, SP002
CP003 Neysa says its AI fabric is built on RoCEv2 at 3.2 Tb/s per node with 1:1 bisection bandwidth. SP001
CP004 Neysa says its reserved rate covers compute, storage, egress, Kubernetes, Jupyter, MLflow, Weights & Biases, and inference endpoints in one fixed commercial package. SP001
CP005 Neysa’s public pricing page showed inventories of 392 H100 SXM, 200 H200 SXM, 208 L40S, 104 L4, and 32 AMD MI300 units when fetched on 2026-06-25. SP002
CP006 Yotta markets Shakti Cloud as a sovereign AI cloud hosted entirely in India for enterprises, researchers, and startups. SP003
CP007 Yotta’s official and NVIDIA-partner materials say Shakti Cloud powers India’s largest AI deployment with 8,000+ H100 GPUs and additional B200 capacity coming soon. SP003, SP007
CP008 Yotta claims Shakti Cloud delivers 99.95% of NVIDIA benchmark performance and trained Llama 3.1 70B on 256 H100 GPUs at 99.5% of NVIDIA speed-of-light performance. SP003
CP009 Yotta’s pricing page lists monthly H100 AI Lab configurations from ₹27,000 for a 10GB slice to ₹1,504,000 for an 8x80GB workstation and adds ₹70,000 for platform access with unlimited ingress and egress. SP004
CP010 Yotta says Shakti Cloud integrates Azure OpenAI, Azure ML, VS Code, and GitHub Copilot on sovereign infrastructure for Microsoft-native buyers. SP003
CP011 E2E markets itself as India’s GPU cloud for AI and ML and says it runs Indian data centers with MeitY empanelment. SP009, SP010
CP012 E2E publishes list prices of ₹624 per hour for B200, ₹300 per hour for H200, ₹249 per hour for H100, and ₹49 per hour for L4 capacity. SP010, SP011
CP013 E2E’s pricing page claims transparent per-minute or hourly pricing, pay-as-you-go or reserved plans, and pricing that is 60% cheaper than hyperscalers. SP011
CP014 E2E’s GPU cloud page highlights SOC 2 Type II, ISO 27001, ISO 27017, PCI DSS, and MeitY empanelment as trust signals. SP010
CP015 CNBC TV18 reported that E2E started executing a ₹177 crore IndiaAI Mission order with H100 SXM and H200 SXM GPUs allocated to Gnani AI and go-live targeted for January 2026. SP012
CP016 IndiaAI’s official compute-capacity page says eligible users can access AI compute at up to 40% reduced cost and references 18,000+ affordable AI compute units. SP013, SP031
CP017 Official IndiaAI allocation records include a 4,096-GPU Sarvam AI allocation through Yotta and multiple allocations through E2E and NxtGen. SP013
CP018 Fortune India reported that the IndiaAI Mission’s total empanelled GPU pool reached 34,333 after a second round that added about 16,000 GPUs. SP031
CP019 Communications Today reported first-phase commitments of 18,693 GPUs against a 10,000 target and identified Yotta as the largest L1 bidder with 9,216 GPUs. SP033
CP020 AWS’s India page says AWS operates Mumbai and Hyderabad regions and Local Zones in Delhi and Kolkata to address latency and data-residency needs. SP016
CP021 AWS says P5 uses NVIDIA H100 and P5e/P5en use NVIDIA H200, with up to 4x faster time to solution and up to 40% lower training cost than previous-generation GPU instances. SP015
CP022 Azure’s ND-family documentation says those GPU instances are designed for AI training, inference, research, and HPC. SP017
CP023 Azure’s data-residency page says most services stay in the selected geo, but Azure Machine Learning metadata may be stored in the United States and some Foundry model deployment types may process prompts and completions outside the selected geo. SP018
CP024 Google Cloud says it operates 43 regions and 130 zones globally and markets SLA-backed regions for data residency. SP020
CP025 The Google pricing pages reviewed expose AI infrastructure pricing through SKU and region tables rather than a simplified India-sovereign AI bundle. SP021, SP022
CP026 Tata says its sovereign cloud hosts data and metadata within India and aligns with MeitY, SEBI, RBI, and IRDAI mandates. SP023, SP024
CP027 Tata says its AI GPU cloud offers bare-metal H100, H200, and L40S with non-blocking InfiniBand and throughput up to 10x faster than standard PNFS. SP024
CP028 Tata’s official and independent materials claim 50K tokens per second throughput, a 99.9% SLA, no surprise egress fees, and up to 30% lower cloud costs. SP024, SP025, SP026
CP029 Tata Vayu is positioned as a unified cloud fabric combining IaaS, PaaS, AI, security, and connectivity for enterprise and government workloads. SP025, SP026
CP030 Sify launched CloudInfinit+AI as a GPU-as-a-Service platform for AI training, inferencing, analytics, rendering, and scientific simulation on a pay-as-you-go model. SP027
CP031 CoreWeave says its AI-native cloud combines Kubernetes-native GPU compute, purpose-built storage and networking, managed software services, and cluster health management. SP028
CP032 CoreWeave’s S-1 contrasts its AI infrastructure with generalized clouds and describes integrated software for provisioning AI infrastructure and orchestrating AI workloads. SP029
CP033 None of the retained CoreWeave sources provides evidence of India-sovereign hosting, IndiaAI empanelment, or India-specific regulated-workload positioning. SP028, SP029
CP034 TechCircle reported that the government planned to add 20,000 GPUs to an existing base of 38,000 under IndiaAI, signaling continued expansion of India-hosted compute supply. SP030
CP035 Tech Funding News linked NVIDIA’s IndiaAI work with 20,000 GPUs and sovereign-cloud partners, indicating that national industrial policy is broadening the competitive field. SP032
CP036 Neysa’s own comparison page argues Yotta’s control plane, support, or interconnect may be billed separately and that Yotta’s public MLOps and AI-security story is thinner than Neysa’s. SP001
CP037 Neysa’s comparison page argues E2E wins on self-service elasticity but still charges separately for surrounding services such as storage, firewall, load balancer, VPC, SSL, compliance, and egress. SP001
CP038 Neysa’s comparison page argues Tata’s enterprise delivery is mature, but public GPU pricing is less transparent and the AI Studio layer is earlier-stage than Neysa’s stack. SP001
CP039 Across the retained sources, domestic providers compete on DPDP-style data-residency messaging, India billing, and regulated-sector access, while hyperscalers compete on ecosystem breadth and adjacent services. SP003, SP010, SP016, SP018, SP020, SP023
CP040 Switching costs are highest for customers already standardized on hyperscaler ecosystems such as AWS platform services, Azure ML and Foundry, or Google’s global pricing and routing stack. SP015, SP018, SP020, SP021
CP041 Switching costs within domestic clouds are lower at the infrastructure layer because several competitors market Kubernetes or cluster-based GPU access, but reserved capacity, workflow habits, and compliance approval still create stickiness after deployment. SP003, SP010, SP024, SP028
CP042 The immediate competitive threat to Neysa is a broadening domestic field—especially Yotta, E2E, Tata, and other IndiaAI-linked providers—rather than only the three global hyperscalers. SP013, SP031, SP033
CI001 Neysa announced a 2026 capital raise of up to $1.2 billion. SI002, SI003, SI004, SI005
CI002 The announced package pairs up to $600 million of equity with an intended additional $600 million of debt financing, subject to documentation. SI002, SI003, SI004, SI006
CI003 Blackstone is publicly described as taking a majority stake in Neysa alongside Teachers’ Venture Growth, TVS Capital, 360 ONE, and Nexus Venture Partners. SI003, SI004, SI005
CI004 TechCrunch reported that most of the new capital is earmarked for compute, networking, and storage expansion, with a smaller portion for research and software platform build-out. SI004, SI010
CI005 The 2026 financing announcement ties the raise to a planned deployment of more than 20,000 GPUs in India. SI002, SI003, SI006
CI006 Before the 2026 round, Neysa had publicly announced a $20 million seed round in March 2024 and a $30 million Series A in October 2024. SI016, SI008
CI007 Independent trackers describe Neysa as having raised about $650 million of equity funding to date, excluding the planned debt tranche. SI007, SI008
CI008 Filing-derived company registries show Neysa Networks Private Limited with authorized capital of about ₹5.005 crore and paid-up capital of about ₹1.95 crore, with the latest balance sheet dated March 31, 2025. SI017, SI018
CI009 Neysa’s public monetization stack includes managed VM instances, bare-metal GPU servers, managed Kubernetes clusters, and public, private, and hybrid deployment models. SI001, SI013, SI014
CI010 Neysa publicly lists L4, L40S, H100, H200, and MI300X GPU configurations across VM and 8-GPU bare-metal offers. SI001, SI014
CI011 Neysa discloses both hourly/on-demand and term-commit pricing, including 1- to 36-month commit options. SI001, SI014
CI012 Neysa advertises reserved-capacity savings of up to 40% relative to on-demand rates. SI001
CI013 Neysa says customers pay for GPU compute consumption without separate ingress, egress, or inference-transaction fees on the public pricing page. SI001
CI014 Neysa positions its platform as offering 40% to 60% lower TCO or unit economics than general-purpose hyperscalers. SI012, SI014
CI015 Neysa’s public product surfaces combine GPU infrastructure with AI Studio, orchestration, observability, inference, and security, implying platform-layer monetization beyond bare compute. SI013, SI025, SI026
CI016 The Pipeshift partnership extends Neysa’s monetization path into dedicated managed inference endpoints delivered through OpenAI-compatible APIs. SI026, SI013
CI017 Regulated sectors are a central GTM wedge because Neysa explicitly pitches sovereign or audit-ready infrastructure to BFSI, public-sector, healthcare, and compliance-sensitive buyers. SI015, SI002, SI021
CI018 Public sources describe Neysa customers across financial services, technology, healthcare, and public services. SI002, SI003, SI006
CI019 Neysa’s Series A release says the company had secured orders from paying customers across AI-first digital natives, media, service providers, software vendors, and the public sector. SI016
CI020 TIFIN says moving to Neysa cut its GPU cloud spend by 65% versus hyperscalers. SI021
CI021 Innoviti says moving production inference to Neysa reduced total cost of ownership by 60% versus a general-purpose cloud setup. SI020
CI022 ITQ says Neysa reduced total cost of ownership by 40% versus general-purpose cloud for a high-volume inference workload. SI022
CI023 TIFIN reports 99.95% system uptime on Neysa Velocis for regulated financial workloads. SI021
CI024 ITQ reports Neysa reduced P99 latency from 14 seconds to under two seconds while sustaining roughly 2,500 tokens per second throughput. SI022
CI025 Innoviti says Neysa-supported workflows handle 7,000-plus service ticket logs per day and keep per-report processing under 30 seconds. SI020
CI026 Nurix AI reported a 3x reduction in time to first token on the Neysa-Pipeshift inference stack. SI026
CI027 Neysa and Pipeshift say production deployments can move from evaluation to production in under two weeks. SI026
CI028 The IISc Bangalore case study says Neysa bare metal supported continuous generation of 33 million sketch-image pairs plus full 7B and 13B model training without queue contention. SI019
CI029 Customer testimonials on Neysa’s home page repeatedly frame hyperscaler cost, latency, configurability, or compliance limits as the reason to switch. SI011
CI030 Neysa’s published hardware offer is asset-heavy, combining high-end GPUs with NVMe storage, high-core-count servers, and 1600 to 3200 Gb/s interconnects. SI014, SI001
CI031 TechCrunch reported Neysa had about 1,200 GPUs live before the Blackstone financing closed. SI004
CI032 Using the announced $600 million debt tranche alone implies about $30,000 of capital per targeted GPU, while allocating the full $1.2 billion package implies about $60,000 per targeted GPU; both are heuristic upper bounds because some capital also funds software and R&D. SI002, SI004, SI005
CI033 Indian GPU-backed debt typically finances 40% to 70% of GPU value and can carry interest rates up to 14%, according to NewsBytes citing local loan economics after Neysa’s raise. SI030
CI034 CoreWeave’s 2025 S-1 shows the specialized GPU-cloud model can scale quickly but still be loss-making at scale, with 2024 revenue of $1.9 billion and net loss of $863 million. SI027
CI035 CoreWeave disclosed total debt commitments of $12.9 billion through December 2024 and highlighted asset-backed debt as a core financing tool for capacity growth. SI027
CI036 CoreWeave says a majority of AI compute capacity is lost to system inefficiencies, with observed performance often in the 35% to 45% range of peak FLOPs. SI027
CI037 Compute Forecast argues H100 rental rates fell roughly 64% to 75% from peak to about $2.99 per hour, materially weakening the revenue assumptions behind shortage-era GPU debt. SI028
CI038 Compute Forecast argues GPU collateral can lose economic relevance within 18 to 24 months as new hardware generations alter the workloads buyers are willing to pay for. SI028
CI039 CNBC reports a “GPU debt treadmill” concern in which lenders finance long-lived data center projects against GPUs with shorter useful lives and recurring upgrade pressure. SI029
CI040 Neysa does not publicly disclose revenue, ARR, gross margin, EBITDA, cash balance, backlog, or utilization metrics on the evidence reviewed for this chapter. SI007, SI008, SI017, SI018
CI041 Tofler places Neysa’s revenue band at ₹10 crore to ₹25 crore on the latest public company-detail page. SI018
CI042 Tracxn places NEYSA NETWORKS PRIVATE LIMITED revenue at ₹10 crore to ₹50 crore as of March 31, 2025 and lists employee count at 122 as of May 1, 2026. SI008
CI043 The public third-party revenue bands are broad and unaudited, so they are directional traction proxies rather than underwriteable revenue evidence. SI018, SI008
CI044 Neysa’s pricing and case studies imply that attractive unit economics require high occupancy on reserved or private deployments so depreciation, power, networking, and support costs are spread across committed workloads. SI021, SI020, SI022, SI027
CI045 The medtech case study frames Neysa as moving a customer from restrictive CAPEX procurement into a predictable OPEX model built around long-term H100 bare-metal commitments. SI023
CI046 Sacra flags hyperscaler price competition and regulatory shifts as the most important strategic risks to Neysa’s economics. SI007
CI047 Because Neysa has not disclosed debt tenor, coupon, collateral package, covenants, or contracted backlog, outside investors cannot judge whether its planned borrowing resembles refinanceable project finance or spot-market GPU leverage. SI028, SI029, SI002
CI048 Financially, Neysa looks stronger than a pure spot GPU marketplace because it layers reserved capacity, private cloud, and managed inference onto a regulated-enterprise GTM motion, but the business still screens as capital-intensive and financing-dependent until it discloses utilization, margin, and debt-service coverage. SI016, SI020, SI021, SI022, SI026, SI028, SI029
CI049 TechCrunch reported that Neysa aims to more than triple revenue next year, but the company did not disclose the baseline revenue figure. SI004
CI050 Public source counts vary between about 1,200 live GPUs and about 2,000 GPUs, so Neysa’s installed base should be treated as approximate until management reconciles the metric and date stamps. SI004, SI010
CE001 Velocis is publicly positioned as a full-stack AI acceleration cloud that combines GPU infrastructure, an AI platform layer, cost governance, observability, and security rather than only raw compute rentals. SE001, SE002
CE002 Neysa says Velocis supports the full lifecycle from training and fine-tuning through deployment and production inference. SE001, SE002
CE003 Velocis can be deployed in public cloud, private cluster, or hybrid modes. SE002, SE003
CE004 The public architecture names AI cluster management as a core component of the Velocis control plane. SE003
CE005 The public architecture names an AI scheduler and resource manager as distinct orchestration functions. SE003
CE006 Velocis advertises support for GPUs delivered as bare metal, virtual machines, or containers. SE003
CE007 Neysa explicitly lists GitHub or GitLab, Docker, MLflow, Kubeflow, Airflow, and enterprise IAM connections in the public architecture surface. SE003
CE008 Neysa says every core function is exposed through secure APIs so Velocis can connect into CI/CD pipelines, monitoring stacks, IDEs, and existing ML workflows. SE003
CE009 Velocis markets pre-integrated developer environments and open-source toolchains including Jupyter, PyTorch, Hugging Face, MLflow, and Kubeflow. SE002, SE003
CE010 The architecture page says Neysa can integrate identity and access, data and storage, Dev and MLOps, SIEM, and hybrid cloud connectivity into one stack. SE003
CE011 Neysa says its single dashboard exposes GPU utilization, disk utilization, NVMe allocation, and custom metrics for observability. SE002
CE012 Neysa documents a zero-trust security model in which services, users, and processes are authenticated and isolated by default. SE004
CE013 Neysa documents granular RBAC by project, persona, or asset with SSO and IAM integration. SE004, SE003
CE014 Neysa says every action, access event, and system event is logged and exportable for compliance teams. SE004
CE015 Neysa says data and model artifacts are encrypted at rest and in transit and can use customer-managed keys with enterprise KMS integration. SE004
CE016 Neysa publicly claims ISO/IEC 27001:2022 certification and SOC 2 compliance for the Velocis environment. SE004, SE015
CE017 The CSA STAR registry lists Neysa Velocis with both a Level 1 self-assessment record and a Level 2 certification record. SE015
CE018 Neysa repeatedly frames Velocis as infrastructure whose workloads and data remain inside Indian data centers or Indian legal jurisdiction. SE013, SE016, SE009
CE019 Public Neysa materials consistently tie that sovereign operating model to BFSI, healthcare, government, research, and voice-AI workloads that are sensitive to compliance, latency, or localization. SE013, SE010, SE001, SE024, SE025, SE026
CE020 Neysa publicly prices three H100 classes: a 10GB fractional slice, a 40GB fractional slice, and a full 80GB H100 SXM configuration. SE005
CE021 The published H100 pricing page shows a 10GB fractional H100 starting at $0.79 per hour on demand and as low as $0.36 per hour on a three-year term. SE005
CE022 The full H100 SXM configuration is publicly described with 48 vCPU, 288GB RAM, and 1000GB NVMe alongside the GPU. SE005
CE023 Neysa publicly references both H100 and H200 generation GPUs across product, blog, and case-study materials rather than only one accelerator generation. SE006, SE010, SE020
CE024 NVIDIA documentation shows why those SKUs matter by pairing H100 and H200 with high memory bandwidth, NVLink interconnects, and large-model training and inference performance characteristics. SE019, SE020
CE025 Neysa’s public architecture roadmap explicitly says Velocis is designed to add new GPU SKUs, new model formats, agents, fine-tuning, and vector databases over time. SE003
CE026 The Velocis product page marks the marketplace ecosystem as coming soon rather than generally available. SE002
CE027 The Neysa-Pipeshift partnership extends Velocis with single-tenant, OpenAI-compatible real-time inference for open-source models including Gemma, Qwen, Llama, DeepSeek, and Mistral. SE009, SE018, SE021
CE028 Neysa and Pipeshift say typical deployment timelines from evaluation to production are under two weeks. SE009
CE029 Neysa and Pipeshift say early deployments include Nurix AI, which reportedly achieved a threefold reduction in time to first token for voice-AI inference in India. SE009, SE024
CE030 Neysa and Pipeshift say Arrowhead AI had a fine-tuned model live as an inference endpoint within a day and also runs SLMs and ASR containers on the platform. SE009
CE031 TIFIN says it moved production AI workloads for training, experimentation, and inference onto Neysa Velocis. SE010, SE022
CE032 TIFIN says its Velocis deployment used NVIDIA H200, H100 SXM, and L40S GPU virtual machines backed by high-performance NVMe storage. SE010
CE033 TIFIN says the Neysa deployment delivered a 65% reduction in GPU cloud spend and 99.95% system uptime. SE010
CE034 TIFIN says its previous local neocloud provider suffered chronic outages and latency spikes, highlighting reliability as a core purchase criterion for Velocis-class workloads. SE010
CE035 Innoviti says it moved a custom Qwen 3.0 VL-powered multimodal verification workflow onto a dedicated inference stack on Neysa Velocis. SE011, SE023
CE036 Innoviti says the Neysa deployment delivered deterministic sub-30-second latency, 60% lower total cost of ownership, and 96% automated verification accuracy. SE011
CE037 Innoviti says the stack handled 50 parallel LLM inference requests and more than 7,000 service ticket logs per day across a 50,000-merchant network. SE011
CE038 IISc says it used Neysa bare-metal compute to generate 33 million sketch-image pairs and train 7B and 13B open-weight sketch vision models. SE012, SE028
CE039 IISc says the resulting O3SLM work was accepted at AAAI 2026 and beat GPT-4o and Gemini 1.5 Pro across four sketch benchmarks. SE012
CE040 Neysa’s public signals in 2026 include expanded case studies, conference demos, and whitepapers rather than a formal detailed public product roadmap. SE012, SE013, SE008
CE041 Neysa’s public job openings include Linux systems and storage, network and cloud security, SOC analyst, threat detection, and incident-management roles, implying ongoing investment in platform and security operations. SE014
CE042 Blackstone says the capital raise is meant to help Neysa deploy more than 20,000 GPUs in India and scale mission-critical AI infrastructure for enterprises and government entities. SE016, SE017
CE043 TechCrunch says Neysa had about 1,200 GPUs live at reporting time and that new capital would go into compute, networking, storage, and software for orchestration, observability, and security. SE017
CE044 Homepage customer quotations and partner references show Neysa being publicly associated with voice AI, banking speech-to-text, research, and retail operations rather than only generic cloud infrastructure marketing. SE001, SE024, SE025, SE026
CE045 Nurix, Smallest AI, and Navana all publicly describe production voice or enterprise conversational AI workloads, which corroborates the type of customers Neysa highlights for latency-sensitive inference. SE024, SE025, SE026
CE046 Neysa’s inference-endpoints materials describe autoscaling, telemetry, authentication, encryption, and compliance monitoring as baseline design patterns for the inference layer. SE007
CE047 Neysa’s neocloud whitepaper frames the stack around open integration, modular architecture, multi-deployment models, telemetry-driven scaling, and interoperability with hyperscalers. SE008
CE048 Neysa’s public product materials repeatedly prefer open-source and open-weight model workflows over dependence on proprietary black-box APIs. SE001, SE002, SE006
CE049 Digit’s India AI Summit interview with Neysa’s CPO says India still needs more GPU availability, new data centers, and enough power supply to keep pace with AI data-center rollout. SE027
CE050 TechCrunch says Neysa is expanding in a market still constrained by specialized-chip supply and data-center capacity, which means sector-level bottlenecks remain a live execution risk even after financing. SE017
CU001 Neysa explicitly targets banks, NBFCs, and fintechs for fraud detection, credit risk modeling, and customer intelligence on a compliant AI cloud. SU001, SU004
CU002 Neysa explicitly targets insurers for claims processing, underwriting, risk scoring, and compliance workflows. SU001, SU005
CU003 Neysa explicitly targets eCommerce and retail teams for recommendations, pricing intelligence, and churn prediction. SU001, SU006
CU004 Neysa explicitly targets manufacturers for predictive maintenance, quality inspection, and process optimization workloads. SU001, SU007
CU005 Neysa explicitly targets education and research institutions with AI labs, experimentation, and model-development infrastructure. SU001, SU008
CU006 Neysa explicitly markets fast GPU access, open-source compatibility, and usage-based pricing to AI-native startups. SU001, SU009
CU007 Neysa frames sovereign or in-region data handling as a core procurement requirement for regulated Indian AI customers. SU004, SU005, SU016
CU008 Neysa’s public pricing is designed to cover PoCs, testing, production, and variable workloads. SU002, SU009
CU009 Neysa says committed usage can save up to 40% and that customers avoid hidden egress, API-call, and inference-transaction fees. SU002
CU010 Neysa recruits channel partners, system integrators, MSPs, consulting teams, ISVs, and alliances to reach customers. SU003
CU011 Neysa offers discounts, referral fees, revenue sharing, co-marketing, and co-selling to influence partner-led customer acquisition. SU003
CU012 TIFIN India includes both MyFI and TIFIN India Enterprise, confirming an India-specific customer-side operating footprint. SU026, SU028
CU013 Neysa’s TIFIN case says TIFIN serves India’s largest mutual funds and wealth management firms. SU010, SU026
CU014 TIFIN said investor data and model workloads had to remain within India for tier-1 financial clients. SU010, SU016
CU015 TIFIN said its previous local neocloud provider had chronic outages and latency spikes before the move to Neysa. SU010
CU016 TIFIN said it cut GPU cloud spend by 65% compared with hyperscalers after moving onto Neysa Velocis. SU010
CU017 TIFIN said Neysa delivered 99.95% system uptime for its production AI workloads. SU010
CU018 TIFIN said Neysa shortened the path from experimentation to live commercial deployment. SU010
CU019 Innoviti’s field-support network covers more than 50,000 merchants across 2,000 cities and supports enterprise retailers including Reliance Retail, Shoppers Stop, and DMart. SU011, SU025
CU020 Innoviti says it processes over ₹80,000 crore annually on its payments network. SU011, SU025
CU021 Innoviti said the Neysa deployment moved its AI field-operations system from proof of concept into a production-ready environment. SU011, SU025
CU022 Innoviti reported a 60% reduction in total cost of ownership on the Neysa-backed deployment. SU011, SU025
CU023 Innoviti reported 7,000-plus service ticket logs processed daily, 96% automated verification accuracy, and deterministic sub-30-second latency. SU011, SU025
CU024 ITQ is Travelport’s exclusive regional partner across India, Sri Lanka, the Maldives, and Bhutan, connecting thousands of agencies to airline inventory. SU012
CU025 ITQ said its AI services had scaled to hundreds of billions of tokens per month. SU012
CU026 ITQ said self-hosting on general-purpose cloud GPUs produced poor unit economics and unpredictable performance while serverless frontier models were cost-prohibitive at its scale. SU012
CU027 The ITQ case positions Neysa as enterprise inference infrastructure for airline-policy interpretation at production volume. SU012
CU028 Neysa’s medtech case describes a global customer using the platform for medical robotics, AI-powered cancer diagnostics, and cell-therapy workloads. SU014
CU029 The medtech case says the customer had more than 100 data scientists and wanted to escape hardware procurement and management friction. SU014
CU030 Neysa’s IISc case and the O3SLM project page together show that the Visual Computing Lab used a Neysa-backed workflow for work labeled AAAI 2026. SU013, SU033
CU031 The O3SLM project page describes 7B and 13B variants, making the IISc proof more specific than a generic logo reference. SU013, SU033
CU032 Neysa’s Pipeshift press release frames the joint offer as real-time inference for open-source models fully deployed within India. SU015, SU021
CU033 Economic Times said the Neysa-Pipeshift offering had already been deployed with AI startups Nurix and Arrowhead AI. SU021
CU034 The GreyLabs partnership targets banking, insurance, and financial-services customers with enterprise-scale voice analytics. SU020
CU035 Elets said more than 90% of customer interactions in Indian BFSI are voice-based and that manual audits typically cover less than 1% of calls. SU020
CU036 WEKA says Neysa supports customers processing more than 100 million tokens daily. SU017
CU037 TechCrunch, Moneycontrol, and Entrackr all reported that Neysa had about 1,200 GPUs live and was targeting more than 20,000 over time. SU018, SU019, SU029
CU038 The Times of India named Juspay, Swiggy, and Perfios as key Neysa customers. SU023
CU039 The Times of India said Neysa sees demand from enterprises, startups, government bodies, research institutions, and global frontier labs. SU018, SU023
CU040 The reviewed public materials do not disclose NRR, GRR, churn, contract length, minimum commitments, or top-customer revenue share for Neysa. SU001, SU010, SU011, SU012, SU023
CU041 Neysa’s own vertical pages repeatedly describe pilot-to-production delays, data-localization review, GPU cost pressure, and integration overhead as customer buying frictions. SU004, SU005, SU006, SU007
CU042 ClusterMAX rated Neysa Bronze and said the platform had security, usability, onboarding, scheduling, and monitoring gaps relative to international competitors. SU024
CU043 Economic Times reported that a Neysa client still struggled with token costs and latency because the workload itself was unoptimized. SU021, SU022
CU044 Public customer proof is concentrated in a small set of verticals: wealth and fintech, payments and retail operations, travel distribution, research, and one unnamed medtech account. SU010, SU011, SU012, SU013, SU014, SU023
CU045 Neysa’s partner program plus the Pipeshift and GreyLabs examples show that channel influence is real, but public evidence does not reveal partner-sourced revenue contribution. SU003, SU020, SU021
CU046 Neysa’s transparent hourly pricing and reserved discounts likely lower friction for trial workloads, but production expansion still depends on compliance sign-off and workflow proof. SU002, SU004, SU005, SU010
CU047 Digit and Economic Times both describe 2026 as a period when Indian AI customers are moving from pilots toward production, which supports Neysa’s demand narrative but also implies immature deployment cohorts. SU030, SU031
CU048 WEKA, TechCrunch, and The Times of India together imply that Neysa’s visible customer mix spans startups, enterprises, research users, and regulated sectors, but not the revenue share of each group. SU017, SU018, SU023
CR001 Neysa’s February 2026 financing paired up to $600 million of primary equity with an intended $600 million of debt to support expansion. SR010, SR017, SR018
CR002 Management described the business as capital-intensive and indicated another fundraise was likely as additional infrastructure is deployed. SR016, SR018
CR003 Public reporting in February 2026 said Neysa had about 1,200 GPUs live and was targeting deployments of more than 20,000 GPUs over time. SR010, SR017
CR004 Neysa’s public pricing advertises material committed-use discounts versus on-demand H100 and H200 pricing. SR002, SR003
CR005 Neysa’s pricing page says it does not charge extra for data ingress, egress, or inference transactions. SR002
CR006 Economic Times reported a $1.4 billion enterprise valuation for the Blackstone transaction and a resulting majority stake for Blackstone. SR018
CR007 Business Standard quoted Sharad Sanghi saying Neysa’s installed GPU base was about 95% Nvidia with some AMD capacity. SR016
CR008 Sharad Sanghi told Economic Times that supply-chain resilience was required to access GPUs quickly. SR018
CR009 AWS says its P5, P5e, and P5en UltraClusters can scale to 20,000 H100 or H200 GPUs. SR027
CR010 Azure says ND H100 v5 deployments can scale to thousands of GPUs with 3.2 Tbps of interconnect bandwidth per VM. SR029
CR011 Google Cloud markets multiple accelerator-optimized machine families for AI training, fine-tuning, and inference with several consumption models. SR028
CR012 Neysa markets 8xH100 and 8xH200 bare-metal nodes with 3200 Gbps bandwidth and instant deployment positioning. SR004, SR005
CR013 NVIDIA lists H100 memory at 80GB or 94GB depending on form factor and configurable thermal design power up to 700W. SR030
CR014 ETEnergyWorld projected India’s data-centre operational electricity demand to rise from 1 GW in 2025 to 13 GW by FY32. SR024
CR015 The same ETEnergyWorld analysis estimated that a 13 GW data-centre load could require 30–40 GW of renewable generation plus storage to meet RTC expectations. SR024
CR016 BusinessLine reported that AI-ready training racks can require 80–120 kW and that a 100 kW AI rack can cost roughly ₹6–7 lakh per month in electricity before cooling or floor-space costs. SR025
CR017 KPMG said India’s data-centre buildout is being accelerated by localization, AI workloads, and 5G but remains bottlenecked by execution complexity. SR026
CR018 Neysa’s privacy policy says customer data inside a dedicated tenant remains the customer’s responsibility while Neysa manages infrastructure security under a shared-responsibility model. SR007
CR019 Neysa’s privacy policy says the policy is incorporated into the site Terms of Use and includes a section on cross-border transfers. SR007
CR020 Neysa’s security page claims RBAC, encryption at rest and in transit, BYOK support, audit logs, ISO/IEC 27001:2022 certification, and SOC2 compliance. SR006
CR021 Neysa’s monitoring job posting describes continuous monitoring, incident classification and escalation, root-cause analysis, patch management, and SLA-oriented response. SR014
CR022 CERT-In’s June 2026 guideline explicitly applies to cloud service providers and requires immediate disclosure of critical or high vulnerabilities to affected organizations and CERT-In. SR022
CR023 The DPDP Act defines obligations of data fiduciaries and significant data fiduciaries and includes a dedicated section on processing personal data outside India. SR023
CR024 PIB said IndiaAI was expanding compute capacity from 38,000-plus GPUs by adding 20,000 more GPUs in 2026. SR020
CR025 PIB said IndiaAI had made existing GPU capacity available at ₹65 per hour under the mission. SR020
CR026 Business Standard reported that Neysa’s customer inflow surged after its IndiaAI empanelment was reflected on the portal. SR016
CR027 Neysa and Pipeshift said production inference demand in India is being pushed by concerns about overseas routing, unpredictable latency, and dollar-denominated APIs. SR011
CR028 Neysa’s public named customer proof is still concentrated in a relatively small number of case studies and testimonials compared with its broad market claims. SR001, SR012, SR013
CR029 Neysa’s public proof set is weighted toward regulated or mission-critical workloads such as BFSI, payment operations, research, and public-sector use cases. SR001, SR012, SR013, SR016
CR030 Innoviti’s case study says it serves 50,000-plus merchants across 2,000 cities and processes more than 7,000 service ticket logs daily on Neysa-supported infrastructure. SR013
CR031 IISc’s case study says its Visual Computing Lab generated 33 million sketch-image pairs and trained 7B and 13B models on dedicated compute. SR012
CR032 Innoviti’s case study says general-purpose cloud APIs limited infrastructure visibility and made production economics unattractive. SR013
CR033 Neysa’s October 2024 Series A materials said it had paying customers across AI-native startups, media and entertainment, service providers, software vendors, public sector, and other enterprise sectors. SR009, SR015
CR034 Homepage testimonials claim some customers moved 100% of AI workloads to Neysa because of performance, compliance, cost, and control advantages over hyperscalers. SR001
CR035 The Pipeshift partnership release says early production deployments included a 3x reduction in time-to-first-token for Nurix AI and live multilingual inference for Arrowhead AI. SR011
CR036 Neysa’s monitoring role is on-site in Mumbai and specifically requires hands-on Linux, incident management, and monitoring-tool expertise. SR014
CR037 Neysa’s public credibility still leans heavily on a founder-led infrastructure pedigree built around the former Netmagic leadership team. SR001, SR008, SR015
CR038 Neysa’s Blackstone release says the company intends to secure the debt component subject to documentation, which means financing execution still matters after the headline announcement. SR010
CR039 Economic Times said Blackstone’s ecosystem could help Neysa engage potential clients such as OpenAI and Anthropic, but those are prospecting advantages rather than disclosed contracted workloads. SR018
CR040 Moneycontrol framed Neysa as a domestic alternative to AWS and Azure whose success depends on securing hardware supply chains and consistent utilization. SR019
CR041 BusinessLine says data-centre cash flows are back-ended, power-sensitive, and heavily influenced by utilization, debt, and depreciation. SR025
CR042 ETEnergyWorld warned that inadequate planning for RTC energy can create localized grid stress and rising balancing costs for Indian data-centre expansion. SR024
CR043 KPMG said fragmented providers and regulatory complexity create delays, unclear responsibilities, and capital-access challenges for India’s data-centre buildout. SR026
CR044 Hyperscaler competition is not just about price because AWS, Azure, and Google each market large GPU clusters, mature tooling, and broad AI service ecosystems. SR027, SR028, SR029
CR045 Neysa’s own alternative pages acknowledge that AWS and other general-purpose clouds retain ecosystem depth, managed-services breadth, and global reach. SR031, SR032
CR046 Velocis promises no queue, zero wait, instant deployment, predictable budget, and full compliance, which raises the execution cost of any future reliability miss. SR004, SR005
CR047 Public descriptions of Neysa’s target market consistently emphasize enterprises, government entities, startups, and regulated sectors in India. SR010, SR016, SR017
CR048 IndiaAI’s compute-capacity page shows that multiple providers, not just Neysa, are participating in the mission’s subsidized allocation ecosystem. SR021
CR049 Neysa’s public security page advertises compliance claims and controls, but the page does not itself provide downloadable audit reports, uptime series, or control exceptions. SR006
CR050 Neysa’s pricing pages show three-year commit discounts that are materially steeper than on-demand pricing, making reserved usage capture important to unit economics. SR002, SR003
CV001 Neysa announced a $1.2 billion financing package consisting of up to $600 million of equity and an intended additional $600 million of debt financing. SV001, SV002, SV003
CV002 Independent reporting places Neysa’s transaction at roughly a $1.4 billion enterprise valuation rather than a fully disclosed common-equity mark. SV003, SV005, SV010
CV003 Blackstone is expected to hold a majority stake in Neysa once the announced capital is fully deployed. SV002, SV003, SV005
CV004 Before the Blackstone transaction, Neysa had raised about $50 million externally, comprising a $20 million seed round and a $30 million Series A. SV002, SV010
CV005 At the time of the financing announcement Neysa had roughly 1,200 GPUs live and was targeting deployments of more than 20,000 GPUs. SV001, SV002, SV010
CV006 Management said Neysa aims to more than triple revenue next year, but the public record does not disclose the revenue base from which that growth starts. SV002
CV007 TechCrunch reported that Neysa employed 110 people across Mumbai, Bengaluru, and Chennai at the time of the Blackstone round. SV002
CV008 Blackstone estimated India had fewer than 60,000 GPUs deployed and could exceed 2 million GPUs over time, framing a large local compute buildout opportunity. SV002, SV003
CV009 India’s FY27 budget introduced a tax holiday through 2047 for companies exporting cloud services from Indian data centers. SV003
CV010 Neysa describes itself as sovereign AI infrastructure aligned with the IndiaAI Mission and aimed at enterprises, government entities, hyperscalers, and global AI labs. SV001, SV011
CV011 Sharad Sanghi previously built Netmagic into India’s largest data-center platform before its sale to NTT, providing relevant execution credibility for a new infrastructure buildout. SV005, SV007
CV012 Customer quotations on Neysa’s site cite data residency, latency, domain-tuned models, support responsiveness, and cost transparency as reasons to move AI workloads off generic infrastructure. SV006
CV013 Neysa’s public materials present a full-stack offer that combines GPU infrastructure with orchestration, MLOps, observability, and AI security rather than only bare-metal rental. SV006, SV009, SV010
CV014 The Velocis public price card lists L4, L40S, H100, H200, MI300, B200, and B300 SKUs, indicating a commercialized hardware catalog and roadmap. SV008
CV015 Public materials do not disclose Neysa’s exact ARR, gross margin, utilization, customer concentration, debt covenants, or liquidation preference stack. SV002, SV003, SV005
CV016 Because half of the announced financing is intended debt and the term sheet is undisclosed, the headline $1.4 billion valuation should be treated as a reference mark rather than a proven common-equity price. SV003, SV005, SV010
CV017 CoreWeave reported $2.078 billion of Q1 2026 revenue while CompaniesMarketCap showed a June 2026 market cap of about $55.03 billion. SV012, SV013
CV018 On annualized Q1 revenue, CoreWeave traded at roughly 6.6x run-rate revenue in June 2026. SV012, SV013
CV019 CoreWeave’s $99.4 billion backlog and more than 1 GW of active power show the scale public investors have already rewarded in AI cloud. SV012, SV014
CV020 Fitch said CoreWeave’s top two customers generated about 65% of Q1 2026 revenue and that its rating remained constrained by high leverage, customer concentration, and negative free cash flow. SV015
CV021 Fitch forecast around $33 billion of 2026 capex and expected incremental debt issuance for CoreWeave, underscoring the capital intensity of scaled AI-cloud growth. SV015
CV022 Nebius reported $399 million of Q1 2026 revenue and CompaniesMarketCap showed a June 2026 market cap of about $65.92 billion. SV016, SV017
CV023 On annualized Q1 revenue, Nebius traded at about 41.3x run-rate revenue, making it a hypergrowth outlier rather than a normal benchmark for AI cloud valuation. SV016, SV017
CV024 Nebius also disclosed up to 1.2 GW of power and land for a new Pennsylvania AI factory while warning in its 20-F context about financing, power, supply chain, and vendor dependence. SV016, SV018
CV025 DigitalOcean reported FY2025 revenue of $901 million, year-end ARR of $970 million, and 2026 guidance that implied roughly $1.09 billion of revenue. SV020, SV021
CV026 DigitalOcean’s June 2026 market cap of about $15.5 billion implied roughly 14.2x guided 2026 revenue or about 16x ARR. SV019, SV021
CV027 DigitalOcean said more than 70% of its AI customer ARR came from inference services and core cloud rather than bare metal, suggesting managed workloads can command better multiples than pure GPU rental. SV021
CV028 Oracle generated $18.1 billion of FY2026 OCI revenue, $34.0 billion of total cloud revenue, and $638 billion of remaining performance obligations. SV022, SV023
CV029 Oracle raised $43 billion of debt and $5 billion of equity in FY2026 to fund AI datacenter expansion, even after customer prepayments and customer-supplied GPUs reduced capital needs. SV023
CV030 Lambda’s February 2025 round valued it at $2.5 billion, and management said the platform had well over 25,000 GPUs and more than 5,000 customers. SV024
CV031 Lambda then raised more than $1.5 billion in November 2025 to build gigawatt-scale AI factories, showing that specialist GPU clouds often need repeated large financings before hyperscaler-scale capacity exists. SV024, SV025
CV032 Crusoe’s October 2025 Series E valued the company above $10 billion and highlighted a vertically integrated model spanning energy sourcing, AI-optimized data-center construction, and cloud services. SV026
CV033 Crusoe said bookings grew 5x in the first three quarters of 2025 and that the first phase of its 1.2 GW Abilene campus was live roughly a year after construction began. SV026
CV034 Sacra estimated Together AI had reached about $1 billion of annualized revenue in February 2026 and was in talks to raise about $1 billion at a $7.5 billion pre-money valuation. SV027
CV035 Together AI’s implied multiple of roughly 7.5x annualized revenue sits much closer to CoreWeave than to Nebius, giving a plausible private-market benchmark for revenue-visible AI infrastructure. SV012, SV016, SV027
CV036 Across visible comps, revenue-backed valuation bands run roughly from about 6.6x to about 16x for scaled cloud providers, while Nebius’s roughly 41x reflects a public hypergrowth outlier. SV012, SV013, SV016, SV017, SV019, SV021
CV037 At a $1.4 billion valuation, Neysa would need about $212 million of revenue at 6.6x, $187 million at 7.5x, $140 million at 10x, and $93 million at 15x to sit inside the visible comp band. SV003, SV013, SV019, SV027
CV038 A Nebius-like 41.3x multiple would require only about $34 million of revenue, but that threshold comes from an extreme public outlier with disclosed hypergrowth and public-market liquidity. SV016, SV017
CV039 Because Neysa has not publicly disclosed revenue, public evidence cannot yet show whether it is anywhere near the $34 million, $93 million, or $212 million thresholds implied by the comp set. SV002, SV003, SV015
CV040 The bull case is that India-specific data residency, local support, sponsor-backed procurement, and an integrated software layer make Neysa the default sovereign AI cloud for regulated sectors. SV001, SV006, SV009, SV010
CV041 Neysa’s own competitive material argues that hyperscaler bills can inflate 30-40% above advertised rates once egress, management, storage, and support charges are counted. SV009
CV042 The base case is that Neysa scales meaningfully but remains a niche sovereign provider whose fair value depends on utilization and managed-service mix rather than on raw GPU scarcity alone. SV010, SV021, SV027
CV043 The bear case is that GPU allocation delays, underutilization, or hyperscaler price competition leave Neysa servicing debt against capacity that is not fully monetized. SV003, SV010, SV029
CV044 Data Center Knowledge reported that colocation providers now prioritize investment-grade credit, end-customer visibility, utilization certainty, and balance-sheet durability over aggressive pricing. SV029
CV045 Sacra flags hyperscaler price competition as a risk that could quickly erode Neysa’s cost advantage and force competition on service rather than on economics. SV010
CV046 Forbes quoted Sanghi that hyperscalers are Neysa’s main competitors because they can cross-subsidize GPU capacity inside broader enterprise agreements. SV005
CV047 Neysa’s own comparison page acknowledges that Blackwell is not yet deployed and that thousand-GPU single training runs are still a build-out goal, limiting proof versus global frontier-scale clouds. SV009
CV048 The current recommendation is research-more with medium confidence, high risk, and a stretched valuation stance because the financing is real but revenue visibility and economics are not. SV002, SV003, SV010, SV029
CV049 An upgrade toward track or buy would require private diligence showing revenue or contracted utilization consistent with roughly a 10x-15x revenue band and without punitive debt or preference terms. SV013, SV019, SV027, SV029
CV050 An avoid posture becomes more likely if debt closes on hard terms, customer concentration proves high, or hyperscaler pricing compresses Neysa’s economics before utilization scales. SV010, SV015, SV029
CV051 Public evidence supports sponsor recapitalization or strategic partnership outcomes more clearly than a near-term IPO path, because Neysa lacks the revenue disclosure already visible in public cloud comps. SV005, SV021, SV023
来源
编号出版方标题引文
SO001 Neysa Neysa - AI Acceleration Cloud System Trusted by Leading Startups, Institutions and Enterprises.
SO002 Neysa About Neysa | AI Infrastructure Platform Built for the Future
SO003 Neysa Contact Our Offices in India: Mumbai, Bengaluru, Chennai.
SO004 Neysa Neysa Raises $30 Million in Series A Funding co-led by Nexus Venture Partners, NTTVC and Z47 to Accelerate GenAI Adoption Neysa’s flagship platform, Neysa Velocis, launched in July 2024, which enables on-demand access to high-performance computing infrastructure, is now generally available.
SO005 Neysa Neysa Raises $20 Million in Seed Funding to Accelerate Generative AI Adoption for Enterprises Neysa is co-founded by India’s recognized technology leaders Sharad Sanghi (CEO) and Anindya Das (CTO).
SO006 Neysa Blackstone Leads Funding of Over $1 Billion to Neysa Blackstone and co-investors have provided equity capital of up to $600 million, on the basis of which Neysa intends to secure an additional $600 million of debt financing.
SO007 Neysa Neysa and Data Science Wizards (DSW) partner to launch advanced Insurance AI Cloud platform for Indian insurance sector
SO008 Neysa Neysa and Pipeshift launch real-time inference for open-source AI models, fully deployed within India The platform ... keeps prompts, inference, and enterprise data fully within India.
SO009 Neysa Join Neysa at India AI Impact Summit 2026
SO010 Neysa How IISc Bangalore fine-tuned vision models to read hand-drawn sketches IISc ran every phase of O3SLM’s development on Neysa Velocis, from synthetic data generation through final evaluation.
SO011 Neysa How Innoviti engineered a 60% TCO reduction in field support operations with custom multimodal AI
SO012 Neysa How TIFIN cut GPU cloud spend by 65% while scaling its AI solutions for India's biggest financial institutions
SO013 Neysa ITQ brings transparency, and trust to airline ticket change and cancellation charges with production AI
SO014 Neysa Digital & AI Native Startups
SO015 Neysa Banking & Financial Services
SO016 Neysa Technical Education & Research
SO017 Neysa GPU-as-a-Service Built for AI: 3200 Gbps bandwidth, NVMe-backed storage, fast interconnects.
SO018 Neysa AI Cloud Pricing Starts at $1.17 / hour ... L4; $1.95 / hour ... L40S; $4.39 / hour ... H100 SXM; $4.73 / hour ... H200 SXM.
SO019 Neysa Neysa brings veteran business technology leader Anup Purohit on board as Strategic Advisor
SO020 Z47 Meet the Founders of Neysa Networks | Sharad Sanghi | Anindya Das Sharad & longtime NTT tech leader Anindya Das started Neysa Networks - spotting the opportunity in AI Cloud infrastructure early.
SO021 TechCrunch Blackstone backs Neysa in up to $1.2B financing as India pushes to build domestic AI compute The Mumbai-headquartered startup also plans to raise an additional $600 million in debt financing as it expands GPU capacity.
SO022 SiliconANGLE India’s Neysa raises $1.2B to expand its AI-optimized cloud platform
SO023 Moneycontrol Explained: What Neysa AI does, why it raised $1.2 billion, and why it matters
SO024 Entrackr Gen AI startup Neysa turns unicorn after Blackstone-led $1.2 Bn funding
SO025 VCCircle Blackstone Picks Up Majority Stake in AI Infra Startup Sanghi set up the startup in January 2023, along with former Netmagic executive Anindya Das, who is the chief technology officer of Neysa.
SO026 Fortune India How Sharad Sanghi built Neysa into India’s latest unicorn
SO027 The Economic Times Blackstone leads $600 million raise in AI cloud startup Neysa at $1.4 billion valuation
SO028 Tracxn Neysa - 2026 Company Profile, Team, Funding, Competitors & News Neysa has raised $650M in funding ... with a current valuation of $1.4B.
SO029 Sacra Neysa valuation, funding & news Hyperscaler price competition ... could force the company to compete primarily on features and service quality rather than economics.
SO030 Outsource Accelerator NTT Data, Neysa Networks announce $1.2Bn Hyderabad AI center The upcoming Hyderabad facility will feature a 400MW data center cluster equipped with 25,000 GPUs.
SM001 Prime Minister of India Cabinet approves ambitious IndiaAI mission to strengthen the AI innovation ecosystem
SM002 IndiaAI INDIAai | Pillars
SM003 Press Information Bureau India AI Governance Guidelines: Enabling Safe and Trusted AI Innovation
SM004 Reserve Bank of India Storage of Payment System Data
SM005 Amazon Web Services Global Infrastructure Regions & AZs
SM006 Google Cloud Global Locations - Regions & Zones
SM007 Oracle Where to Find Oracle Public Cloud Regions
SM008 Shakti Cloud Yotta Shakti Cloud: Sovereign Cloud & AI Cloud Platform India
SM009 Neysa GPU Cloud Pricing / Rent H100 On Demand
SM010 Neysa Neysa homepage
SM011 Blackstone Blackstone Leads Funding of Over $1 Billion to Neysa to Work Towards Building India's Leading AI Infrastructure Platform
SM012 TechCrunch Blackstone backs Neysa in up to $1.2B financing as India pushes to build domestic AI compute
SM013 The Economic Times Blackstone leads $600 million raise in AI cloud Neysa at $1.4 billion enterprise value
SM014 CRN Asia Neysa to deploy over 20,000 GPUs in India with Blackstone-led $1.2 billion funding
SM015 ETGovernment India scales AI compute infrastructure to 58,000 GPUs: Transforming global AI dynamics
SM016 EY India The AIdea of India 2026: Sovereign AI in India
SM017 JLL India - The new frontier for AI GPU clusters
SM018 Cushman & Wakefield India's Data Center Market is Emerging as a Key Growth Engine in APAC
SM019 S&P Global Will datacenter growth in India propel country to global hub status?
SM020 Arizton India Data Center Market - Investment Analysis & Growth Opportunities 2026-2031
SM021 CRN Asia India's AI infrastructure build opens new ground for channel partners
SM022 Financial Express Microsoft to invest $17.5 billion in India for AI buildout
SM023 Moneycontrol Explained: What Neysa AI does, why it raised $1.2 billion, and why it matters
SM024 Neysa Blackstone leads funding of over $1 billion to Neysa
SM025 Yotta Labs GPU Cloud for AI Training & Inference | Yotta Labs
SP001 Neysa Top 5 Hyperscaler Alternatives in India for AI and GPU Workloads (2026 Updated)
SP002 Neysa Neysa Velocis-RFP
SP003 Yotta Yotta Shakti Cloud: Sovereign Cloud & AI Cloud Platform India
SP004 Shakti Cloud Yotta Shakti Cloud | Explore on Plans & Pricing
SP005 Yotta Yotta & Markets and Markets on India’s Sovereign AI Shift
SP006 Yotta Yotta Empaneled in India AI Mission to Accelerate AI Adoption with Advanced GPU & AI Cloud Services
SP007 NVIDIA Yotta Built India’s First Sovereign AI Infrastructure With Shakti Cloud
SP008 NVIDIA India Fuels Its AI Mission With NVIDIA
SP009 E2E Networks E2E Networks | India's GPU Cloud for AI & ML
SP010 E2E Networks GPU Cloud India | Rent NVIDIA GPUs from Rs49/hr
SP011 E2E Networks GPU Cloud Pricing | H200/H100 Starting ₹49/hr | E2E Networks
SP012 CNBC TV18 E2E Networks starts work on ₹177 crore IndiaAI Mission order, to go live by January 2026
SP013 IndiaAI IndiaAI Compute Capacity
SP014 Amazon Web Services Global Infrastructure Regions & AZs
SP015 Amazon Web Services Amazon EC2 P5 Instances
SP016 Amazon Web Services AWS Regions in India
SP017 Microsoft Learn ND family virtual machine size series - Azure Virtual Machines
SP018 Microsoft Azure Data Residency in Azure
SP019 Microsoft Learn Azure compliance documentation
SP020 Google Cloud Global Locations - Regions & Zones
SP021 Google Cloud VM instance pricing
SP022 Google Cloud Documentation View and download prices for Google's cloud services
SP023 Tata Communications Tata Communications Secure Enterprise Growth
SP024 Tata Communications Enterprise GPU cloud provider for AI with predictable costs
SP025 Tata Group Move With Vayu
SP026 Data Center Dynamics Tata Communications launches Vayu cloud
SP027 Nasdaq / GlobeNewswire Sify Technologies announces the launch of GPU Cloud Sify CloudInfinit+AI
SP028 CoreWeave The Essential Cloud for AI
SP029 U.S. Securities and Exchange Commission CoreWeave, Inc. Form S-1
SP030 TechCircle India AI Summit 2026 news wrap: GPUs, data centers, sovereign AI
SP031 Fortune India IndiaAI Mission scales up to 34,000+ GPUs, backs Indian AI models for the future
SP032 Tech Funding News NVIDIA fuels IndiaAI with 20K GPUs, sovereign clouds, and startup boost
SP033 Communications Today The IndiaAI GPU procurement makes progress
SI001 Neysa AI Cloud Pricing Save up to 40% when you commit. Both on-demand and reserved pricing options available.
SI002 Neysa $1.2 Billion Capital Raise: Neysa AI Secures Blackstone Funding Blackstone and co-investors have provided equity capital of up to $600 million, on the basis of which Neysa intends to secure an additional $600 million of debt financing, subject to documentation.
SI003 Blackstone Blackstone Leads Funding of Over $1 Billion to Neysa to Work Towards Building India’s Leading AI Infrastructure Platform This funding provides a material impetus to Neysa’s planned scale-up and deployment of over 20,000 GPUs in India.
SI004 TechCrunch Blackstone backs Neysa in up to $1.2B financing as India pushes to build domestic AI compute The startup also plans to raise an additional $600 million in debt financing as it expands GPU capacity.
SI005 The Economic Times Blackstone leads $600 million raise in AI cloud startup Neysa at $1.4 billion enterprise value The fundraise would be split equally between a primary equity round and debt.
SI006 CRN Asia Neysa to deploy over 20,000 GPUs in India with Blackstone-led $1.2 billion funding Blackstone and co-investors have committed up to $600 million in equity capital. Based on this, Neysa intends to secure an additional $600 million of debt financing, subject to documentation.
SI007 Sacra Neysa valuation, funding & news Hyperscaler price competition: Major cloud providers like AWS, Google Cloud, and Microsoft Azure have greater resources and can engage in sustained price competition to defend market share.
SI008 Tracxn Neysa NEYSA NETWORKS PRIVATE LIMITED ... Revenue | ₹10 - ₹50 Cr (As on Mar 31, 2025) | Latest Employee Count | 122 (As on May 01, 2026).
SI009 Moneycontrol Explained: What Neysa AI does, why it raised $1.2 billion and why it matters Its platform, Velocis, operates on a GPU-as-a-Service model, allowing companies to rent high-performance computing resources on demand.
SI010 SiliconANGLE India’s Neysa raises $1.2B to expand its AI-optimized cloud platform The platform is reportedly powered by about 2,000 graphics processing units.
SI011 Neysa Neysa - AI Acceleration Cloud System With Neysa, we got compute tuned for high-throughput, low-latency workloads that actually meet enterprise-grade accuracy requirements.
SI012 Neysa Why Choose Neysa? Transparent pricing, 40–60% lower TCO vs. general-purpose hyperscaler clouds.
SI013 Neysa Neysa Velocis A full-stack AI Acceleration Cloud system built for speed, control, and scale.
SI014 Neysa GPU Cloud for AI 3200 Gbps bandwidth, NVMe-backed storage, fast interconnects ... Up to 40–60% lower unit economics than hyperscalers.
SI015 Neysa AI Infrastructure for BFSI Transparent Cost Controls — Predictable, project-level billing aligned with enterprise financial audit reporting requirements.
SI016 Neysa Neysa raises $30 million in Series A funding Neysa has secured orders from paying customers across various sectors.
SI017 Falcon Ebiz NEYSA NETWORKS PRIVATE LIMITED / U72900MH2022PTC395489 This company is registered at Registrar of Companies(ROC), RoC-Mumbai I with an Authorized Share Capital of ₹5,00,50,000 and paid-up capital is ₹1,95,02,310.
SI018 Tofler Neysa Networks Financials | Company Details It’s authorized share capital is INR 5.00 cr and the total paid-up capital is INR 1.95 cr. ... Revenue ₹ 10-25 cr.
SI019 Neysa How IISc Bangalore fine-tuned vision models to read sketches on Neysa Velocis The pipeline ran continuously and produced 33 million sketch-image pairs before training could begin.
SI020 Neysa Innoviti case study 60% reduction in total cost of ownership.
SI021 Neysa TIFIN case study 65% reduction in GPU cloud spend compared to hyperscalers.
SI022 Neysa ITQ case study 40% Reduction in Total Cost of Ownership vs. General Purpose cloud.
SI023 Neysa Building a happy home for life-saving AI We provided a cost-effective solution that transitioned the company from a restrictive CAPEX model to a predictable OPEX model.
SI024 Neysa AI Unleashed: cutting costs, accelerating innovation, and scaling smart with neocloud 35% say their expensive GPUs sit idle.
SI025 Neysa Neysa empowers India with open-weight sovereign AI control With Velocis, we have built a platform that ends the dependency on closed models that can only be billed on tokens, by giving users full transparency, predictable economics, and the freedom to build on their own terms.
SI026 Neysa Neysa and Pipeshift launch real-time inference for open-source AI models fully deployed within India Nurix AI achieved a 3x reduction in Time to First Token (TTFT) for its voice AI deployments in India.
SI027 U.S. Securities and Exchange Commission CoreWeave, Inc. Form S-1 Our revenue was $16 million, $229 million, and $1.9 billion for the years ended December 31, 2022, 2023, and 2024, respectively. ... our net loss ... was $31 million, $594 million, and $863 million, respectively.
SI028 Compute Forecast Private Credit GPU Infrastructure Risk Is Underexamined H100 GPU cloud rental rates fell 64 to 75% from their peak within 14 months.
SI029 CNBC AI data center boom “stress tests” insurers as private capital floods in There is a core tension in data center project finance: lenders typically want asset lives that exceed loan tenors by a comfortable margin, and the shorter useful life of GPUs challenges that assumption.
SI030 NewsBytes Neysa raised $600 million using GPU-backed debt In India, these GPU-backed loans usually cover 40% to 70% of the GPU value, with interest rates up to 14% on GPU-backed debt.
SE001 Neysa Neysa - AI Acceleration Cloud System
SE002 Neysa Neysa Velocis
SE003 Neysa Neysa Velocis: Platform Architecture & Design
SE004 Neysa AI Infrastructure Security
SE005 Neysa Rent H100 GPUs On-demand Prices from $0.36/hr - Neysa
SE006 Neysa AI Has Advanced. Infrastructure Hasn't.
SE007 Neysa Inference Endpoints Explained: Architecture, Use Cases, and Ecosystem Impact
SE008 Neysa AI Unleashed - Cutting Costs, Accelerating Innovation and Scaling Smart with Neocloud
SE009 Neysa Neysa and Pipeshift launch real-time inference for open-source AI models, fully deployed within India
SE010 Neysa How TIFIN cut GPU cloud spend by 65% while scaling its AI solutions for India's biggest financial institutions
SE011 Neysa How Innoviti engineered a 60% TCO reduction in field support operations with custom multimodal AI
SE012 Neysa How IISc Bangalore fine-tuned vision models to read hand-drawn sketches
SE013 Neysa MLDS 2026
SE014 Neysa Job Openings
SE015 Cloud Security Alliance STAR Registry Listing for | CSA
SE016 Blackstone Blackstone Leads Funding of Over $1 Billion to Neysa to Work Towards Building India’s Leading AI Infrastructure Platform
SE017 TechCrunch Blackstone backs Neysa in up to $1.2B financing as India pushes to build domestic AI infrastructure
SE018 Express Computer Neysa and Pipeshift launch real-time inference for open-source AI models, fully deployed within India - Express Computer
SE019 NVIDIA NVIDIA H100 GPU
SE020 NVIDIA NVIDIA H200 GPU
SE021 Pipeshift Inference Platform: Deploy AI models in Production | Pipeshift
SE022 TIFIN AI for Financial Prosperity | TIFIN
SE023 Innoviti Innoviti | Online payments for modern Indian businesses
SE024 Nurix Nurix AI - Conversational AI for Sales and Support
SE025 Smallest AI Voice AI Platform - TTS, STT & Voice Agents | Smallest AI
SE026 Navana.ai Navana.ai
SE027 Digit Neysa AI shows how Indian businesses are using AI at India AI Summit 2026
SE028 Indian Institute of Science Indian Institute of Science
SU001 Neysa Neysa - AI Acceleration Cloud System
SU002 Neysa AI Cloud Pricing - Neysa Velocis
SU003 Neysa Partner With Neysa – Accelerate AI Growth Together
SU004 Neysa AI Cloud for Banking and Financial Services - Neysa Velocis
SU005 Neysa AI Cloud for Insurance - Neysa Velocis
SU006 Neysa AI Cloud for eCommerce and Retail - Neysa Velocis
SU007 Neysa AI Cloud for Manufacturing - Neysa Velocis
SU008 Neysa AI Cloud for Education and Research Institutes - Neysa Velocis
SU009 Neysa AI Cloud for Digital and AI Native Startups - Neysa Velocis
SU010 Neysa How TIFIN cut GPU cloud spend by 65% while scaling its AI solutions for India's biggest financial institutions Neysa has been a force multiplier. Velocis provides the mission-critical stability and instant scalability of a global hyperscale AI cloud, but with an efficiency that transforms our GPU infrastructure from a cost center into a distinct competitive advantage.
SU011 Neysa How Innoviti engineered a 60% TCO reduction in field support operations
SU012 Neysa AI Inference Case Study for Travel Industry | Neysa & ITQ
SU013 Neysa How IISc Fine-Tuned Vision Models to Read Hand-Drawn Sketches
SU014 Neysa Building a Happy Home for Life-Saving AI
SU015 Neysa Neysa and Pipeshift launch real-time inference for open-source AI models, fully deployed within India
SU016 Neysa The Case of Sovereign AI in India: Local Data, Global Potential
SU017 WEKA Neysa Networks Achieves Speed and Scale with WEKA
SU018 TechCrunch Blackstone backs Neysa in up to $1.2B financing as India pushes to build domestic AI infrastructure Neysa operates in this emerging segment, positioning itself as a provider of customized, GPU-first infrastructure for enterprises, government agencies, and AI developers in India.
SU019 Moneycontrol Explained: What Neysa AI does, why it raised $1.2 billion, and why it matters
SU020 Elets BFSI Neysa partners with GreyLabs AI to deliver enterprise-scale voice insights for BFSI
SU021 The Economic Times Neysa and Pipeshift team up for AI inference play in India
SU022 The Economic Times Tokenomics 2.0: The battle against AI costs
SU023 The Times of India Blackstone leads $1.2 billion raise for homegrown AI firm Neysa
SU024 ClusterMAX by SemiAnalysis Neysa Review 2026: Bronze Tier GPU Cloud Our testing revealed that their current platform has gaps in security and usability when compared to international competitors.
SU025 Innoviti Scaling AI-Powered Field Operations Across 50,000+ Retail Locations Through its collaboration with Neysa, Innoviti built a dedicated AI inference environment that enabled reliable, large-scale deployment of its field operations intelligence platform.
SU026 TIFIN TIFIN announces international expansion around its mission of using AI for wealth with the launch of TIFIN India and strategic investment from DSP Group.
SU027 TIFIN TIFIN Expands AI Operations Creating Global Hub for Financial LLM Innovation and Expanding Access Through Multilingual AI
SU028 TIFIN TIFIN.AI
SU029 Entrackr Gen AI startup Neysa turns unicorn after Blackstone-led $1.2 Bn funding
SU030 Digit Neysa AI shows how Indian businesses are using AI at India AI Summit 2026
SU031 The Economic Times AI to move from pilots to production, see wider adoption in 2026: Neysa’s Sharad Sanghi
SU032 Fortune India Neysa Eyes IPO in 24–36 Months After $1.2 Billion Blackstone-Led Funding, Says CEO Sharad Sanghi
SU033 Visual Computing Lab, IISc Bangalore O3SLM | VCL | IISc
SR001 Neysa Neysa homepage A full-stack AI Acceleration Cloud system built for speed, control, and scale.
SR002 Neysa Pricing Save up to 40% when you commit.
SR003 Neysa Rent H100 on demand Neysa offers three configurations to match your workload and budget.
SR004 Neysa Neysa Velocis Instant sovereign GPU access with full compliance.
SR005 Neysa GPU Cloud for AI Up to 40–60% lower unit economics than hyperscalers.
SR006 Neysa AI Infrastructure Security ISO/IEC 27001:2022 certified and SOC2 compliant*.
SR007 Neysa Privacy Policy v2 Customers are solely responsible for the privacy and security of the data they store or process within their tenant.
SR008 Neysa Seed funding press release Neysa is planning to release its services in Q3 2024.
SR009 Neysa Series A press release Neysa has secured orders from paying customers across various sectors.
SR010 Neysa Blackstone funding press release This funding provides a material impetus to Neysa’s planned scale-up and deployment of over 20,000 GPUs in India.
SR011 Neysa Neysa and Pipeshift launch real-time inference Typical deployment timelines from evaluation to production are under two weeks.
SR012 Neysa IISc Bangalore case study The pipeline ran continuously and produced 33 million sketch-image pairs before training could begin.
SR013 Neysa Innoviti case study Innoviti was processing more than 7,000 service ticket logs every day.
SR014 Neysa Monitoring desk associate (Linux) job posting You will be responsible for continuously monitoring Neysa’s AI platforms and infrastructure for any performance issues, system alerts, or service disruptions.
SR015 Business Standard Sharad Sanghis AI cloud startup Neysa raises $30 mn in Series A funding
SR016 Business Standard Neysa joins IndiaAI Mission, eyes global growth and fresh fundraise At present, we have 95 per cent graphics processing units (GPUs) from Nvidia and some from AMD.
SR017 TechCrunch Blackstone backs Neysa in up to $1.2B financing as India pushes to build domestic AI infrastructure The startup currently has about 1,200 GPUs live and plans to sharply scale that capacity, targeting deployments of more than 20,000 GPUs over time.
SR018 Economic Times Blackstone leads $600 million raise in AI cloud startup Neysa at $1.4 billion enterprise value We require supply chain resilience to ensure we can access GPUs quickly.
SR019 Moneycontrol Explained: What Neysa AI does, why it raised $1.2 billion and why it matters
SR020 Press Information Bureau 20,000 additional GPUs power India’s next phase of AI leadership Under the Mission, the existing 38,000 plus high-end GPUs have been made available at ₹65 per hour.
SR021 IndiaAI IndiaAI Compute Capacity
SR022 CERT-In Guidelines regarding AI-Accelerated Vulnerability Protection and Response Requirements Any Critical or High severity vulnerability affecting deployed cloud services should be communicated to affected organizations and CERT-In immediately upon discovery or confirmation.
SR023 India Code The Digital Personal Data Protection Act, 2023
SR024 ETEnergyWorld Why 13 GW of data centre power load may require 40 GW of renewables India’s data centre operational electricity demand is estimated to grow from 1 GW in 2025 to 13 GW by financial year 2031-2032.
SR025 The Hindu BusinessLine How AWS, Microsoft, Google, Adani and Reliance are driving India’s data-centre boom AI-ready racks can require 80-120 kW for training.
SR026 KPMG India India’s data centre revolution: The integrated lifecycle blueprint 2026-2030 The biggest roadblock is the complexity of meeting the demand.
SR027 AWS Amazon EC2 P5 instances P5, P5e, and P5en instances in EC2 UltraClusters can scale up to 20,000 H100 or H200 GPUs.
SR028 Google Cloud Accelerator-optimized machine family
SR029 Microsoft Azure ND H100 v5 series ND H100 v5-based deployments can scale up to thousands of GPUs with 3.2 Tbps of interconnect bandwidth per VM.
SR030 NVIDIA NVIDIA H100 H100 uses breakthrough innovations based on the NVIDIA Hopper architecture.
SR031 Neysa AWS alternatives in India
SR032 Neysa Hyperscaler alternatives in India AWS and the other general purpose cloud providers retain advantages in ecosystem depth, managed services breadth, and global reach.
SV001 Neysa $1.2 Billion Funding Round: Blackstone Leads Investment in Neysa Blackstone and co-investors have provided equity capital of up to $600 million, on the basis of which Neysa intends to secure an additional $600 million of debt financing.
SV002 TechCrunch Blackstone backs Neysa in up to $1.2B financing as India pushes to build domestic AI infrastructure The Mumbai-headquartered startup also plans to raise an additional $600 million in debt financing as it expands GPU capacity, a sharp increase from the $50 million it had raised previously.
SV003 The Economic Times Blackstone leads $600 million raise in AI cloud startup Neysa at $1.4 billion valuation People familiar with the transaction said Neysa secured an enterprise valuation of $1.4 billion.
SV004 CRN Asia Neysa to deploy over 20,000 GPUs in India with Blackstone-led $1.2 billion funding
SV005 Forbes India Our main competitors are not Indian players, but hyperscalers: Neysa CEO I think the main competitors are not the Indian players, but hyperscalers themselves.
SV006 Neysa Neysa - AI Acceleration Cloud System
SV007 Neysa About Us
SV008 Neysa Neysa Velocis-RFP
SV009 Neysa Top 5 Hyperscaler Alternatives in India for AI and GPU Workloads (2026 Updated) Blackwell is not yet deployed. Its in the pipeline – but most providers in India don’t offer blackwell, anyway.
SV010 Sacra Neysa valuation, funding & news Hyperscaler price competition: Major cloud providers like AWS, Google Cloud, and Microsoft Azure have greater resources and can engage in sustained price competition to defend market share.
SV011 INDIAai INDIAai | Pillars
SV012 CoreWeave via Business Wire CoreWeave Reports Strong First Quarter 2026 Results Revenue backlog1 was $99.4 billion as of March 31, 2026.
SV013 CompaniesMarketCap CoreWeave (CRWV) - Market capitalization
SV014 Last10K 10-Q Quarterly Report Fri May 08 2026
SV015 Fitch Ratings Fitch Affirms CoreWeave's IDR at 'BB-'; Outlook Positive The ratings remain constrained by high leverage, customer concentration and negative FCF during the current investment phase.
SV016 Nebius via Business Wire Nebius reports first quarter 2026 financial results
SV017 CompaniesMarketCap Nebius Group (NBIS) - Market capitalization
SV018 Nebius SEC Filings
SV019 CompaniesMarketCap DigitalOcean (DOCN) - Market capitalization
SV020 DigitalOcean Investor Relations DigitalOcean, LLC - Financials - SEC Filings
SV021 Last10K 10-K Annual Report Tue Feb 24 2026
SV022 CompaniesMarketCap Oracle (ORCL) - Market capitalization
SV023 Oracle via PR Newswire Oracle Announces Record Q4 and FY 2026 Results Driven by Cloud Infrastructure & Cloud Applications
SV024 CNBC / Reuters AI cloud startup Lambda raises $480 million in new round; Nvidia among investors: Reuters
SV025 Lambda Lambda Raises Over $1.5B from TWG Global, USIT to Build Superintelligence Cloud Infrastructure
SV026 Crusoe Crusoe, the AI factory company, raising $1.375 billion at a valuation above $10 billion to power the future of AI infrastructure
SV027 Sacra Together AI revenue, valuation & funding
SV028 TechCrunch The billion-dollar infrastructure deals powering the AI boom
SV029 Data Center Knowledge Neocloud Storm Gathers as Data Center Deals Stall Operators say pricing no longer determines who wins capacity. Providers are prioritizing credit strength, visibility into end-customer demand, confidence in long-term utilization, and balance sheet durability.
SV030 Nebius Nebius Investor Hub
SV031 U.S. News / Reuters Equinix Forecasts Annual Sales Above Estimates on AI Data Center Demand
SV032 Converge Digest Digital Realty Hits Record Bookings as AI Drives Multi-Gigawatt Expansion