Deepgram
产品势头强劲、但私人财务数据不透明的语音 AI 基础设施平台
Deepgram 仍是实时语音 AI 领域可信的品类龙头,但在经审计的财务分母和留存指标披露前,1.3B 美元的 C 轮融资仍只能维持跟踪判断。
覆盖范围与披露说明
本次更新纳入 2026 年 1 月的 C 轮融资和 2026 年 8 月的私募市场经营估算。经审计的 ARR、毛利率、NRR 和客户集中度仍未披露。
封面要素
公司概况
Deepgram 是一家 2015 年成立、注册于特拉华州且总部位于旧金山的语音 AI 基础设施公司。公司面向开发者、AI 平台 ISV 和企业买家,销售 API 优先的语音转文本、文本转语音及语音代理产品;目前披露已有 1,300+ 个组织和 200,000+ 名开发者在其平台上构建应用。公开证据显示其商业势头强劲,但要完整评估这项业务所需的核心财务分母仍未披露。
- 创始人
- Scott Stephenson, Adam Sypniewski
- 总部
- San Francisco, California, United States
- 产品
- 覆盖语音转文本、文本转语音和语音代理的 API 优先语音 AI 平台。
- 客户
- 开发者、AI 平台 ISV 和企业语音自动化买家
- 商业模式
- 按使用量计费的 API 消费,加上企业合同和定制部署。
- 阶段
- growth-stage private
- 融资情况
- 2026 年 1 月以 $1.3B 投后估值完成 $130M C 轮融资
执行摘要
主要优势
- 低延迟语音 AI 平台覆盖语音转文字、文字转语音和语音代理,处于品类领先位置。
- 开发者和企业客户基础庞大且仍在扩张,已有 1,300 多家机构和 200,000 多名开发者。
- 融资支持强劲,战略投资者组合有望为国际扩张提供资金。
主要风险
- 经审计的 ARR、毛利率、NRR 和客户集中度仍未披露。
- 1.3B 美元估值建立在相互冲突的第三方收入估算和有限的非公开披露之上。
- 语音代理、TTS 和基础设施领域的竞争可能挤压定价权和利润率。
未决问题
- 经验证的 ARR、毛利率、NRR 和客户集中度仍是评估 C 轮价格时的主要障碍。
- OfOne 的整合经济性,以及 API、企业合同和餐厅软件之间的收入构成,均无公开信息。
- 从公开证据中看不到股权结构表条款、清算优先权和任何二级交易定价。
目录
01公司概览
1.1 公司身份与技术基础
Deepgram, Inc. 于 2015 年在特拉华州注册,并在加利福尼亚州旧金山设有主要执行总部。公司由 Scott Stephenson 和 Adam Sypniewski 创立;两人最初在中国一座地下暗物质探测器上开展粒子天体物理学研究时,开发了深度学习波形分析技术。他们意识到,商业语音识别仍受传统隐马尔可夫模型和多阶段音系流水线限制,于是构建了端到端深度学习架构,直接把原始音频波形转录成文本。经过 11 年的运营扩张,Deepgram 已从专注自动语音识别的供应商,转型为多模态语音 AI 平台。其生产套件包括基于 Nova-3 的语音转文本、基于 Aura-2 的企业级文本转语音、基于 Flux 的对话轮次检测,以及通过 Voice Agent API 实现的全双工语音到语音编排,服务全球 1,300 多家商业组织和超过 200,000 名注册开发者。[CO001, CO002, CO003, CO004, CO010]
| 指标 | 报告值 / 状态 | 截止日期 | 信心度 | 尽调缺口 / 验证路径 |
|---|---|---|---|---|
| 估值 | $1.3 billion 投后 | 2026-01-13 | 高 | 已与 AVP 领投的 C 轮公告核对 |
| 累计募资 | 股权轮累计 $215M+ | 2026-01-13 | 中 | 私人数据库中的历史累计总额;完整的债务 / 股权拆分尚未确认 |
| 年度经常性收入(ARR) | $100M 估计 / $21.8M(2024) | 2026-08-28 | 低 | 经审计的 GAAP 收入、ARR 定义和收入拆分仍未披露 |
| 平台组织数 | 1,300+ 家企业和 ISV 组织 | 2026-01-13 | 高 | C 轮公告披露;未报告客户层级集中度 |
| 开发者生态 | 200,000+ 名注册开发者 | 2026-01-13 | 高 | 平台指标已披露;活跃开发者与注册开发者的拆分未经验证 |
| 总部与地理布局 | 加利福尼亚州旧金山(特拉华州注册公司) | 2026-08-01 | 高 | 总部位于旧金山 Market St 548 号;业务分布于 20+ 个州和 5+ 个国家 |
主要公司和财务指标汇编自 C 轮披露及私人市场数据库;ARR 和经验证的利润率仍未确认。
[CO001, CO006, CO007, CO010, CO015]1.2 管理层与关键人物结构
Deepgram 由原有技术联合创始人和经验丰富的运营高管共同领导。联合创始人兼首席执行官 Scott Stephenson 拥有密歇根大学物理学博士学位,是公司的主要战略发声者;他主张语音是人机交互的终极接口,同时负责公司资本运作和伙伴拓展。联合创始人兼首席技术官 Adam Sypniewski 负责研发、深度学习架构,以及不断扩大的美国专利组合,专利覆盖统一 Transformer 语音识别和硬件高效并行处理。首席运营官 Shadi Baqleh 负责运营扩张与行政执行,带领商业企业运营团队;组织采用远程优先模式,覆盖 20 多个州和 5 个国家。从治理和风险角度看,Deepgram 保持制度化控制,并执行标准服务条款:争议必须在旧金山进行具有约束力的个人仲裁,且排除集体诉讼。[CO002, CO005, CO013]
| 人员 | 职务 | 背景 | 职能覆盖与市场匹配 | 关键人物依赖 |
|---|---|---|---|---|
| Scott Stephenson | 联合创始人兼首席执行官 | 密歇根大学物理学博士;曾在中国开展地下暗物质探测器波形分析研究。 | 负责深度学习语音模型的高管领导、战略融资和产品愿景。 | 高:主要外部布道者和技术联合创始人,掌舵长期战略。 |
| Adam Sypniewski | 联合创始人兼首席技术官 | 密歇根大学物理学博士;从事暗物质波形和计算物理研究。 | 负责核心语音 AI 架构、专有神经网络训练和专利组合管理。 | 高:技术愿景负责人,推动端到端深度学习模型突破。 |
| Shadi Baqleh | 首席运营官 | 资深运营高管,具备软件即服务运营规模化经验。 | 负责全球商业运营、法律 / 合规基础设施和组织规模化。 | 中:对日常执行至关重要,但运营职责可以交接。 |
管理层背景和职能重点汇编自官方公司披露及高管资料。
[CO001, CO002, CO005]1.3 资本化历程与投资人阵容
自完成 Y Combinator 2016 年冬季批次加速器项目以来,Deepgram 已在 7 轮融资中累计募集超过 $215M 的股权资本。2026 年 1 月,公司以 $1.3B 投后估值完成 $130M C 轮融资,跻身独角兽。领投方 AVP 是一家独立的欧美成长型投资平台,管理资产超过 2.5B 欧元;公司选择 AVP,旨在加速国际市场进入。C 轮融资显示现有投资人信心很强,Madrona Venture Group、Alkeon Capital、Tiger Global、Wing 和 BlackRock 管理的基金继续参与。更重要的是,本轮还引入 Twilio、ServiceNow Ventures、SAP 和 Citi Ventures 等知名企业战略投资人;它们既是资本伙伴,也把 Deepgram 的实时语音 API 嵌入大型企业软件生态,成为商业分销渠道。[CO006, CO007, CO008]
| 利益相关方 | 角色 / 轮次 | 控制权或经济重要性 | 尽调要求 |
|---|---|---|---|
| AVP | 领投方,C 轮($130M,2026 年 1 月) | 独立科技成长平台,管理资产规模超过 €2.5B;领投 C 轮并推动全球 / 欧洲扩张。 | 核实董事席位安排、投票权及对未来融资的否决权控制。 |
| Madrona Venture Group | 领投方,B 轮扩展轮($47M,2022 年 11 月);参与 C 轮 | 长期风险投资支持者,聚焦太平洋西北地区和企业 AI;参与多轮融资。 | 审查按比例跟投权和董事代表任期。 |
| Alkeon Capital Management | 联合领投方,B 轮;参与 C 轮 | 成长股权和交叉投资管理机构,为规模化阶段提供资本。 | 检查二级股份购买及信息权协议。 |
| Y Combinator | 基础种子加速器(W16 批次) | 早期机构催化者,也是最初企业客户引荐的网络来源。 | 确认初始种子股权契约及持续顾问身份。 |
| Strategic Corporate Syndicate (Twilio, ServiceNow, SAP, Citi) | 战略企业投资者,C 轮(2026 年 1 月) | 将 Deepgram API 接入企业工作流的商业验证方和渠道分销伙伴。 | 评估商业最低承诺、API 排他性和 IP 授权条款。 |
| 机构交叉策略基金(Tiger Global、BlackRock) | 主要机构投资者,C 轮(2026 年 1 月) | 以大规模机构资产管理能力支撑独角兽估值和资产负债表储备。 | 审查清算优先权、级次分配瀑布和登记权。 |
机构和企业风投支持者汇编自公开融资公告及监管银团通知;具体股权比例和清算优先权未公开披露。
[CO006, CO007, CO008]1.4 商业规模、单位经济与运营里程碑
Deepgram 通过按使用量计费的开发者 API 和企业量级合同,将基础语音模型变现。Nova-3 的主打转录价格较为激进:流式音频起价 $0.0077 / 分钟,批量音频为 $0.0043 / 分钟。但第三方定价尽调显示,生产环境必需的附加功能——说话人分离($0.002/min)和 PII 脱敏($0.002/min)——各自会使基础成本增加 26%,将全功能企业流水线成本推高至 $0.0130 / 分钟,比原始转录贵 69%。商业上,截至 2026 年 8 月,该平台的年度经常性收入估计已达到 1 亿美元,此前 2024 年报告收入为 2,180 万美元。2026 年 1 月收购 Y Combinator 支持的 OfOne 后,平台战略边界进一步扩大:Deepgram for Restaurants 可自动化快餐店得来速点单,订单自助完成率超过 95%,并由符合 HIPAA、SOC 2 和 GDPR 的零留存企业基础设施提供支持。[CO009, CO010, CO011, CO012, CO014, CO015]
Deepgram 创始研究、风险融资轮次、基础模型发布和公司收购的时间脉络。
里程碑日期反映公开新闻公告和平台发布通知。
[CO001, CO002, CO006, CO007, CO008, CO009, CO010, CO014]1.5 展示项
02市场分析
2.1 市场边界、品类定义与纳入支出
语音人工智能市场包含多个技术层,覆盖语音识别、语音合成和实时代理编排。Deepgram 将核心市场边界定义为实时开发者 API:提供自动语音转文本(STT)转录和文本转语音(TTS)生成,而不是面向消费者的打包应用或传统人工转录服务。纳入支出包括 API 消费额度、自定义语言模型训练、多语言语音识别和流式语音代理流水线基础设施。相反,通用云计算托管、通用大语言模型 token 生成、传统硬件电话 PBX 系统和人工医疗转录劳动均不计入。平台把转录和合成统一到单一对话式 API 架构中,面向希望消除多供应商语音流水线集成摩擦的软件构建者和企业自动化团队。[CM001, CM002, CM009]
| 类别 | 纳入支出 | 排除支出 | 买方 / 付款方 | 相关性 |
|---|---|---|---|---|
| 语音转文字(STT)API | 流式 API 使用、批量音频处理、自定义词汇模型、说话人分离功能 | 人工转录服务、终端用户会议录音消费软件、硬件录音设备 | 工程副总裁、产品负责人、开发者平台负责人 | 核心收入引擎,为语音代理和联络中心工作流提供低延迟转录 |
| 文字转语音(TTS)合成 | 实时语音生成 API、富有表现力的对话式语音模型、低延迟音频流 | 录音棚配音制作机构、广播电台设备、有声书分发平台 | 产品经理、对话式 AI 设计师、客户体验负责人 | 高增长的相邻技术栈,承接语音人格支出并实现自然的轮流对话 |
| 语音代理编排 | 单次调用语音到语音 API、轮流对话上下文引擎、LLM 语音集成管线 | 通用 LLM 推理 tokens、云托管计算实例、传统 PBX 电话硬件 | CTO、AI 自动化负责人、联络中心现代化负责人 | 高价值平台层,将 STT、LLM 编排和 TTS 统一为自动化对话代理 |
| 垂直领域语音自动化 | 餐厅免下车语音自动化、专业医疗转录管线、领域词汇 | 销售点硬件终端、厨房显示系统、临床电子健康记录软件 | 运营高级副总裁、首席信息官、医疗 IT 采购负责人 | 通过自动化劳动密集型语音交互,以垂直领域解决方案获得溢价 |
该分类将基础设施软件 API 支出与实体电话硬件、人工转录及通用云计算区分开来。
[CM001, CM002, CM009, CM013]2.2 多重分析视角下的可触达市场测算
测算可触达机会时,需要分别评估不同市场视角,不能只依赖一个宽泛预测。独立行业研究估计,全球语音转文本 API 市场单独到 2030 年就可能达到 $8.3 billion,复合年增长率为 14.1%。与此同时,文本转语音合成市场到 2029 年还代表着 $10.0 billion 的机会,年增长率为 19.1%。除单点组件 API 外,Deepgram 还覆盖更广泛的联络中心和客户服务自动化市场;随着企业用自主语音代理替换传统交互式语音应答树,该市场全球年支出超过 $40 billion。在高合规专业垂直领域,仅医疗语音转录市场到 2027 年就代表 $4.89 billion 的可触达细分市场。以 2026 年 8 月估计的 $100 million ARR 运行率里程碑衡量,Deepgram 在可服务可触达市场中仍有很大空间。[CM003, CM004, CM005, CM007, CM011, CM014]
| 发布方 | 规模视角 | 地理范围 | 数值 | CAGR | 方法 | 信心度 | 局限 |
|---|---|---|---|---|---|---|---|
| Sacra / 行业预测 | 全球语音转文字 API 市场 | 全球 | 2030 年 $8.3B | 14.1% | API 消费建模与企业开发者工作负载汇总 | 中 | 将批量媒体转录与实时语音代理需求合并计算 |
| Sacra / 行业预测 | 全球文字转语音市场 | 全球 | 2029 年 $10.0B | 19.1% | 数字助理、媒体和 CX 渠道中的合成语音采用情况 | 中 | 包含企业 B2B API 之外的消费媒体旁白和游戏支出 |
| Sacra / 行业基准 | 联络中心对话式 AI 支出 | 全球 | $40.0B+ | 20.0%+ | 企业联络中心运营软件和劳动力替代支出 | 中 | 顶层 TAM 包含联络中心坐席薪资和完整 CCaaS 套件软件 |
| Sacra / 分析师估计 | 医疗语音与转录垂直市场 | 全球 | 2027 年 $4.89B | 12.5% | 医疗服务提供商文档支出和临床转录合同 | 中 | 不含医院 EHR 平台许可和专业临床硬件 |
市场预测汇编自 Sacra 研究和公开分析师估计;数值反映软件毛 TAM,而非 API 平台净收入。
[CM003, CM004, CM005, CM007]从整体联络中心自动化,到实时语音代理 SAM 和当前运行率 SOM 的可服务市场分层。
各层宽度仅表示层级,不代表数值比例。
市场分层综合 Sacra 市场预测、分析师预测和 Deepgram 披露的客户增长基准。
[CM003, CM004, CM005, CM011, CM014]2.3 买家分层、预算归属与采购路径
企业和开发者采用分为三个客户群,采购逻辑各不相同。开发者和 AI 原生层包括构建对话代理、自动会议摘要和客服机器人的产品工程师,通常通过按使用量计费的自助 Pay As You Go 方案接入。中型及成长型软件公司会升级到年度 Growth 合同,起价为每年 $4,000,资金来自部门工程预算。企业层买家包括联络中心高管、首席信息官和运营负责人,他们要走正式供应商采购流程。企业买家重视定制词汇准确率、多租户安全和专用吞吐量保障。收购餐厅得来速平台 OfOne 把公司推向垂直企业工作流,也把买家画像延伸到希望直接降低劳动力成本的零售和快餐运营高管。[CM006, CM010, CM013]
| 客群 | 买方 | 用户 | 付款方 | 工作流 | 预算负责人 | 采用触发因素 |
|---|---|---|---|---|---|---|
| AI 原生应用构建者 | 创始人 / 工程负责人 | 软件工程师与 AI 开发者 | 企业信用卡 / 自助服务 | 实时对话代理、交互式数字人、实时语音助理 | 工程 / 研发预算 | 传统超大规模云厂商语音 API 出现延迟峰值或准确率下降 |
| 企业联络中心与 CX | 客户体验副总裁 / CIO | 联络中心工程师、支持运营 | 企业集中采购 | 实时坐席辅助、自动化客户来电解决、通话后合规分析 | 客户支持与运营预算 | 呼叫中心劳动力成本上涨、客户满意度和等待时间恶化 |
| 受监管医疗与金融 | 首席信息安全官 / IT 副总裁 | 临床人员、合规官 | 企业 IT 指导委员会 | 医疗就诊转录、财务顾问合规记录、审计轨迹 | 企业 IT 与风险 / 合规预算 | HIPAA/SOC 2 合规及本地数据隔离的监管要求 |
| 快餐与零售自动化 | 运营高级副总裁 / 首席数字官 | 门店经理、免下车员工 | 零售技术运营 | 免下车自动接单、语音自助终端交互、订单路由 | 门店技术 / 运营资本开支 | 零售工资压力剧增、免下车订单错误率高、免下车吞吐量受阻 |
客群图谱体现 Deepgram 覆盖自助开发者、企业 IT 和专业垂直软件的多层次市场进入路径。
[CM002, CM006, CM007, CM013]2.4 采用驱动因素、监管瓶颈与尽调风险
市场扩张由企业对类人语音代理的需求推动;这类代理需要亚秒级延迟和自然的对话轮转,而传统大型云厂商的语音引擎往往难以做到。但企业合规、安全治理和切换壁垒,正显著拖慢采用速度。医疗和金融服务等强监管行业面临严格的数据隐私门槛;医疗数据泄露平均每起造成 $10.93 million 损失,因此企业在把敏感音频送入云端之前,要求严格的 HIPAA、SOC 2 和 GDPR 保障。旧式电话系统集成和供应商锁定也会增加运营摩擦、拉长销售周期。投资尽调的关键缺口仍是:云 API 量级业务与私有或本地部署之间,缺少披露的单位经济和经审计毛利率。[CM002, CM007, CM008, CM012]
2.5 展示项
03竞争格局
3.1 竞争版图与结构分类
2026 年商业语音人工智能市场由四类结构鲜明的竞争者构成:独立专业模型实验室、大型云平台供应商、开放权重自托管替代方案,以及前沿多模态基础模型实验室。Deepgram、AssemblyAI、Speechmatics 和 ElevenLabs 等独立模型实验室专注声学建模和语音处理,主要比拼低延迟、专业领域准确率和开发者 API。Google Cloud Speech-to-Text、Microsoft Azure Speech Services 和 Amazon Transcribe 等大型云供应商,则把语音当作基础设施商品,打包进完整企业云协议。OpenAI Whisper Large v3 等开放权重模型,为愿意承担 GPU 基础设施和推理工程成本的组织提供零许可费替代方案。前沿基础模型实验室通过发布原生语音到语音的多模态模型,正形成新的替代层,把传统语音流水线压缩进单一神经网络。[CP001, CP002, CP004, CP013, CP014]
从语音 AI 技术栈集成广度和部署自主性两个维度,与商品化云服务和开源模型进行战略定位比较。
| 条目 | X (5–8.5) | Y (4–8.5) | 说明 |
|---|---|---|---|
| 1. Deepgram | 8.5 | 8.5 | 全栈语音 AI(Nova-3 STT + Aura-2/Flux TTS + Agent API),支持本地部署 / VPC 运行时。 |
| 2. AssemblyAI | 6 | 4.5 | 面向开发者的 STT 专业厂商,配备 LeMUR 音频智能;仅支持云端,不提供第一方 TTS。 |
| 3. Speechmatics | 5 | 7.5 | 擅长口音和多语言 STT 的领导者,为欧洲企业提供强大的本地部署选项。 |
| 4. ElevenLabs | 7.5 | 4 | 表现力强的 TTS 和对话式 AI 领导者;云端优先,同时具备次要的语音转文本能力。 |
| 5. OpenAI(Whisper / Realtime) | 7 | 6.5 | 开放权重 Whisper 提供最高的部署自主性;Realtime API 提供集成式多模态云端体验。 |
| 6. 超大规模云服务商(GCP / Azure / AWS) | 6.5 | 5 | 覆盖多种服务的广泛云产品目录,将商品化语音 API 打包进更广泛的企业云协议。 |
坐标分数是有证据支持的序数坐标(1 至 10),评估架构广度(第一方 STT、TTS 和代理运行时)与部署主权(本地部署、私有云和自托管经济性)。
[CP001, CP002, CP006, CP008, CP013, CP014]3.2 同业画像与架构差异化
在专业独立供应商中,Deepgram 的架构位置很独特:它同时运营第一方自动语音识别(Nova-3)和第一方语音合成(Aura-2 与 Flux TTS)。直接同业过去通常只专注一种模态:AssemblyAI 严格聚焦语音转文本,使用 Universal-3 模型和 LeMUR LLM 编排层,并依赖第三方伙伴生成语音;Speechmatics 通过 Ursa 模型专注多语言转录,在带口音英语和方言稳健性上保持技术领先;ElevenLabs 则从表现力语音克隆切入,再扩展到对话代理。Deepgram 借助统一声学栈快速扩张,到 2026 年中 ARR 运行率估计达到 00M,并由 30M C 轮融资支持。这一双模态基础使其转录延迟低于 300 毫秒、合成延迟低于 90 毫秒,足以支撑交互式企业语音应用。[CP001, CP002, CP005, CP006, CP008, CP012]
| 竞争对手 | 类别 | 规模与融资 | 目标客群 | 核心差异化 | 战略局限 |
|---|---|---|---|---|---|
| Deepgram | 专业模型实验室 | 00M ARR(Sacra 估计),累计融资 15M(C 轮 30M) | 开发者优先、企业联络中心、语音代理 | 统一的一方 STT/TTS 技术栈、低于 300ms 的延迟、可自托管 | 依赖厂商选择的基准,学术研究产出有限 |
| AssemblyAI | 专业模型实验室 | 0M+ ARR(估计),募资 15M | 开发者、异步媒体转录、AI 摘要 | LeMUR 音频智能层、开发者文档、LLM 工作流 | 没有一方 TTS 引擎;依赖第三方语音合成伙伴 |
| Speechmatics | 专业模型实验室 | 0M+ ARR(估计),募资 2M+ | 欧洲企业、合规、广播、国防 | Ursa 引擎的口音覆盖、方言处理、本地部署 | 仅聚焦转录,没有对话式 TTS 或语音代理 API |
| OpenAI (Whisper / Realtime) | 前沿 AI 实验室 | 数十亿美元 ARR,资本超过 0B | 通用开发者、消费 AI、对话式界面 | 开放权重 Whisper 分发、GPT-4o 原生多模态语音 API | 自托管 Whisper 缺少托管式流运行时;Realtime API 是昂贵的黑盒 |
| Google Cloud (Speech-to-Text) | 超大规模云 | 数十亿美元云收入基础 | 全球企业、现有 GCP 客户、呼叫中心 | 全球语言覆盖、Chirp 模型、捆绑的 Dialogflow CX 工具 | 每分钟定价更高,开发者迭代更慢,支持不够专业化 |
规模和 ARR 估计源自 Sacra 2026 年情报及行业报道;融资反映截至 2026 年累计风险投资轮次。局限体现架构取舍。
[CP001, CP002, CP004, CP005, CP008]3.3 能力矩阵与技术取舍
对企业语音供应商的评估显示,延迟、部署灵活性和对话轮转动态之间存在明显技术取舍。在高吞吐联络中心场景中,Deepgram 的话轮结束转录延迟低于 300 毫秒,优于通常超过 600 毫秒的大型云厂商通用 API。Deepgram 的 Flux 引擎把对话轮转直接融入声学处理,使语音代理无需外部静音计时器也能处理人类打断。Deepgram 还通过 Enterprise Runtime 覆盖受监管行业,支持部署到客户虚拟私有云或本地环境,并提供 SOC 2 Type II 和 HIPAA 合规。但独立尽调也提示关键限制:供应商公布的基准测试显示词错误率低 40% 至 54%,却高度依赖供应商自行选择的评测语料。在非标准声学环境的真实企业部署中,独立评测显示其相对 Speechmatics 和优化后的 Whisper 流水线的性能差距更窄。[CP002, CP003, CP006, CP008, CP009, CP011, CP012]
| 采购标准 | Deepgram | AssemblyAI | Speechmatics | OpenAI Whisper / Realtime | 超大规模云(GCP/Azure/AWS) |
|---|---|---|---|---|---|
| 一方 STT 模型 | Nova-3(专有端到端神经网络) | Universal-3(专有神经网络) | Ursa(专有自监督) | Whisper Large v3(开放权重) | Chirp / Neural Speech(专有云) |
| 一方 TTS 合成 | Aura-2 与 Flux TTS(<90ms 延迟) | 无(需要伙伴集成) | 无(未提供) | TTS-1 & Realtime API | 云 TTS / 神经语音 |
| 语音代理编排 | 集成式 Voice Agent API(.50/hr) | LeMUR + 外部框架 | 无(开发者自行搭建) | OpenAI Realtime API(统一) | Dialogflow CX / Amazon Lex / Azure Bot |
| 部署灵活性 | 云 API、VPC 和本地运行时 | 仅云 API | 云 API、VPC、本地部署 | 自托管权重或云 API | 带选定混合边缘的云 API |
| 轮流对话 / 打断逻辑 | Flux 对话式轮流引擎 | 需要外部编排 | 需要外部编排 | 原生服务端 VAD(Realtime API) | 可配置的语音静默计时器 |
能力评估源自官方文档、AI Wiki 竞争分析和截至 2026 年的产品发布规格。
[CP002, CP003, CP006, CP009, CP012]3.4 定价、产品打包与护城河持久性
Deepgram 采用激进的量导向定价,在转录和对话式语音两个层级都形成竞争压力。Pay-as-you-go 方案中,批量音频价格为 /bin/bash.0043 / 分钟,流式音频为 /bin/bash.0077 / 分钟,明显低于 Google Cloud 和 Microsoft Azure 每分钟 /bin/bash.0160 至 /bin/bash.0240 的目录价。其统一 Voice Agent API 的价格为每对话小时 .50,显著低于 OpenAI Realtime API 音频令牌每小时 8 至 4 的成本,以及拼接独立 STT、LLM 和 TTS 端点所需的 至 2 成本。企业语音 AI 的切换成本主要来自定制声学微调、词汇适配和生产遥测流水线,而不是 API 协议复杂度。不过,Deepgram 的长期护城河仍有风险:Whisper 等开放权重模型会持续压低商品化转录价格上限,前沿多模态语音模型最终也可能完全绕过中间转录层。[CP003, CP004, CP007, CP010, CP014]
| 定价维度 | Deepgram | AssemblyAI | Speechmatics | OpenAI Whisper / Realtime | 超大规模云可比方(GCP/AWS) |
|---|---|---|---|---|---|
| 预录制 STT(每分钟) | /bin/bash.0043/min(按量付费 Nova-3) | /bin/bash.0065/min(标准异步) | /bin/bash.0070–/bin/bash.0100/min(Ursa 批处理) | 免费开放权重(托管 /bin/bash.0060/min) | /bin/bash.0160–/bin/bash.0240/min 标价 |
| 流式 STT(每分钟) | /bin/bash.0077/min(实时 Nova-3) | /bin/bash.0090/min(实时流) | /bin/bash.0100–/bin/bash.0140/min(实时) | /bin/bash.0600/min(Realtime 音频输入) | /bin/bash.0160–/bin/bash.0240/min 标价 |
| 语音代理 / Realtime API | .50/小时对话运行时 | 组件聚合成本(–/小时) | N/A(仅转录) | 8.00–4.00/小时,音频 tokens 估计 | 拼接式 STT + LLM + TTS(–5/小时) |
| 企业承诺层级 | 成长(k/年)、企业(5k/年+) | 定制企业合同 | 年度企业许可 | 按用量计费档位或 Azure 企业方案 | 云承诺折扣协议 |
费率反映截至 2025–2026 年公布的按量付费标价。语音代理的小时费率在打包销售时包含 STT、模型编排和 TTS 交付。大客户批量折扣另行协商。
[CP003, CP007, CP010]3.5 展示项
04财务
4.1 收入来源与定价架构
Deepgram 以多产品 API 变现模式服务开发者、技术 ISV 和企业联络中心。公司的核心商业引擎长期围绕自动语音转文本(STT)转录,既提供预录音频批处理,也提供低延迟实时流式处理。Nova-3 批量音频转录基础目录价为每分钟 $0.0043,实时流式则收取每分钟 $0.0077 的溢价。客户通过消费额度或合同承诺把音频量变现。2025 年末,Deepgram 推出 Voice Agent API,扩大了变现边界:这一统一语音到语音复合服务按每对话小时 $4.50 计费,在单一端点中编排转录、语言模型推理和语音生成。Aura-2、Flux TTS 等补充 TTS 模型还按字符和流式生成收取增量费用;摘要、实体识别和情感分析等 Audio Intelligence 附加功能,则为高通话量呼叫中心带来交叉销售机会。[CI003, CI004, CI005, CI006]
| 收入来源 | 变现机制 | 单位 / 指标 | 当前进展 / 状态 | 收入质量 | 尽调需核实 |
|---|---|---|---|---|---|
| 语音转文字(STT)API | 按消耗计费的 API,含分档折扣 | $0.0043/min(批处理)至 $0.0077/min(流式 Nova-3 标价) | 核心收入来源;已处理 50,000+ 音频年 | 使用频次高且经常性强;开源方案带来商品化定价压力 | 前 20 大客户的实际每分钟有效费率 |
| 语音代理 API | 语音到语音的复合 API 计量 | 每小时对话 $4.50 | 2025 年末上线;正拓展至联络中心自动化 | 高价值复合 SKU;每次交互带来的毛收入更高 | 每个对话小时的计算毛利拆分 |
| 文字转语音(TTS)API | 按消耗计量的音频生成 | 按字符 / 按流计费的消耗档位 | Flux TTS 和 Aura-2 已上线;与 STT 互补 | 随对话代理快速增长;竞争激烈 | 独立 TTS 用量与打包代理部署的对比 |
| 企业许可与私有 VPC | 年度企业软件及专用基础设施合同 | 年度许可费(基础价 $15,000+)+ 定制模型训练 | 服务受监管的银行及医疗客户(SOC 2、HIPAA) | 留存率高且有合同锁定;销售周期更长 | 合同期限、续约条款及自托管支持毛利 |
标价反映公开文档中的标准 Pay As You Go 价格;实际客户费率还会计入 Growth(最高 20%)及 Enterprise 定制批量折扣。
[CI003, CI004, CI005, CI006]4.2 市场进入动作与变现机制
Deepgram 把产品驱动的自助获客漏斗,与企业直销和联合销售结合起来。开发者入门可获得 $200 免费 API 额度,工程团队先基准测试准确率和延迟,再进入商业计费。初始额度用完后,账户转入按消费计费的 Pay As You Go 方案;预付余额低于可配置阈值时,系统会通过信用卡自动充值。对于增长型应用,Deepgram 提供年度 Growth 方案,要求至少预付 $4,000 年度额度,换取最高 20% 的阶梯折扣。年度承诺 $15,000 或以上的企业账户,可以协商定制量价、专用 VPC 或本地部署许可、自定义模型微调和企业 SLA。合同严格规定 API 使用额度取消后不退款,使年度方案为 Deepgram 提供前置营运资金;但 Sierra、Decagon 和 Twilio 等大型平台伙伴的实际每分钟净价,仍远低于公开目录价。[CI004, CI005, CI006, CI007, CI011]
| 产品 / 档位 | 标价 | 合同 / 最低承诺结构 | 标价与实际价格 | 折扣与未知项 | 来源与核验 |
|---|---|---|---|---|---|
| Pay As You Go | $0.0043-$0.0077/min STT;$4.50/hr 语音代理 | 无最低承诺;自动充值余额 | 标价直接从信用卡扣款 | 无批量折扣;按条款充值余额不可退款 | Deepgram 服务条款与官方文档 |
| Growth 档位 | 按用量计费,需预付承诺 | 年度预付额度承诺 $4,000+ | 年度预付现金支付后按折扣标价计费 | 标准 API 费率最高 20% 折扣 | Sacra 研究与 Deepgram 服务条款 |
| Enterprise 档位 | 定制协商的批量价格 | 年度合同最低 $15,000+ | 实际费率较标价大幅折扣 | 批量折扣、定制模型训练、专用 SLA | Sacra 研究与 Enterprise 解决方案备案 |
| 语音代理复合服务 | 每个对话小时 $4.50 | 按用量计费,或打包进企业支持协议 | 综合标价涵盖 STT、编排和 TTS | BYO-LLM 选项会改变实际净费率 | Deepgram 上线公告与分析师报道 |
Deepgram 服务条款规定,取消后预付余额和订阅费均不可退款;实际企业折扣需审阅客户合同。
[CI004, CI005, CI006, CI007]4.3 收入转化与运营流程
要把企业对话流量转成持久毛利,需要同时平衡推理效率、网络基础设施和分层变现。Deepgram 客户组织在联络中心、销售智能应用和客服机器人中接入音频,把实时语音数据流入 Deepgram 的专用深度学习运行时。系统按转录分钟数或双工对话小时数计量原始音频,并套用分层价格表。语音 AI 工作负载要求实时执行、延迟低于 300 milliseconds,因此服务成本主要由 GPU 基础设施和数据中心吞吐量驱动。Deepgram 采用专有端到端深度神经架构,而不是级联式多供应商流水线,据称相对 Google Cloud 和 Microsoft Azure 等通用云服务商拥有 2-5x 的性价比优势。这种结构效率让高量企业流量摊薄核心模型训练成本,并在多租户云环境中维持软件式毛利率。[CI003, CI004, CI008, CI011, CI012]
展示原始企业对话量如何经 API 编排转化为毛利的结构化流程。
| 节点 / 连接 | 节点 / 起点 | 终点 | 说明 |
|---|---|---|---|
| 节点 1 | 企业语音流量接入 | 从 1,300 多家客户组织和开发者处接入超过 50,000 个音频年、1 万亿个词。 | |
| 节点 2 | 统一 API 处理 | 通过端到端深度学习模型执行,包括 Nova-3 STT、Flux TTS 和 Voice Agent 运行时。 | |
| 节点 3 | 多层计量架构 | 转录按分钟计费($0.0043-$0.0077/分钟),对话代理按小时计费($4.50/小时)。 | |
| 节点 4 | 合同变现层 | 收入来自按用量自动充值、Growth 预付款($4K+/年)和 Enterprise 合同($15K+/年)。 | |
| 节点 5 | 基础设施与模型服务成本 | GPU 计算、数据中心托管、模型推理延迟优化和合规协议。 | |
| 节点 6 | 毛利率与现金实现 | 预计毛利率超过 85%;随着规模扩大,高音频量摊薄模型训练投入后,这部分毛利得以实现。 |
节点顺序反映运营变现机制;利润率数据是行业估算,不是经审计的 GAAP 披露。
[CI003, CI004, CI008, CI012]4.4 成本结构与单位经济分析
Deepgram 的销货成本(COGS)包括云托管、模型推理所需的专用 GPU 计算实例、数据中心网络和合规开支。医疗和银行等受监管行业的企业部署必须严格遵守 HIPAA、SOC 2 和 GDPR,因此需要专用基础设施分区和数据隔离,增加交付运营成本。尽管推理需要大量资本投入,Sacra 的独立研究仍估计 Deepgram 毛利率维持在 85% 或更高,反映专有模型优化带来的强单位经济。Sacra 还估计 Deepgram 于 2026 年 8 月达到 $100 million ARR,并在 2025 年以 115% 的同比 ARR 增长进入新阶段,原因是企业转向实时语音代理。但这些指标没有经审计 GAAP 财务文件佐证;净收入留存率(NRR)、客户获取成本(CAC)回收期,以及头部语音代理平台账户的收入集中度等关键承保输入,仍属于未披露的私人指标。[CI008, CI009, CI010, CI012, CI013, CI014]
| 单位指标 | 报告值 / 估算值 | 置信度 | 对投资判断的意义 | 尽调核实事项 |
|---|---|---|---|---|
| 毛利率 | 85%+(Sacra 估算) | 中 | 尽管深度学习推理密集,仍显示出软件级经济性 | 经审计 GAAP 毛利率及 GPU 计算托管拆分 |
| 年度经常性收入(ARR) | $100M ARR(Sacra 估算,2026 年 8 月) | 中 | 验证 ARR 同比增长 115% 的轨迹及独角兽规模 | 月度 ARR 桥接表、客户数核对及流失率 |
| 净收入留存率(NRR) | 未披露(未公开披露) | 低 | 语音 AI 用量扩大后,衡量账户扩张的关键指标 | 各档位的队列留存曲线及净美元扩张率 |
| 客户集中度 | 未披露(1,300+ 家机构基于平台构建产品) | 低 | 衡量对头部语音代理平台(如 Sierra、Decagon)的收入暴露 | 前 5 大及前 10 大客户贡献的收入比例 |
毛利率和 ARR 指标来自第三方研究估算(Sacra);Deepgram 是一家私营公司,尚未公布经核验的审计财务报表。未披露项表示私募投资判断中的信息缺口。
[CI008, CI009, CI010, CI011]4.5 资本充足性、现金跑道与投资判断
Deepgram 于 2026 年 1 月完成 $130 million C 轮融资、投后估值 $1.3 billion,大幅加强资产负债表;领投方为全球科技投资平台 AVP,现有风险投资人 Tiger Global、Madrona、In-Q-Tel、BlackRock 及企业战略投资人继续参与。管理层表示,资金将用于拓展欧洲和亚太市场、采购大规模计算基础设施、扩大专利组合,以及整合餐饮语音自动化初创公司 OfOne。二级投资数据库显示,公司多轮累计融资在 $215 million 至 $221 million 之间,但第三方对过去盈利能力的说法相互冲突。从投资判断看,Deepgram 展现出强劲商业 traction 和鲜明技术定位,有望成为下一代 AI 代理的基础语音层。然而,$1.3 billion 估值对应激进的企业价值倍数;若没有验证经审计 ARR、毛利率、现金消耗率和客户流失动态的确认性尽调,无法完整评估这一估值。[CI001, CI002, CI009, CI010, CI013, CI014]
4.6 展示项
05产品与技术
5.1 产品组合与能力
Deepgram 提供面向企业、API 优先的语音 AI 平台,把语音识别、音频智能和语音合成统一到一致的开发者界面。核心能力包括高准确率语音转文本(STT)、自然流畅的文本转语音(TTS)和实时 Voice Agent 编排,省去拼接多家供应商点状方案的运营复杂度。平台既支持通过 WebSocket 连接传输实时双向音频,也支持通过标准 REST 端点批处理预录音频文件。Deepgram 的模型系列同时提供 Nova-3 通用转录,以及 Flux 对话优化处理;后者原生理解轮转,并能处理交互式语音代理中的打断。除核心音频转录外,Deepgram 还提供说话人分离、敏感号码自动脱敏、情感分析和自定义关键词提示等音频智能能力,以处理专业领域术语。[CE001, CE002, CE003, CE004, CE005]
| 模块 / 资产 | 目标用户 / 买方 | 成熟度 / 状态 | 技术差异化 | 尽调缺口 |
|---|---|---|---|---|
| Nova-3 语音转文字 | 企业开发者与 CX 工程师 | 正式商用 | 端到端深度学习架构;基础费率 $0.0048/min | 不同口音音频上的独立词表外 WER 基准 |
| Flux STT / TTS | 对话式 AI 代理开发者 | 正式商用 | 具备原生轮次检测和打断处理的对话感知语音模型 | 多说话人背景噪声下的生产环境打断准确率 |
| Aura-2 文字转语音 | 语音应用开发者 | 正式商用 | 40+ 种富有表现力的声音人格,以及低于 200ms 的首字节流式延迟 | 大型企业并发下字符到音频的计算成本曲线 |
| 语音代理 API | AI 管家和联络中心构建者 | 正式商用 | 统一单一 API 编排 STT、LLM 推理和 TTS;支持自带模型 | 第三方 LLM 限流期间的编排故障转移行为 |
产品线和模型已根据截至 2026 年 8 月的 Deepgram 官方文档及定价指南核验。
[CE001, CE002, CE003, CE004]5.2 运行架构与部署模式
Deepgram 的技术基础不同于传统多阶段语音识别流水线,而是采用端到端深度学习架构,直接把声学波形特征映射到文本和对话状态。该架构带来低于 300 milliseconds 的流式转录中位延迟,以及低于 200 milliseconds 的文本转语音首字节响应时间,使对话节奏接近自然人速。Deepgram 支持灵活的部署拓扑:既提供跨全球可用区的多租户托管云基础设施,也提供面向受监管企业的完全隔离本地容器化环境。统一 Voice Agent API 在单一网络连接中管理实时音频传输、端点检测和 LLM 编排,减少延迟叠加和网络往返。公司还支持自定义模型训练和微调,让医疗、法律服务等专业领域的企业客户能针对专有词汇优化声学和语言表示。[CE006, CE007, CE008, CE009, CE010]
| 用户任务 | 当前工作流 | 公司方案 | 可衡量收益 | 技术限制 |
|---|---|---|---|---|
| 实时联络中心转写 | 使用独立声学模型的级联式开源管线 | 带说话人分离的 Nova-3 流式 WebSocket STT | 延迟低于 300ms,批处理中 30 秒处理 1 小时音频 | 附加功能使实际每分钟消耗成本翻倍 |
| 交互式语音代理执行 | 多供应商拼接 STT、LLM 和 TTS,存在往返延迟 | Deepgram 统一语音代理 API | 音频交付低于 200ms,并检测对话轮次 | 需要可靠的客户端音频流和网络稳定性 |
| 医疗与临床笔记听写 | 手工转写或不合规的云端语音服务 | 配置 HIPAA、带 BAA 且不保留模型训练数据的 STT | 加密传输及自动 PII 脱敏 | 传输健康信息前必须正式签署 BAA |
| 受监管的本地语音分析 | 数据驻留要求禁止向云 API 传输 | 自托管容器化 Deepgram 推理引擎 | 无需向外部传输数据的隔离式语音智能 | 本地 GPU 内存占用高,内部硬件维护负担重 |
根据公开部署模式及定价文档评估生产工作流取舍。
[CE005, CE006, CE007, CE008]Deepgram 语音 AI 平台从音频接入到部署环境的六层架构栈。
- 客户接入与传输层
- 双向 WebSocket 音频流和 REST 批量接入,配合客户端 SDK
- 对话编排层
- Voice Agent API 协调原生轮次检测、打断处理和 LLM 桥接
- 核心神经模型层
- 专有 Nova-3 STT、Flux 对话模型和 Aura-2 TTS 神经引擎
- 音频智能与分析层
- 说话人分离、自动脱敏、自定义关键词提示和情感分析
- 企业安全与治理层
- HIPAA BAA 边界、SOC 2 Type II 控制,以及按请求选择不用于训练的控制
- 基础设施与部署运行时
- 多区域托管公有云和隔离网络自托管容器化运行时
基于 Deepgram 架构出版物和服务规格提炼的概念性功能分层。
[CE003, CE006, CE007, CE008, CE010, CE012]5.3 集成、合规与技术风险
Deepgram 通过记录在案的生产模式满足企业治理要求,覆盖 HIPAA 合规、SOC 2 Type II 认证和 GDPR 对齐。处理受保护健康信息的企业客户,必须在传输患者音频前签署正式的业务伙伴协议(BAA);API 客户也能在请求层明确选择不将音频载荷用于模型再训练。尽管治理机制已经成型,部署层面仍有技术和尽调风险。Deepgram 官方转录模型目前覆盖约 36 至 40 种语言,落后于 Gladia 等专业竞争者的 100+ 语言广度。商业语音代理部署也有成本上升风险:基础转录价格从每分钟 $0.0048 起,叠加说话人分离、关键词提示和脱敏后可能超过每分钟 $0.0101。综合正常运行时间历史、准确的本地 GPU 硬件加速依赖,以及公开 SDK 基准测试套件仍属私有,构成重大尽调缺口。[CE011, CE012, CE013, CE014, CE015]
| 层 / 组件 | 角色与功能 | 关键依赖 | 已识别的技术风险 |
|---|---|---|---|
| 声学特征编码器 | 将原始音频波形转换为嵌入的端到端神经网络 | NVIDIA GPU 加速及专用 CUDA 深度学习内核 | GPU 计算强度高,硬件供应可能受限 |
| 对话轮次引擎 | Flux 模型原生的轮次检测和打断处理 | 双向 WebSocket 音频流协议 | 易受客户端声学回声消除和数据包抖动影响 |
| 语音合成引擎 | Aura-2 流式 TTS,生成富有表现力的多口音语音输出 | 字符级音位模型和神经声码器 | 非英语外来词和自定义缩写的发音漂移 |
| 治理与训练隔离 | 零数据留存标记执行及 BAA 合规政策 | 内部 API 遥测过滤和安全日志隔离 | 客户配置错误,导致每次请求未发送退出选择标头 |
根据工程规格和服务条款综合分析架构组件风险。
[CE009, CE010, CE011, CE012]5.4 展示项
06客户
6.1 客户基础分层与买家画像
Deepgram 的客户基础分成两端:大规模企业平台和自下而上的开发者团队。企业层由 Sierra、Decagon、Cresta 和 Kore.ai 等 AI 应用独立软件供应商(ISV)领头,也包括 IBM、Twilio 和 Spotify 等把对话智能直接嵌入核心产品工作流的全球企业技术伙伴。技术买家主要是首席技术官、工程副总裁和 AI 负责人;他们看重 Deepgram 的实时流式延迟、嘈杂真实声学环境下的生产准确率,以及统一的语音代理编排。同时,Deepgram 还拥有超过 200,000 名注册从业者的广泛自助开发者层,他们通过按消费计费的应用程序编程接口(API)接入,并升级到付费 Growth 方案。[CU001, CU002, CU003, CU004]
| 细分市场 | 买方 / 用户 / 付款方 | 主要用例 | 规模与触达 | 收入 / 战略价值 | 尽调缺口 |
|---|---|---|---|---|---|
| AI 平台 ISV 与代理 | CTO / AI 负责人 / 工程负责人 | 语音代理编排、实时客户体验机器人 | 高(如 Sierra、Decagon、Cresta、Kore.ai) | 高用量消耗;平台增长的核心驱动 | 未披露客户层面的流失率和合同期限 |
| 企业 IT 与全球科技 | 企业架构师 / 工程副总裁 | 嵌入式语音服务、企业代理构建、通话智能 | 高(如 IBM watsonx、Twilio、Spotify、NASA、Citibank) | 战略性多年渠道合作及企业许可 | 直销与嵌入式合作伙伴档位之间的收入分配未知 |
| 开发者 / Growth 档位 | 独立软件开发者与初创公司 | API 转写、音频智能、会议摘要 | 已注册开发者 200,000+;自助注册 | 自助消费额度,为 Growth 档位($4k/yr)导流 | 从 $200 免费额度转为付费账户的转化率未报告 |
| 垂直领域 | 客户体验副总裁 / 临床运营 | 联络中心自动化、医疗合规听写、零售 | 多垂直领域(医疗、金融、通过 OfOne 覆盖的酒店业) | 高级功能增购(脱敏、专用模型、VPC) | 垂直集中度和净收入留存率(NRR)未经核验 |
客户细分综合自 Deepgram 官方客户披露、IBM 合作公告、Tooliverse 评价及 Sacra 行业研究。
[CU001, CU002, CU003, CU004]6.2 采用轨迹、用量扩张与开发者速度
Deepgram 的商业采用轨迹显示,转录和实时语音代理接口的用量都在复合增长。根据 Sacra 的独立行业研究,Deepgram 截至 2024 年连续 4 年实现每年 3.3x 的平台用量增长,服务超过 400 家企业客户;到 2026 年 8 月,ARR 估计超过 $100 million。2026 年 1 月,已有超过 1,300 个不同组织在 Deepgram 基础设施上构建应用,累计处理音频超过 50,000 年,转录词数超过 1 trillion。基于用量的扩张形成自我强化飞轮:音频输入增加,直接改善专有深度学习模型对不同口音和声学环境的适应,进一步降低企业错误率,并加速单个账户的用量扩张。[CU005, CU006, CU007, CU008]
| 指标 | 数值 | 日期 | 来源 | 置信度 | 含义 | 缺失分母 |
|---|---|---|---|---|---|---|
| 机构总数 | 1,300+ 家机构 | 2026-01 | Sacra / Silicon Valley Invest Club | 中 | 在企业和初创公司领域实现广泛市场渗透 | 活跃付费机构与免费 / 休眠机构的拆分 |
| 开发者社区 | 200,000+ 名开发者 | 2026-02 | IBM Newsroom / Deepgram | 高 | 开发者采用量巨大,为自下而上的自助渠道供给流量 | 月活跃 API 用户与累计注册数 |
| 企业客户群 | 400+ 个企业账户 | 2024-12 | Sacra Company Profile | 中 | 高价值企业账户基础规模可观 | ACV 分布及最低消费门槛 |
| 用量增长 | 每年 3.3x(连续 4 年) | 2024-12 | Sacra Company Profile | 中 | 转写和语音 API 的消费量持续复合增长 | 每年处理的音频基准小时数 |
| 累计处理音频 | 50,000+ 音频年 | 2026-01 | Deepgram / Sacra | 高 | 广泛的专有数据反馈闭环推动模型准确率 | 年化运行速率小时数与历史总量 |
轨迹数据汇总自 Sacra 研究、IBM 合作公告及 Silicon Valley Invest Club 资料;私营公司的财务分母仍未经核验。
[CU005, CU006, CU007, CU008]6.3 已点名客户部署与企业生产验证
一线企业技术领导者和高速成长的对话式 AI 供应商,都为其生产采用提供了佐证。2026 年 2 月,IBM 宣布战略合作,选择 Deepgram 作为首个语音伙伴,把 Deepgram 的语音转文本和文本转语音模型直接嵌入 IBM watsonx Orchestrate Agent Builder,为全球金融和医疗客户提供对话式企业工作流。与此同时,Sierra 和 Decagon 等对话式 AI 领导者依靠 Deepgram 实时 API 执行面向客户的语音客服和代理工作流,并强调其任务关键型可靠性和低延迟。电信巨头 Twilio 将 Deepgram 的流式识别嵌入实时通信渠道。这些生产部署验证了 Deepgram 在企业规模下提供运营商级可用性和低延迟的能力。[CU001, CU009, CU010, CU011]
| 客户 | 细分市场 | 部署 / 用例 | 生产环境还是试点 | 结果 / 收益 | 限制 / 风险 |
|---|---|---|---|---|---|
| IBM (watsonx Orchestrate) | 企业 IT / 科技合作伙伴 | 生成式 AI 代理构建器中嵌入语音 STT/TTS | 生产环境合作集成 | 为 watsonx 增加自然对话语音和地区方言能力 | 取决于 IBM 客户对 watsonx 语音能力的采用 |
| Sierra | AI 平台 ISV | 面向客户体验 AI 代理的实时语音栈 | 生产环境部署 | 为关键客户服务提供企业级准确率和可靠性 | 客户合同条款及 Deepgram 消费分成未披露 |
| Decagon | 对话式 AI 平台 | 面向 AI 管家代理的企业级语音能力 | 生产环境部署 | 真实客户交互中可靠性高、延迟低且准确 | 需接受与替代语音 API 的竞争基准比较 |
| Twilio | 通信平台 | 将语音识别嵌入客户语音工作流 | 生产环境部署 | 低延迟流式转写实现规模化集成 | Twilio 运营多个合作伙伴集成,并保留战略选项 |
具名客户已通过 Deepgram 官方客户公告、IBM 合作公告及经核验的 Tooliverse 客户资料确认。
[CU001, CU009, CU010, CU011]6.4 扩张动力、开发者摩擦与留存持久性
客户通常先通过自助开发者试用接入,再逐步转向多年期企业承诺。开发者可获得 $200 不过期免费额度和快速软件开发工具包(SDK)集成,几分钟内就能基准测试实时 WebSocket 流式处理,然后启用 Pay-As-You-Go 自动充值或每年起价 $4,000 的 Growth 合同。企业账户会从基础语音转文本扩展到 Aura-2 文本转语音和统一 Voice Agent API,后者价格为每小时 $4.50。但客户尽调也暴露出两个阻力:经验证的评价指出,说话人分离等高级功能存在不小的学习曲线,文档有时落后于产品发布;同时,净收入留存、毛收入留存和客户集中度等私人财务留存指标仍未披露。[CU012, CU013, CU014]
客户旅程从开发者发现和评估,经过生产部署、年度承诺,扩展到多产品语音技术栈。
[CU012, CU013]6.5 展示项
07风险
7.1 高管风险分类与剩余敞口
评估 Deepgram 的风险,需要把其语音转文本技术实力,与对话代理编排和多模态市场动态带来的新脆弱性放在一起看。Deepgram 用统一架构整合语音转文本、文本转语音和编排,以降低生产语音应用的延迟;但这种统一界面也把不同失效模式紧密耦合。随着语音 AI 从被动批量转录转向实时对话代理,执行风险会从单纯优化词错误率,转向轮转可靠性、打断处理和连续音频流稳定性。若客户把语音模型垂直整合,对话式 AI 平台和联络中心伙伴的客户集中度会让 Deepgram 暴露于下游流失。若专业音频竞争者在合成能力上超过 Deepgram,或客户要求增加支持复杂度的本地治理架构,剩余敞口仍会升高。[CR001, CR013]
按可能性、影响、缓解成熟度和剩余暴露,评估核心运营、监管、竞争、合作伙伴和财务风险。
可能性和影响等级反映第三方构建者评估、服务条款约定和市场估值倍数。
[CR001, CR003, CR005, CR010, CR013]7.2 监管、隐私与法律敞口
语音 AI 平台在全球各司法辖区面临日益严格的数据保护、消费者隐私和生物特征信息监管。Deepgram 服务条款严格禁止对模型权重进行逆向工程,也禁止竞争性基准测试或模型训练,反映出基础模型激烈竞争下的强势知识产权防御。为处理法律争议、降低集体诉讼敞口,Deepgram 要求客户争议依据加州法律,通过具有约束力的个人仲裁解决,并放弃集体诉讼。合规方面,医疗部署需要业务伙伴协议和数据保护措施,以避免受保护健康信息泄露带来的高昂责任。虽然 Deepgram 为 HIPAA、SOC 2 和 GDPR 工作负载发布参考架构,企业客户仍需承担区域隐私框架下的生物特征同意获取和数据最小化责任。[CR002, CR006, CR007]
| 规则 / 许可 / 案件 | 管辖区 | 状态 | 可能性 | 严重性 | 缓释措施 | 剩余暴露 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| HIPAA / HITECH 合规 | 美国 | 运营标准 | 低 | 严重 | 商业伙伴协议、零留存默认设置及实时 PHI 脱敏参数 | 客户配置错误,导致 PHI 暴露在客户应用日志中 | 审计已签署的 BAA 模板及医疗客户实施架构样本 |
| GDPR / 欧盟 AI 法案语音处理 | 欧盟 | 生效中的监管 | 中 | 高 | 欧盟区域 API 端点、数据最小化及传输和存储全程加密 | 生物特征分类审查及跨境音频传输限制 | 审阅欧盟端点架构及数据处理附录文件 |
| 服务条款中的逆向工程限制 | 美国 / 全球 | 合同约定 | 中 | 中等 | 明确的知识产权限制、禁止将输出用于竞争,以及 API 速率监控 | 难以发现第三方模型蒸馏和合成数据抓取 | 检查 API 遥测监控工具和滥用检测协议 |
| 具有约束力的仲裁与争议解决 | 美国(加州) | 强制条款 | 低 | 中等 | 强制 AAA 个别仲裁条款,并设有 60 天非正式协商期 | 大规模仲裁申请或州法院挑战代表诉讼弃权条款 | 查阅标准商业主服务协议,确认是否存在非标准争议修改条款 |
风险评估基于 Deepgram 服务条款、公开的企业合规文件,以及美国 / 欧盟监管标准。
[CR001, CR002, CR006, CR007]7.3 运营基础设施、模型边界案例与安全
在生产环境部署实时语音代理,会带来与传统异步 API 服务明显不同的运营挑战。Deepgram 的 Voice Agent API 比其转录模型更新,在复杂打断处理上可能存在瑕疵,尤其是在嘈杂的多人声学环境中。为应对企业数据主权和安全问题,Deepgram 默认采用零留存处理,并提供可配置的实时脱敏参数,以满足 PCI 和 SSN 数据的 SOC 2、HIPAA 及 GDPR 标准。处理敏感生物特征音频的企业客户,还能把 Deepgram 部署在自托管或 VPC 环境中,让语音数据留在内部网络边界内。但自托管部署需要定制模型交付、客户工程投入和复杂维护,而公开定价表并不透明。[CR003, CR008, CR009]
| 故障模式 | 可能性 | 严重性 | 缓解成熟度 | 剩余暴露 | 未解决缺口 |
|---|---|---|---|---|---|
| 轮次切换与打断延迟 | 中 | 高 | 中等 | 实时用户对话中语音代理突然截断,以及令人尴尬的响应停顿 | 覆盖高噪声联络中心声学流的生产边界案例基准数据 |
| 线性用量计费冲击 | 高 | 中等 | 初期 | $200 免费试用额度用完后,开发者流失、原型被放弃 | 透明的用量折扣阶梯和自助式支出上限控制 |
| 生物特征语音数据泄露 | 低 | 严重 | 成熟 | 未加密语音录音或临时内存缓冲区被未经授权访问 | 独立 SOC 2 Type II 审计报告,核实零留存内存清除机制 |
| 自托管部署复杂度 | 中 | 中等 | 中等 | 工程支持负担重,企业销售上线周期漫长 | 面向客户的部署自动化脚本和隔离网络更新遥测 |
运营风险来自独立开发者评测、架构指南和技术部署文档。
[CR003, CR004, CR008, CR009]7.4 伙伴依赖、平台集中度与市场风险
Deepgram 的市场定位同时承受底层基础设施约束和语音技术栈竞争替代方案的压力。独立开发者评价指出,Deepgram 免费层按额度计费而非永久免费,因此开发者账单会随音频量线性增长。Deepgram 的文本转语音产品比 ElevenLabs 和 Fish Audio 等专业音频平台提供更少的声音选项,也缺少情绪标签或定制声音克隆。基础设施方面,企业语音基础设施仍容易受到云托管集群中断和各区域可用区 GPU 容量瓶颈影响。与此同时,Deepgram 于 2026 年 1 月以 $1.3 billion 估值完成 $130 million C 轮融资;相对历史收入,这形成较高估值倍数,需要持续三位数用量扩张才能支撑。[CR004, CR005, CR010, CR012]
| 依赖项 | 交易对手方 | 角色 | 集中度 | 失败情景 | 严重性 | 缓解措施 | 剩余暴露 |
|---|---|---|---|---|---|---|---|
| GPU 集群托管 | 主要云服务商(AWS / GCP / CoreWeave) | 实时模型推理的计算基础设施 | 高 | 数据中心宕机或 GPU 配额限流,导致 API 延迟飙升 | 严重 | 多区域故障转移路由,以及容器化自托管集群选项 | 大范围云基础设施中断期间出现短期区域性 API 降级 |
| 对话平台分发 | 代理框架(Sierra / Decagon / Vapi) | 下游企业语音代理应用部署 | 高 | 关键合作伙伴转用其他语音供应商或将模型内化 | 高 | 联合营销协议和深度架构联合开发 | 头部代理平台分散音频 API 供应商后,收入收缩 |
| 文本转语音功能落后 | 音频竞争者(ElevenLabs / Fish Audio) | 备选语音合成和声音克隆供应商 | 中 | 客户采用多供应商竞争方案,以获得更高的语音保真度 | 高 | 统一单一 API 计费和低延迟 Voice Agent WebSocket 打包方案 | 失去需要情感语音控制的创意及媒体语音客户 |
| 估值下行重估 | 后期资本市场(Series C 银团) | 成长资本和投资者估值预期 | 高 | 增长放缓触发倍数压缩,估值跌破 $1.3B | 高 | 扩大高毛利 Voice Agent API 用量和企业年度合同 | 增长停滞时面临大幅股权稀释或下轮融资风险 |
评估外部云基础设施、合作伙伴分发、竞争性语音技术和资本市场暴露。
[CR005, CR010, CR011, CR012]7.5 监测指标、尽调重点与论点破坏触发器
要评估 Deepgram 的增长轨迹,必须严密监测私人财务指标、企业留存持久性和竞争替代信号。第三方估计 Deepgram 于 2026 年 8 月达到 $100 million ARR,但未经审计的私人数据让合同持久性无法验证。因此,未经验证的私人财务、毛利率走势和净收入留存率,是 2026 年投资人的必要持续治理尽调项目。关键论点破坏触发器包括:主要对话式 AI 客户自建专有语音模型;大型云厂商大幅下调实时语音 API 价格;以及 Voice Agent API 在任务关键型联络中心轮转失败,导致企业流失。尽调必须关注客户群续约趋势、OfOne 收购整合后的利润率,以及 GPU 计算成本上升。[CR011, CR014]
7.6 展示项
08估值
8.1 投资论点与反论点
Deepgram 处在一个令人羡慕的位置:它是 API 优先的基础语音 AI 基础设施平台,为超过 1,300 家企业和 200,000 名开发者处理实时语音转文字、文字转语音以及全双工语音代理交互。看多投资逻辑押注于其专有端到端深度学习架构、受 2025 年已获授权专利保护的硬件效率,以及向高价值对话式语音自动化的快速扩张。反方逻辑则指出,核心单位经济性存在严重信息不对称;第三方跟踪机构给出的年度经常性收入估计从 $21.8M 到 $240M 不等且未获验证;超大规模云厂商和前沿模型实验室带来激烈的平台竞争;最近 $130M 的增长资本还带来了显著的清算优先权悬垂。[CV001, CV002, CV003, CV005, CV006, CV010, CV012]
| 视角 | 核心论点 | 支持证据 | 哪些情况会改变判断 |
|---|---|---|---|
| 投资论点 | 低延迟、拥有专有深度学习模型的品类定义型 Voice AI 基础设施 | 超过 1,300 家客户组织、20 万名开发者、低于 200ms 的延迟,以及硬件效率专利 | 企业开发者采用放缓,或无法将对话式语音代理变现 |
| 投资论点 | 企业覆盖面扩大并整合并购,形成可守住的垂直壁垒 | 为 QSR 得来速场景收购 OfOne,以及 AVP、ServiceNow 和 EDBI 的全球支持 | 餐厅自动化执行失败,或并购资产整合不佳 |
| 投资反论点 | 财务极不透明,第三方年度经常性收入估算未经核实 | 公开估算差异巨大:$21.8M(2024 Latka)、$100M(Sacra)和 $240M(Multiples.vc) | 提供经审计财务报表,证明 ARR 超过 $100M 且单位经济性为正 |
| 投资反论点 | 面对超大规模云服务商和前沿模型实验室定价时,估值容易下行 | OpenAI、Google 及专门的开放权重模型压缩实时 STT 和 TTS API 的经济性 | 证明客户拥有定价能力,且净收入留存率持续超过 130% |
核心论点来自经核实的技术能力、专利申请、客户基础披露和第三方财务跟踪。
[CV001, CV002, CV003, CV005, CV006, CV007, CV010, CV012]8.2 融资背景与入场纪律
2026 年 1 月,Deepgram 获得由 AVP 领投的 $130M C 轮融资,投后估值为 $1.3B,较 2022 年 11 月 $379M 的 B 轮扩展轮上升 3.4 倍。本轮吸引了 Alkeon、Madrona、Tiger Global 等顶级机构投资者,以及 ServiceNow Ventures、Twilio、SAP 等战略投资者;公司同时收购了餐饮 AI 免下车场景专家 OfOne。新加坡 EDBI 于 2026 年 8 月提供战略性扩展支持,进一步强化了其全球 APAC 扩张计划。不过,公司在多轮优先股融资中累计募集的股权资本达到 $215M 至 $229M,若按表面上的 $1.3B 估值进入成长股权投资,就必须严格执行入场纪律:公开指标缺少经审计的净留存率、客户集中度拆分和毛利率验证。[CV001, CV002, CV004, CV007, CV008, CV009, CV013, CV014]
| 维度 | 评估 | 证据依据 | 决策含义 |
|---|---|---|---|
| 建议 | 跟踪 | 技术实力强,但私人公司财务透明度低,且对外标示的价格为 $1.3B | 在数据室审计完成前,按季度跟踪,不投入资本 |
| 置信度 | 中 | 多份独立报告确认融资规模为 $130M,但对 ARR($21.8M-$240M)的说法相互冲突 | 对二级市场估值标记打折,等待一手财务披露 |
| 风险评级 | 高 | 前沿实验室商品化、API 价格下滑,以及 $130M 高级清算优先权形成的悬压 | 任何直接投资都要求结构性下行保护或高级证券 |
| 估值立场 | 合理 | 按 Sacra 估算的 $100M ARR 计算,隐含 13.0x 倍数,与成长型 AI 基础设施基准一致 | 在客户集中度和流失率缺乏清晰可见性的情况下,切入价格缺乏吸引力 |
评估综合 Series C 交易证据、Sacra 收入研究,以及截至 2026 年 10 月的私营 AI 可比倍数。
[CV001, CV003, CV005, CV006, CV011]将技术差异化和增长规模连接到估值风险及最终跟踪建议的决策路径。
| 节点 / 连接 | 节点 / 起点 | 终点 | 说明 |
|---|---|---|---|
| 节点 1 | 1,300+ 家企业客户与自研 STT/TTS 平台 [scale] | 超过 200,000 名开发者基于 Nova-3、Aura-2 和 Flux 模型构建产品 | |
| 节点 2 | $130M Series C 轮,估值 $1.3B [funding] | AVP 领投,ServiceNow、Twilio 和 SAP 战略参与 | |
| 节点 3 | 全球布局扩张与设于新加坡的亚太总部 [extension] | EDBI 领投 Series C 轮追加融资,支持国际增长和企业客户覆盖 | |
| 节点 4 | 严重的财务与 ARR 披露不透明 [opacity] | 第三方估算在 $21.8M、$100M 和 $240M 之间相互冲突,且没有经验证的审计 | |
| 节点 5 | Frontier Lab 商品化压力与优先清算权负担 [risk] | 基础模型供应商带来的价格竞争,加上 $215M+ 的高级优先股堆叠 | |
| 节点 6 | 审慎立场:跟踪建议 [decision] | 在获得数据室访问权限并完成单位经济审计之前,暂不评估增长资本投入 | |
| 连接 1 | 1,300+ 家企业客户与自研 STT/TTS 平台 [scale] | $130M Series C 轮,估值 $1.3B [funding] | |
| 连接 2 | $130M Series C 轮,估值 $1.3B [funding] | 全球布局扩张与设于新加坡的亚太总部 [extension] | |
| 连接 3 | 全球布局扩张与设于新加坡的亚太总部 [extension] | 严重的财务与 ARR 披露不透明 [opacity] | |
| 连接 4 | 严重的财务与 ARR 披露不透明 [opacity] | Frontier Lab 商品化压力与优先清算权负担 [risk] | |
| 连接 5 | Frontier Lab 商品化压力与优先清算权负担 [risk] | 审慎立场:跟踪建议 [decision] |
概念性流程图,综合定性优势、量化融资里程碑和尽调决策门槛。
[CV001, CV002, CV003, CV005, CV009, CV011]8.3 情景分析与市场可比公司
我们的估值框架构建看多、基准和悲观三种情景,评估 Deepgram 到 2027 年在不同商业执行路径下的表现。基准情景假设独立研究跟踪机构 Sacra 建模的 $100M ARR 以及扩大的语音代理市场,认为 Deepgram 配得上 13.0x 倍数,对应 $1.3B 估值。看多情景预计语音优先的企业采用加速,推动 ARR 达到 $150M,按 18.0x 倍数对应 $2.7B 估值;悲观情景则假设竞争极度激烈、定价承压、模型商品化,私人市场 ARR 在约 $50M 附近停滞,最终按 8.0x 倍数将估值下调 69% 至 $400M。SoundHound AI 等对话式 AI 同业,以及 C3.ai 等企业 AI 软件公司的交易倍数表明,增长放缓时,私有基础设施公司的估值倍数很容易暴露于市场风险。[CV005, CV006, CV008, CV011, CV012, CV013]
| 情景 | 概率信号 | 关键假设 | 2027 年估值目标 | 下行 / 回报逻辑 |
|---|---|---|---|---|
| 乐观情景 | 25%(选择性) | 语音代理市场快速扩张;ARR 达到 $150M;毛利率维持在 80% 以上;企业合同扩大 | $2,700M(18.0x ARR) | 相对 Series C 投后估值的 MOIC 为 2.1x;有望走向高倍数上市 |
| 基准情景 | 50%(核心展望) | API 用量稳步增长;ARR 增至 $100M;企业客户覆盖达到 2,000 家;利润率温和压缩 | $1,300M(13.0x ARR) | MOIC 为 1.0x;Deepgram 成长为其 2026 年成长轮融资规模的同时,估值持平 |
| 悲观情景 | 25%(下行风险) | 前沿实验室将 STT/TTS 商品化;ARR 停滞在 $50M;大幅降价,客户流失至开源模型 | $400M(8.0x ARR) | -69% 估值回撤;严重的下轮融资风险触发高级清算优先权,抹掉初级股权 |
估值倍数以私营 AI 基础设施融资轮和企业软件可比公司为基准;收入输入反映建模区间。
[CV001, CV004, CV005, CV006, CV008, CV011, CV013]8.4 投资建议与尽调重点
我们给 Deepgram 的投资建议是跟踪,信心中等、风险高,估值判断为合理。Deepgram 的产品延迟表现出色,SOC 2 Type 2 合规扎实,也已获得战略性企业客户进展;但在无法直接接触私人财务分母的情况下,按 $1.3B 估值承销增长资本,意味着承担了未获补偿的风险。若要改为买入建议,必须完成三方面的确定性验证:经审计的利润表,确认毛利率是否超过 80%;按客户 cohort 分组的净收入留存率,验证语音代理合同是否扩张;以及精确的股权结构表清算分配瀑布,明确高级清算优先权和员工期权池稀释。[CV003, CV009, CV010, CV011, CV014]
8.5 展示项
免责声明
本报告是基于公开证据的尽调快照,不构成投资建议。重要的财务、法律、技术和合同事实仍未公开,任何投资决策前都应直接向管理层及通过一手文件核实。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | Deepgram, Inc. was founded in 2015 as a Delaware corporation and maintains its corporate headquarters at 548 Market Street in San Francisco, California. | 高 | SO002, SO005 |
| CO002 | Deepgram was co-founded by University of Michigan physicists Scott Stephenson and Adam Sypniewski following machine learning waveform analysis research conducted on an underground dark matter detector in China. | 高 | SO002, SO005 |
| CO003 | Deepgram delivers an API-first voice AI infrastructure platform spanning speech-to-text via Nova-3, text-to-speech via Aura-2, and conversational speech-to-speech via its Voice Agent API. | 高 | SO001, SO004 |
| CO004 | Deepgram utilizes an end-to-end deep learning neural network architecture that processes raw audio waveforms directly rather than decomposing speech into intermediate phonetic components. | 高 | SO002, SO004 |
| CO005 | Deepgram's executive leadership team includes co-founder and Chief Executive Officer Scott Stephenson, co-founder and Chief Technology Officer Adam Sypniewski, and Chief Operating Officer Shadi Baqleh. | 中 | SO005 |
| CO006 | Deepgram announced a $130 million Series C equity financing round in January 2026 led by AVP at a $1.3 billion post-money valuation. | 高 | SO001, SO004, SO006 |
| CO007 | Deepgram has raised more than $215 million in cumulative venture capital financing across seven investment rounds since graduating from Y Combinator's Winter 2016 cohort. | 中 | SO004, SO005 |
| CO008 | Deepgram's institutional investor syndicate includes AVP, Madrona, Alkeon, Tiger Global, Wing, In-Q-Tel, BlackRock accounts, and corporate strategic backers Twilio, ServiceNow Ventures, SAP, and Citi Ventures. | 高 | SO001, SO004, SO006 |
| CO009 | Deepgram acquired Y Combinator-backed restaurant voice AI startup OfOne in January 2026 to launch Deepgram for Restaurants and automate quick-service drive-thru ordering workflows. | 高 | SO001, SO006 |
| CO010 | Deepgram reported that more than 1,300 commercial organizations and over 200,000 developers build on its voice AI API platform. | 高 | SO001, SO004 |
| CO011 | Deepgram prices its Nova-3 speech-to-text API at $0.0077 per minute for streaming and $0.0043 per minute for pre-recorded audio on pay-as-you-go tiers, with annual Growth plans requiring a $4,000 minimum spend. | 中 | SO003, SO004 |
| CO012 | Independent pricing analysis reveals that add-on features including speaker diarization ($0.002/min) and PII redaction ($0.002/min) each add approximately 26% to base transcription costs, increasing a fully-featured pipeline price by up to 69%. | 中 | SO003 |
| CO013 | Deepgram's standard Terms of Service mandate binding individual arbitration seated in San Francisco, California, expressly waiving class action participation and jury trials. | 中 | SO007 |
| CO014 | Deepgram provides zero-storage processing defaults and real-time redaction parameters designed to comply with HIPAA, SOC 2, and GDPR enterprise regulatory standards. | 中 | SO008 |
| CO015 | Third-party private-market researchers estimate Deepgram achieved $100 million in annual recurring revenue by August 2026, following historical reported revenue of $21.8 million in 2024, though audited financial statements remain undisclosed. | 中 | SO004, SO005 |
| CM001 | Deepgram positioned its real-time API platform to underpin the B2B Voice AI economy, raising $130 million in Series C financing at a $1.3 billion post-money valuation in January 2026. | 高 | SM001, SM009 |
| CM002 | Enterprise demand in speech recognition is shifting from offline batch transcription toward real-time conversational agents that require low-latency streaming and duplex turn-taking. | 高 | SM002, SM003 |
| CM003 | The global speech-to-text API market is forecasted by industry analysts to reach $8.3 billion by 2030, expanding at a 14.1% compound annual growth rate. | 中 | SM004 |
| CM004 | The global text-to-speech market represents an adjacent $10.0 billion addressable opportunity by 2029, growing at an annual rate of 19.1%. | 中 | SM004 |
| CM005 | Conversational voice agent automation intersects with a global contact center software and operational market exceeding $40 billion annually. | 中 | SM004 |
| CM006 | Deepgram monetizes across self-serve Pay As You Go consumption plans and annual Growth commitments starting at $4,000 annually, with enterprise contracts tailored for custom scale. | 高 | SM004, SM006 |
| CM007 | Highly regulated industry verticals such as healthcare require specialized HIPAA and SOC 2 compliance, where medical transcription spend is projected to reach $4.89 billion by 2027. | 中 | SM004, SM007 |
| CM008 | Healthcare data breaches average $10.93 million per incident, creating stringent security and data residency adoption barriers for cloud-hosted speech-to-text pipelines. | 中 | SM008 |
| CM009 | Production voice agents require end-to-end integration across speech-to-text, LLM orchestration, and speech synthesis to overcome compound latency bottlenecks. | 高 | SM002, SM003 |
| CM010 | Independent evaluation datasets rank Deepgram as one of the fastest streaming speech-to-text engines, while noting it is less suited for casual pre-recorded file transcription. | 中 | SM005 |
| CM011 | Sacra estimated Deepgram attained a $100 million annual recurring revenue run-rate milestone in August 2026, driven by real-time voice agent usage expansion. | 中 | SM004 |
| CM012 | Deploying voice AI in enterprise environments entails substantial switching friction due to legacy telephony integrations, custom vocabulary adaptation, and compliance sign-offs. | 中 | SM002, SM007 |
| CM013 | Deepgram expanded its addressable vertical footprint into quick-service restaurant drive-thru automation through its acquisition of AI voice platform OfOne. | 高 | SM001, SM009 |
| CM014 | The addressable market pyramid transitions from a broad $40B+ contact center operations layer down to a $1.5B-$2.5B immediate real-time voice agent SAM and Deepgram $100M ARR base. | 中 | SM004, SM001 |
| CP001 | The 2026 commercial speech AI landscape divides into specialized independent model labs (Deepgram, AssemblyAI, ElevenLabs, Speechmatics) and hyperscale cloud platforms (Google Cloud, Microsoft Azure, AWS). | 高 | SP001, SP009 |
| CP002 | Deepgram operates both first-party automatic speech recognition (Nova-3) and text-to-speech synthesis (Aura-2, Flux TTS), contrasting with STT-only specialists AssemblyAI and Speechmatics and TTS-led ElevenLabs. | 高 | SP001, SP003 |
| CP003 | Deepgram's Voice Agent API unifies speech-to-text, text-to-speech, and large language model orchestration into a single endpoint priced at .50 per hour of conversation. | 高 | SP002, SP003 |
| CP004 | OpenAI Whisper Large v3 serves as the primary open-weights substitute for batch transcription, though self-hosting incurs GPU infrastructure, inference latency, and streaming engineering overhead. | 高 | SP001, SP009 |
| CP005 | Sacra estimates Deepgram reached 00M in annual recurring revenue in August 2026 following 115% ARR growth entering 2025, supported by a 30M Series C round valuing the firm at .3B. | 高 | SP005, SP009 |
| CP006 | Deepgram models achieve sub-300 millisecond end-of-utterance transcription latency and 80 to 90 millisecond time-to-first-byte synthesis latency optimized for interactive conversational turn-taking. | 中 | SP001, SP004 |
| CP007 | Deepgram structures list pricing across Pay As You Go (/bin/bash.0043/min batch, /bin/bash.0077/min streaming for Nova-3), Growth (,000+/year with up to 20% discounts), and Enterprise (5,000+/year) tiers. | 中 | SP009 |
| CP008 | Speechmatics maintains an advantage in accented English and low-resource languages, whereas Deepgram's Nova-3 emphasizes real-time multilingual code-switching across major commercial languages. | 中 | SP001, SP010 |
| CP009 | Deepgram provides SOC 2 Type II certification, HIPAA compliance implementation guidelines, and self-hosted private cloud deployment options for regulated banking and healthcare customers. | 高 | SP006, SP007, SP008 |
| CP010 | Enterprise switching costs in voice AI are driven by bespoke audio vocabulary adaptation, custom acoustic fine-tuning, and production telemetry rather than raw API protocol lock-in. | 中 | SP001, SP009 |
| CP011 | Independent evaluations indicate that vendor-published benchmarks overstate real-world accuracy differentials because evaluation corpora are selected by the model provider. | 中 | SP001 |
| CP012 | Deepgram's Flux model family introduces conversation-aware end-of-turn detection and interruption handling directly into the speech runtime to mitigate multi-turn voice agent latency. | 中 | SP003, SP004 |
| CP013 | Well-funded voice AI challengers including Cartesia, Bolna AI, and Ringg AI raised fresh venture rounds in early 2026 to compete in low-latency voice infrastructure and regional deployments. | 中 | SP005, SP010 |
| CP014 | Multimodal real-time APIs from frontier labs like OpenAI present long-term displacement pressure on pipeline architectures that combine separate STT, LLM, and TTS components. | 中 | SP001, SP009 |
| CI001 | Deepgram announced a $130 million Series C financing round in January 2026 valuing the company at $1.3 billion. | 高 | SI001, SI002, SI005 |
| CI002 | The Series C financing was led by investment platform AVP with participation from Tiger Global, Madrona, In-Q-Tel, BlackRock, and strategic corporate investors. | 高 | SI001, SI002, SI005 |
| CI003 | Deepgram monetizes its platform through usage-based APIs spanning speech-to-text, text-to-speech, and unified conversational voice agent services. | 中 | SI003, SI004 |
| CI004 | Deepgram charges $4.50 per hour for its unified Voice Agent API, which integrates speech-to-text, LLM orchestration, and speech synthesis. | 中 | SI004, SI009 |
| CI005 | Deepgram structures pricing across Pay As You Go, Growth ($4,000+ annually), and Enterprise ($15,000+ annually) subscription plans. | 高 | SI004, SI006 |
| CI006 | Deepgram's Pay As You Go terms specify consumption-based auto-reloading credit balances, and terms stipulate that subscription fees and credits are non-refundable. | 中 | SI006 |
| CI007 | Developer sign-up begins with $200 in free credits, serving as a low-friction adoption funnel to generate self-serve API consumption. | 中 | SI004 |
| CI008 | Industry analyst Sacra estimates that Deepgram maintains gross margins of 85% or higher across its cloud and enterprise speech AI workloads. | 中 | SI004 |
| CI009 | Sacra estimates that Deepgram reached $100 million in annual recurring revenue in August 2026, after growing ARR 115% year-over-year entering 2025. | 中 | SI004 |
| CI010 | Deepgram grew its active customer base from 400+ enterprise accounts in 2024 to more than 1,300 organizations building on its platform by January 2026. | 中 | SI001, SI004, SI009 |
| CI011 | Deepgram claims its models are 2-5x more affordable than legacy cloud competitors while providing lower latency for real-time applications. | 中 | SI004 |
| CI012 | Enterprise compliance architectures supporting HIPAA and SOC 2 require dedicated deployment patterns that protect sensitive speech data against costly breaches. | 中 | SI007, SI008 |
| CI013 | Deepgram has not publicly disclosed verified audited revenue figures, gross margins, or profitability, resulting in conflicting third-party database estimates. | 中 | SI009 |
| CI014 | Deepgram's net revenue retention and customer concentration across major AI platform clients like Sierra and Decagon remain undisclosed private metrics. | 中 | SI003, SI004, SI009 |
| CE001 | Deepgram offers a unified voice AI platform encompassing speech-to-text, text-to-speech, and voice agent APIs across streaming and batch modalities. | 高 | SE001, SE003, SE004 |
| CE002 | Nova-3 serves as Deepgram flagship general-purpose speech-to-text model family, with streaming base pricing starting at $0.0048 per minute. | 中 | SE001, SE002 |
| CE003 | Deepgram provides an end-to-end Voice Agent API unifying transcription, LLM orchestration, and speech synthesis into a single WebSocket connection. | 中 | SE003, SE004 |
| CE004 | The Flux model family is engineered for conversational voice agents, integrating native turn-taking awareness and interruption handling. | 中 | SE001, SE003 |
| CE005 | Aura-2 text-to-speech provides over 40 enterprise voice personas with streaming latency emphasized at sub-200 milliseconds time-to-first-byte. | 中 | SE001, SE005 |
| CE006 | Deepgram processes audio using an end-to-end deep learning architecture rather than traditional multi-stage acoustic and language modeling pipelines. | 中 | SE001, SE005 |
| CE007 | Deepgram achieves streaming transcription latency under 300 milliseconds and batch processing throughput of approximately 30 seconds per hour of audio. | 中 | SE001, SE004 |
| CE008 | Deepgram supports both multi-tenant public cloud API access and self-hosted on-premise containerized enterprise deployments. | 高 | SE001, SE003, SE004 |
| CE009 | Deepgram allows custom model training and fine-tuning for specialized enterprise vocabularies and industry-specific acoustic conditions. | 中 | SE001, SE004 |
| CE010 | Deepgram Voice Agent API offers Bring-Your-Own-LLM and Bring-Your-Own-TTS configuration options with separate rate-card pricing tiers. | 中 | SE002, SE003 |
| CE011 | Deepgram terms of service require a signed Business Associate Agreement before transmitting Protected Health Information under HIPAA regulations. | 高 | SE006, SE007 |
| CE012 | Deepgram API customers can opt out of model training on a per-request basis by specifying request parameters. | 高 | SE006, SE008 |
| CE013 | Deepgram speech models support between 36 and 40 languages, trailing specialized competitors offering broader language portfolios exceeding 100 languages. | 中 | SE001, SE004 |
| CE014 | Combining transcription add-ons including speaker diarization, redaction, and keyterm prompting can raise effective streaming STT costs to $0.0101 per minute. | 中 | SE001, SE002 |
| CE015 | Deepgram product terms reference a consumer voice assistant application named Saga subject to distinct user settings. | 中 | SE006 |
| CU001 | IBM partnered with Deepgram in February 2026 to embed Deepgram's speech-to-text and text-to-speech models into watsonx Orchestrate as its first voice partner. | 高 | SU001, SU003 |
| CU002 | Deepgram serves multiple customer segments including AI platform builders, global enterprises, self-serve software developers, and domain-specific contact centers. | 中 | SU003, SU005 |
| CU003 | Deepgram offers subscription and billing tiers consisting of Pay-As-You-Go with auto-reload, Growth annual commitments starting at $4,000 per year, and custom Enterprise contracts. | 高 | SU002, SU006 |
| CU004 | Enterprise customers utilize compliance safeguards including HIPAA BAA support, SOC 2 Type II certification, zero-retention defaults, and real-time PII redaction. | 中 | SU002, SU007, SU008 |
| CU005 | Deepgram is utilized by over 1,300 organizations worldwide across startup and enterprise tiers as of January 2026. | 中 | SU004, SU005 |
| CU006 | More than 200,000 software developers build applications on top of Deepgram's voice platform APIs. | 中 | SU001, SU004 |
| CU007 | Deepgram grew audio usage 3.3x annually for four consecutive years through 2024 while serving over 400 enterprise customers. | 中 | SU005 |
| CU008 | Deepgram has transcribed over 1 trillion words and processed more than 50,000 years of audio across its foundational model platform. | 中 | SU001, SU005 |
| CU009 | IBM clients utilize Deepgram through watsonx Orchestrate Agent Builder to automate enterprise workflows and customer care with natural speech. | 中 | SU001 |
| CU010 | Conversational AI agents at Sierra and Decagon use Deepgram APIs to power real-time voice concierges and customer experience workflows. | 中 | SU003, SU005 |
| CU011 | Twilio integrates Deepgram's streaming speech recognition APIs to deliver low-latency voice infrastructure to communications customers. | 中 | SU002, SU004 |
| CU012 | Developer reviews indicate a steep learning curve for advanced capabilities such as speaker diarization and report that documentation occasionally lags fast product releases. | 中 | SU002 |
| CU013 | Deepgram provides $200 in free non-expiring credits and integrates in minutes, allowing developers to test low-latency streaming before committing to annual plans. | 中 | SU002, SU005 |
| CU014 | Public evidence does not disclose verified net revenue retention, gross revenue retention, or customer concentration percentages for Deepgram. | 中 | SU005 |
| CR001 | Deepgram's platform architecture unifies speech-to-text, text-to-speech, and orchestration to mitigate latency in production voice applications. | 高 | SR001, SR003 |
| CR002 | Deepgram terms of service strictly prohibit reverse engineering of model weights and competitive benchmarking or model training. | 高 | SR006, SR001 |
| CR003 | Deepgram's Voice Agent API is newer than its transcription models and can experience rough edges in complex interruption handling. | 中 | SR001 |
| CR004 | Independent builder reviews note that Deepgram's free tier is credit-based rather than permanent, causing developer bills to scale linearly with audio volume. | 中 | SR001 |
| CR005 | Deepgram's speech synthesis portfolio offers fewer voice options and lacks emotion tags or custom voice cloning compared to specialized audio platforms. | 中 | SR002 |
| CR006 | Deepgram mandates individual binding arbitration and class-action waivers for resolving customer disputes under California governing law. | 中 | SR006 |
| CR007 | Healthcare compliance deployments require Business Associate Agreements and data protection safeguards to prevent costly protected health information breach liabilities. | 中 | SR007, SR008 |
| CR008 | Deepgram maintains zero-retention default processing and configurable real-time redaction parameters for PCI and SSN data to meet SOC 2, HIPAA, and GDPR standards. | 中 | SR007 |
| CR009 | Enterprise customers handling sensitive biometric audio can deploy Deepgram in self-hosted or VPC environments to keep voice data within internal network boundaries. | 中 | SR001, SR005 |
| CR010 | Deepgram completed a $130 million Series C financing at a $1.3 billion valuation in January 2026, establishing high valuation multiples relative to historical revenue. | 中 | SR004, SR005 |
| CR011 | Third-party estimates suggest Deepgram reached $100 million in annual recurring revenue in August 2026, though unaudited private figures leave contract durability unverified. | 中 | SR005 |
| CR012 | Enterprise voice infrastructure remains vulnerable to cloud hosting cluster interruptions and GPU capacity bottlenecks across regional availability zones. | 中 | SR001, SR007 |
| CR013 | Customer concentration among conversational AI platforms and contact center partners exposes Deepgram to downstream churn if clients vertically integrate speech models. | 中 | SR003, SR005 |
| CR014 | Unverified private financials, gross margin trajectories, and net revenue retention rates represent essential ongoing governance diligence checks for 2026 investors. | 中 | SR004, SR005 |
| CV001 | Deepgram raised $130 million in Series C funding at a $1.3 billion post-money valuation led by AVP in January 2026. | 高 | SV001, SV002 |
| CV002 | Deepgram equips more than 1,300 organizations and 200,000 developers with autonomous voice agent and speech API infrastructure. | 高 | SV001, SV002 |
| CV003 | Deepgram has publicly disclosed no verified total funding figure and third-party databases report conflicting historical funding totals between $215 million and $229 million. | 中 | SV005 |
| CV004 | Multiples.vc reports Deepgram's recorded revenue as $240 million at Series C, implying a 5.4x revenue multiple against its $1.3 billion valuation. | 中 | SV004 |
| CV005 | Sacra independently estimates that Deepgram reached $100 million in annual recurring revenue in August 2026 driven by expansion across its Voice Agent API stack. | 中 | SV006 |
| CV006 | Silicon Valley Investclub tracks Deepgram's prior calendar 2024 revenue at $21.8 million, highlighting substantial uncertainty and wide variance across unverified public financial benchmarks. | 中 | SV005 |
| CV007 | Alongside its Series C financing, Deepgram announced the acquisition of restaurant voice ordering startup OfOne and the launch of its Powered by Deepgram initiative. | 中 | SV001, SV007 |
| CV008 | Deepgram's Series C valuation of $1.3 billion reflects a 3.4x step-up over its November 2022 Series B valuation of $379 million. | 中 | SV004 |
| CV009 | Deepgram holds independent SOC 2 Type 1 and Type 2 certifications and maintains regional endpoints for HIPAA and GDPR compliance. | 中 | SV009 |
| CV010 | Deepgram's terms of service prohibit customers from accessing services or outputs for competitive model training, benchmarking, or developing competing services. | 中 | SV010 |
| CV011 | In a bear scenario involving price compression from frontier model labs and ARR slowing to $50 million, Deepgram faces down-round valuation repricing toward $400 million. | 中 | SV004, SV005 |
| CV012 | Major institutional investors participating in Deepgram's Series C round include Alkeon, Madrona, Tiger Global, Wing, and strategic corporate backers ServiceNow Ventures, Twilio, and SAP. | 高 | SV007, SV008 |
| CV013 | Deepgram extended its Series C in August 2026 with strategic investment from Singapore EDBI to establish its regional Asia-Pacific headquarters. | 高 | SV003, SV004 |
| CV014 | Institutional underwriting of Deepgram requires primary verification of cap table liquidation preferences, customer concentration, and GAAP gross margin percentages. | 中 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | Deepgram | Deepgram Raises $130M Series C at $1.3B Valuation to Power the Voice AI Economy | Deepgram Raises $130M Series C at $1.3B Valuation to Power the Voice AI Economy |
| SO002 | Deepgram | About Us | Voice AI | STT & TTS - Deepgram | Founded in 2015, Deepgram started with machine learning research for waveform analysis in a dark matter detector in China. CEO and co-founder Scott Stephenson and his teammate later explored deep learning for audio analysis at the University of Michigan. Seeing a gap in the speech-to-text market, they built Deepgram using end-to-end deep learning. |
| SO003 | Toolradar | Deepgram Pricing 2026: Plans, Hidden Costs & Alternatives | Main caveat: the older Enhanced and Base models are more expensive than Nova while being less accurate, always use Nova-3 or Flux unless you have a specific legacy reason. |
| SO004 | Sacra | Deepgram revenue, funding & news | Sacra | Deepgram was valued at $1.3B following its $130M Series C round led by AVP in January 2026. The round included participation from existing investors Alkeon, In-Q-Tel, Madrona, Tiger Global, Wing, Y Combinator, and BlackRock, alongside new investors including Alumni Ventures, Princeville Capital, Twilio, ServiceNow Ventures, SAP, and Citi Ventures. |
| SO005 | Silicon Valley Investclub | Deepgram — $1.3B · Scott Stephenson, Adam Sypniewski | Founded in 2015 by Scott Stephenson, Adam Sypniewski, and Noah Shutty—all former physicists from the University of Michigan—Deepgram started with machine learning research for waveform analysis in a dark matter detector. |
| SO006 | Business Wire | Deepgram Raises $130M Series C at $1.3B Valuation to Power the Voice AI Economy | SAN FRANCISCO--(BUSINESS WIRE)--Deepgram, the real-time API platform underpinning the Voice AI economy, today announced it has raised $130 million in Series C funding at a $1.3 billion valuation. The round was led by AVP, an independent global investment platform dedicated to high-growth technology companies across Europe and North America. |
| SO007 | Deepgram | Terms of Service - Deepgram | These Terms are an agreement between Deepgram, Inc. (“Deepgram”, “we,” “our,” or “us”) and you. By accessing our Services, you agree to these Terms. |
| SO008 | Deepgram | Standard Compliance speech-to-text: HIPAA, SOC 2, GDPR Production Guide | Master standard compliance speech-to-text implementation for HIPAA, SOC 2, and GDPR. Production-tested deployment patterns, API configuration, and testing protocols. |
| SM001 | Crowdfund Insider | Deepgram Closes $130M C Round, Acquires OfOne - Crowdfund Insider | Much like Stripe delivered the API platform underpinning the payments economy, we believe Deepgram is poised to deliver the API platform underpinning the emerging trillion-dollar B2B Voice AI economy |
| SM002 | Tech Funding News | Voice AI Deepgram hits unicorn status with $130M raise led by AVP | With live conversations requiring real-time human interaction, enterprises still struggle with unreliable speech recognition due to high latency, inaccuracies, interruptions, and scalability issues. |
| SM003 | Deepgram | Enterprise Voice AI: STT, TTS & Agent APIs | Deepgram | The first conversation-aware speech platform built to natively understand turn-taking, handle interruptions, and carry conversational context. Built for voice agents. Available in real time or batch, in the cloud or self-hosted. |
| SM004 | Sacra | Deepgram revenue, funding & news | Sacra | The global speech-to-text API market alone is projected to reach $8.3 billion by 2030 with a 14.1% CAGR. By expanding into text-to-speech with Aura, Deepgram taps into an additional market expected to hit $10 billion by 2029, growing at 19.1% annually. |
| SM005 | DIY AI | Deepgram Review 2026: Accuracy, Features, Voice Agents and ... | The short version: Deepgram is one of the strongest speech-to-text APIs for real-time transcription and voice AI. It is not the simplest option for occasional file transcription, nor is it always the best fit for analytics-first buyers. |
| SM006 | Deepgram | Terms of Service - Deepgram | Pay As You Go: This is a consumption-based plan. When your Credit balance drops below a specific threshold, you will be charged the Subscription Fee and your account will be topped off with credits. |
| SM007 | Deepgram | Standard Compliance speech-to-text: HIPAA, SOC 2, GDPR Production ... | Master standard compliance speech-to-text implementation for HIPAA, SOC 2, and GDPR. Production-tested deployment patterns, API configuration, and testing protocols. |
| SM008 | Deepgram | Speech-to-Text Privacy: Enterprise Security, Compliance and Data ... | Healthcare data breaches cost $10.93M per incident. Learn encryption, HIPAA compliance, and privacy-by-design architecture for secure speech-to-text deployment. |
| SM009 | Business Wire | Deepgram Raises $130M Series C at $1.3B Valuation to Power ... - Business Wire | As we rapidly approach a world where billions of simultaneous conversations are powered by Voice AI, enterprises and developers need real-time, reliable infrastructure capable of fully duplex, contextual conversations at scale |
| SP001 | AI Wiki | Deepgram - AI Wiki | The speech AI market in 2026 contains a small number of model labs that train and operate their own production speech systems, alongside the hyperscale cloud providers that offer commodity ASR APIs as part of broader platforms. Deepgram is most directly compared to AssemblyAI, Speechmatics, and the speech APIs offered by Google, Amazon, and Microsoft on the transcription side, and to ElevenLabs and OpenAI on the synthesis side. |
| SP002 | Business Wire | Deepgram Launches Voice Agent API | Voice Agent API Is Industry's Only Offering That Delivers The Single, Real-Time API Experience Developers Love, Combined with Full Controllability Enterprises Need. |
| SP003 | Deepgram | Deepgram Voice AI Platform | Voice AI built for real conversations. |
| SP004 | AI Central Resources | Deepgram Profile and Capabilities | Deepgram is a foundational voice AI company providing developer-first APIs for speech-to-text transcription, text-to-speech synthesis, and real-time conversational voice agents. |
| SP005 | The Economic Times | Voice AI Startups and Deepgram Topic Coverage | Voice AI startup Deepgram has secured $130 million in funding, valuing the company at $1.3 billion. This investment will fuel international expansion into Europe and Asia-Pacific. Deepgram also plans to develop new AI models and pursue strategic acquisitions. |
| SP006 | Deepgram | Deepgram Terms of Service | Welcome to Deepgram! These Terms of Service (“Terms”) apply to our website and the products and services described on our website (collectively, our “Services”), including our Voice AI APIs for speech to text, text to speech, voice agents, and Audio Intelligence |
| SP007 | Deepgram | Master Standard Compliance Speech-to-Text Implementation for HIPAA, SOC 2, and GDPR | Master standard compliance speech-to-text implementation for HIPAA, SOC 2, and GDPR. Production-tested deployment patterns, API configuration, and testing protocols. |
| SP008 | Deepgram | Speech-to-Text Privacy and Healthcare Data Security Architecture | Healthcare data breaches cost $10.93M per incident. Learn encryption, HIPAA compliance, and privacy-by-design architecture for secure speech-to-text deployment. |
| SP009 | Sacra | Deepgram Revenue, ARR and Competition Analysis | Deepgram operates in a market that includes several distinct categories of speech AI competitors, ranging from tech giants to specialized startups and open-source alternatives. |
| SP010 | Silicon Valley Investment Club | Deepgram Company Profile and Competitor Tracking | Voice AI platform powering real-time speech-to-text, text-to-speech, and voice agent APIs for developers and enterprises worldwide |
| SI001 | TechStartups | Voice AI startup Deepgram raises $130M at $1.3B valuation to fuel global expansion | The San Francisco-based startup raised $130 million in a Series C round, valuing the company at $1.3 billion, pushing it firmly into unicorn territory. |
| SI002 | AVP | Deepgram raises $130M Series C at $1.3B valuation to power the voice AI economy | Funding accelerates Deepgram’s API platform for real-time Voice AI and supports the launch of ‘Powered by Deepgram,’ the acquisition of OfOne, an expanded patent portfolio, and a new Voice AI Collaboration Hub in San Francisco |
| SI003 | Deepgram | Enterprise Voice AI: STT, TTS & Agent APIs | Instead of stitching together separate components, Deepgram unifies speech-to-text, text-to-speech, and LLM orchestration into a single API, reducing complexity, latency, and cost. |
| SI004 | Sacra | Deepgram revenue, funding & news | Sacra estimates that Deepgram hit $100M in annual recurring revenue (ARR) in August 2026, with the company announcing the milestone alongside the launch of its Flux TTS model. Deepgram entered 2025 growing ARR 115% year-over-year as voice AI shifted from transcription toward real-time agents. |
| SI005 | Business Wire | Deepgram Raises $130M Series C at $1.3B Valuation to Power the Voice AI Economy | Deepgram, the real-time API platform underpinning the Voice AI economy, today announced it has raised $130 million in Series C funding at a $1.3 billion valuation. The round was led by AVP, an independent global investment platform dedicated to high-growth technology companies across Europe and North America. |
| SI006 | Deepgram | Terms of Service - Deepgram | Pay As You Go: This is a consumption-based plan. When your Credit balance drops below a specific threshold, you will be charged the Subscription Fee and your account will be topped off with credits. |
| SI007 | Deepgram | Standard Compliance speech-to-text: HIPAA, SOC 2, GDPR Production | Master standard compliance speech-to-text implementation for HIPAA, SOC 2, and GDPR. Production-tested deployment patterns, API configuration, and testing protocols. |
| SI008 | Deepgram | Speech-to-Text Privacy: Enterprise Security, Compliance and Data | Healthcare data breaches cost $10.93M per incident. Learn encryption, HIPAA compliance, and privacy-by-design architecture for secure speech-to-text deployment. |
| SI009 | Silicon Valley Investment Club | Deepgram — $1.3B · Scott Stephenson, Adam Sypniewski | The provided results do not verify a revenue figure or profitability statement for Deepgram. |
| SE001 | Gladia | Deepgram review 2026: Is this voice AI platform right for you? | To create this Deepgram review, the platform has been analyzed extensively. Based on the findings, Deepgram is the ideal choice if:...You need a unified platform combining speech-to-text, text-to-speech, and voice agents...You want custom model training for industry-specific vocabulary...You require flexible deployment options, including on-premise...You value having conversational AI capabilities in one integrated stack |
| SE002 | HappyRobot | Deepgram Pricing Explained: Plans, Rates & Voice Agent ... | Deepgram pricing starts at $0.0048 per minute for Nova-3 Monolingual streaming, with separate rates for multilingual transcription, batch processing, text-to-speech, and voice agents. |
| SE003 | Deepgram | Enterprise Voice AI: STT, TTS & Agent APIs | Deepgram | Flux STT listens. Flux TTS speaks. The first conversation-aware speech platform built to natively understand turn-taking, handle interruptions, and carry conversational context. Built for voice agents. Available in real time or batch, in the cloud or self-hosted. |
| SE004 | Voice AI Space | Deepgram — Voice AI Platform | Deepgram provides a comprehensive suite of voice AI technologies, including high-accuracy speech-to-text, natural-sounding text-to-speech, and advanced conversational agent APIs, enabling developers and enterprises to build, scale, and deploy voice-first applications. The platform is designed for real-time and batch processing, supports over 36 languages, and offers both cloud and on-premises deployment options. |
| SE005 | Silicon Valley Investment Club | Deepgram — $1.3B · Scott Stephenson, Adam Sypniewski | Voice AI platform powering real-time speech-to-text, text-to-speech, and voice agent APIs for developers and enterprises worldwide |
| SE006 | Deepgram | Terms of Service - Deepgram | These Terms of Service (“Terms”) apply to our website and the products and services described on our website (collectively, our “Services”), including our Voice AI APIs for speech to text, text to speech, voice agents, and Audio Intelligence, and any software you install on your devices (“Software”), such as the Saga voice assistant. |
| SE007 | Deepgram | Standard Compliance speech-to-text: HIPAA, SOC 2, GDPR Production ... | Master standard compliance speech-to-text implementation for HIPAA, SOC 2, and GDPR. Production-tested deployment patterns, API configuration, and testing protocols. |
| SE008 | Deepgram | Speech-to-Text Privacy: Enterprise Security, Compliance and Data ... | Healthcare data breaches cost $10.93M per incident. Learn encryption, HIPAA compliance, and privacy-by-design architecture for secure speech-to-text deployment. |
| SU001 | IBM Newsroom | Deepgram and IBM Introduce Advanced Voice Capabilities for Enterprise AI | ARMONK, N.Y. and SAN FRANCISCO — February 24, 2026 — IBM (NYSE: IBM) and Deepgram today announced a collaboration to integrate Deepgram’s industry-leading speech-to-text and text-to-speech capabilities into IBM’s watsonx Orchestrate generative AI solution. |
| SU002 | Tooliverse | Deepgram Review 2026 - Voice AI API Platform | Deepgram requires a learning curve to master advanced features like diarization and formatting, a challenge reported in 64 user reviews. |
| SU003 | Deepgram | Enterprise Voice AI: STT, TTS & Agent APIs | Deepgram | Sierra helps businesses build better, more human customer experiences with AI. Our agents handle everything from originating a mortgage to preventing churn and authenticating patients for access to care — and we support some of the world’s largest brands. Deepgram is a key component of Sierra’s voice stack, and the voice capabilities that are powered by Deepgram APIs help us deliver enterprise-grade accuracy and reliability. |
| SU004 | Silicon Valley Invest Club | Deepgram — $1.3B · Scott Stephenson, Adam Sypniewski | The company went through Y Combinator in Winter 2016 and has since grown to serve over 1,300 organizations including NASA, Spotify, Twilio, and Citibank. |
| SU005 | Sacra | Deepgram revenue, funding & news | Usage grew 3.3x annually for four consecutive years through 2024, when Deepgram had 400+ enterprise customers, 200,000+ developers, and 85%+ gross margins. By January 2026, more than 1,300 organizations were building on Deepgram, with the platform having processed over 50,000 years of audio and transcribed more than 1 trillion words. |
| SU006 | Deepgram | Terms of Service | Deepgram | Pay As You Go: This is a consumption-based plan. When your Credit balance drops below a specific threshold, you will be charged the Subscription Fee and your account will be topped off with credits. |
| SU007 | Deepgram | Standard Compliance speech-to-text: HIPAA, SOC 2, GDPR Production Guide | Master standard compliance speech-to-text implementation for HIPAA, SOC 2, and GDPR. Production-tested deployment patterns, API configuration, and testing protocols. |
| SU008 | Deepgram | Speech-to-Text Privacy: Enterprise Security, Compliance and Data Protection | Healthcare data breaches cost $10.93M per incident. Learn encryption, HIPAA compliance, and privacy-by-design architecture for secure speech-to-text deployment. |
| SR001 | scored.tools | Deepgram Review 2026: Voice AI Infrastructure That Actually Ships | Free tier is credit-limited, not permanent. Costs ramp fast at scale — model your bill before you scale... Voice Agent API is newer. It works, but it hasn't been through as many production edge cases as the core STT product. Expect to hit rough edges around interruption handling in complex scenarios. |
| SR002 | Fish Audio | Deepgram Review — Are They Legit? (2026) | Speech Recognition & TTS API Platform | Fewer voice options than dedicated TTS platforms... Better known for transcription than synthesis... Does not support emotion instructions... Limited languages available... Does not offer voice cloning |
| SR003 | Deepgram, Inc. | Enterprise Voice AI: STT, TTS & Agent APIs | Instead of stitching together separate components, Deepgram unifies speech-to-text, text-to-speech, and LLM orchestration into a single API, reducing complexity, latency, and cost. |
| SR004 | Silicon Valley Investclub | Deepgram — $1.3B · Scott Stephenson, Adam Sypniewski | In January 2026, Deepgram raised $130 million in Series C funding at a $1.3 billion valuation, becoming the newest unicorn in the Voice AI space. |
| SR005 | Sacra | Deepgram revenue, funding & news | Sacra estimates that Deepgram hit $100M in annual recurring revenue (ARR) in August 2026, with the company announcing the milestone alongside the launch of its Flux TTS model. |
| SR006 | Deepgram, Inc. | Terms of Service | Except to the extent a restriction is prohibited by law, you agree not to do... reverse engineer, decompile, disassemble, decode, attempt to gain access to, or copy the source code, underlying algorithms, or model weights... use or access our Services or any Output for competitive purposes, including model training, benchmarking and other competitive analysis |
| SR007 | Deepgram, Inc. | Standard Compliance speech-to-text: HIPAA, SOC 2, GDPR Production Guide | Master standard compliance speech-to-text implementation for HIPAA, SOC 2, and GDPR. Production-tested deployment patterns, API configuration, and testing protocols. |
| SR008 | Deepgram, Inc. | Speech-to-Text Privacy: Enterprise Security, Compliance and Data Protection | Healthcare data breaches cost $10.93M per incident. Learn encryption, HIPAA compliance, and privacy-by-design architecture for secure speech-to-text deployment. |
| SV001 | Built In San Francisco | Deepgram Raises $130M Series C at $1.3B Valuation | Deepgram, an API platform offering voice AI capabilities like speech-to-text and text-to-speech, recently raised a $130 million round of Series C funding. The raise, led by AVP, values the company at $1.3 billion. |
| SV002 | Business Wire | Deepgram Raises $130M Series C at $1.3B Valuation to Power the Voice AI Economy | Deepgram, the real-time API platform underpinning the Voice AI economy, today announced it has raised $130 million in Series C funding at a $1.3 billion valuation. The round was led by AVP, an independent global investment platform dedicated to high-growth technology companies across Europe and North America. |
| SV003 | The Economic Times | Deepgram News and Topic Coverage | Voice AI startup Deepgram has secured $130 million in funding, valuing the company at $1.3 billion. This investment will fuel international expansion into Europe and Asia-Pacific. |
| SV004 | Multiples.vc | Deepgram Private Valuation and Comps | Deepgram's latest recorded revenue is $240M, as of its Series C in January 2026. At a $1.3B valuation, that implies a 5.4x revenue multiple. |
| SV005 | Silicon Valley Investclub | Deepgram Company Profile and Valuation Analysis | Deepgram has publicly disclosed no verified total funding figure in the provided results, and the reported figures conflict. |
| SV006 | Sacra | Deepgram Revenue and Business Model Teardown | Sacra estimates that Deepgram hit $100M in annual recurring revenue (ARR) in August 2026, with the company announcing the milestone alongside the launch of its Flux TTS model. |
| SV007 | The SaaS News | Deepgram Raises $130 Million Series C | Deepgram, a San Francisco, CA-based real-time API platform powering the Voice AI economy, has raised $130 million in a Series C round at a $1.3 billion valuation. |
| SV008 | Yahoo Finance | Deepgram Raises $130M Series C at $1.3B Valuation | This is a paid press release. Contact the press release distributor directly with any inquiries. |
| SV009 | Deepgram Developer Documentation | Data Privacy and Compliance Frameworks | Deepgram has achieved SOC 2 Type 1 and Type 2 certification. An independent auditor has evaluated the security controls and procedures we use to protect the data we process in the cloud and has assessed the operational effectiveness of our systems. |
| SV010 | Deepgram Official | Deepgram Terms of Service | use or access our Services or any Output for competitive purposes, including model training, benchmarking and other competitive analysis, or developing competing models, products or services; |