Anysphere (Cursor)

1.1 Identity and Business Model

Anysphere, Inc. operates under the product brand Cursor. The company is incorporated in Delaware and headquartered in San Francisco, California. Founded in 2022 by Michael Truell (CEO), Sualeh Asif (CPO), Aman Sanger (COO), and Arvid Lunnemark (co-founder; departed as CTO in October 2025), all four were students at MIT when they incorporated the business. The core product is Cursor, an AI-native integrated development environment (IDE) that forks Microsoft's open-source Visual Studio Code and layers in generative AI capabilities including multi-agent execution, codebase-wide semantic search, and a proprietary model called Composer. The business model is primarily recurring subscription (SaaS): Individual plans at $0/month (Hobby), $20/month (Pro), $60/month (Pro+), and $200/month (Ultra); team plans at $40/user/month (Business); and custom enterprise contracts. Cursor competes in the AI developer tools segment against GitHub Copilot (Microsoft), Windsurf/Codeium (acquired by OpenAI), Amazon Q Developer, and Claude Code (Anthropic). The product-led growth model relies almost entirely on word-of-mouth and viral developer adoption; marketing spend is reportedly near zero. [CO001][CO002][CO003]

1.2 Founders, Leadership, and Governance

The founding team consists of four MIT students who incorporated Anysphere in 2022 and built an early prototype before raising their first external capital in October 2023. Michael Truell (CEO) leads strategy and product vision; Sualeh Asif (CPO) leads product; Aman Sanger (COO) leads operations and go-to-market. Arvid Lunnemark co-founded the company and served as CTO until October 2025, when he left to found Integrous Research, a safety-focused AI research lab. His departure is a noted key-person risk. Post-Series D (November 2025), each original co-founder held approximately 4.5% of the company, making all four billionaires on paper at the $29.3B valuation. Board composition is not publicly disclosed; Andreessen Horowitz (Series A), Thrive Capital (Series C lead), and Accel/Coatue (Series D leads) are presumed board observers or directors based on their investment terms. Strategic investors Google and Nvidia joined in the Series D. In July 2025, Anysphere hired Travis McPeak (co-founder of Resourcely) to lead security. The company employs roughly 150 people as of August 2025, making it extraordinarily capital- efficient for its revenue scale. [CO004][CO005][CO006][CO007]

1.3 Funding History and Valuation

Anysphere has raised approximately $3.47 billion in disclosed equity across five rounds in roughly 28 months. The October 2023 seed round of $8M was led by the OpenAI Startup Fund, with notable angels including Nat Friedman (former GitHub CEO) and Arash Ferdowsi (Dropbox co-founder). In August 2024, Andreessen Horowitz led a $60M Series A at a $400M post-money valuation. In November 2024, the company was the subject of a competitive unsolicited bidding war among Benchmark, Index Ventures, and others, reportedly valuing Anysphere at approximately $2.5B. In January 2025, Anysphere closed a $105M Series B at a $2.6B valuation. In June 2025, Thrive Capital led a $900M Series C at a $9.9B post-money valuation—at which point the company had passed $500M ARR. On November 13, 2025, Anysphere closed a $2.3B Series D co-led by Accel and Coatue, valuing the company at $29.3B; Google and Nvidia participated as strategic investors. By early 2026, Bloomberg and The Information reported that Anysphere was in discussions to raise ~$5B at $50–60B valuation. On April 21, 2026, xAI (SpaceX's AI subsidiary) announced it had secured an option to acquire Anysphere for $60B or pay $10B for a collaboration arrangement. Total disclosed equity raised: approximately $3.37B. No debt or project finance is reported. [CO008][CO009][CO010][CO011][CO012]

1.4 Scale, Milestones, and Adverse Events

Cursor is widely described as the fastest SaaS company to reach $100M ARR (reached January 2025, approximately 14–16 months after launch) and $1B ARR (reached November 2025). The company has approximately 1 million paying customers and 40,000+ paying enterprise teams as of early 2026. The Fortune 500 penetration is reported by the company at 64% (company-claimed, not independently audited). Anysphere has made three notable acquisitions: Supermaven (AI code completion, November 2024), Koala (CRM startup acqui-hire, July 2025), and Graphite (code review startup, December 2025, valued above $290M). Key adverse events include: (1) April 2025 — an AI help-desk agent named "Sam" invented a non-existent login policy, causing user cancellations before Anysphere apologized and issued refunds; (2) July 2025 — Cursor changed its Pro plan from 500 requests to usage-metered billing, triggering widespread user complaints; the company apologized, rolled back limits, and refunded affected users; (3) October 2025 — co-founder and CTO Arvid Lunnemark departed to found a safety-focused AI lab. Security vulnerabilities CVE-2025-54135 (CurXecute) and CVE-2025-54136 (MCPoison) disclosed in 2025 exposed critical remote code execution risks. [CO013][CO014][CO015][CO016]

1.5 Exhibits

2.1 Market Definition and Boundary

The relevant market for Cursor is AI-assisted developer tooling — software that embeds large-language-model capabilities directly into the software development workflow to accelerate code authoring, debugging, review, and testing. This market is distinct from: (a) general-purpose AI chatbots (ChatGPT, Gemini) that have no deep editor integration; (b) traditional IDEs (VS Code, JetBrains) without AI features; (c) DevOps/CI/CD platforms (GitHub Actions, CircleCI); and (d) application monitoring or observability tools. Adjacent spend that flows into the same budget but is excluded from the core definition includes AI code-review automation (e.g., standalone PR review bots), AI test-generation platforms, and AI documentation generators. Within AI developer tooling, the market segments by delivery model. The two main categories are IDE-native AI coding tools — full-featured environments built from the ground up around AI (Cursor, Windsurf) — and editor extensions/plugins that bolt onto an existing IDE (GitHub Copilot, Tabnine, Codeium, Amazon Q Developer). Cursor competes primarily in the IDE-native category, where it built the product as a VS Code fork; it also captures cross-category buyers who switch to it from extension-based tools. The primary buyer unit is the individual software developer (direct B2C) or the engineering team (B2B enterprise). Revenue can flow from direct individual subscriptions (PLG self-serve), team plans billed to a department head or eng leader, or enterprise agreements managed by IT or procurement. In regulated industries — finance, healthcare, defense — additional approval layers are added, often with data-residency or on-premise requirements.

2.2 Market Sizing — TAM, SAM, and SOM

Estimating the TAM for AI developer tools is complicated by definitional variation and rapid market evolution. Market research firms publish figures that range from $2.6B (narrow, coding-assistant-only, 2024) to $34B+ (broad, encompassing all AI-enhanced software development lifecycle tools, 2034). Three distinct lenses produce a reasonable range: **Bottom-up lens (developer count × ARPU):** There are an estimated 27–29 million professional software developers worldwide (Statista 2024), with roughly 100 million registered on GitHub (Octoverse 2024). Stack Overflow's 2024 survey found that 62% of professional developers already use AI tools, up from 44% in 2023. At blended ARPU of $15-$25/user/month for the portion converting to paid — and assuming 30–50% paid conversion among current adopters — the serviceable revenue is roughly $4–8B per year at today's adoption levels, scaling to $12–18B by 2028 as adoption reaches 70–80%. **Top-down lens (market research):** Third-party estimates compiled in the ai2.work November 2024 analysis place the AI developer tools market at approximately $12.8B by 2026, growing at a 27% CAGR. Grand View Research valued the AI coding assistant sub-market at $13.22B by 2031. Allied Market Research cited a broader $34.1B figure by 2034 including adjacent DevSecOps AI tools. **Comparables lens (incumbent revenues):** GitHub Copilot — the market's current revenue leader — reached approximately 1.8 million paid subscribers and 77,000 enterprise customers in FY2024 (Microsoft 2024 Annual Report), a 180% YoY enterprise gain. Cursor surpassed $2B ARR in February 2026. Together these two players likely account for $3.5–4B+ of the market's current annual revenue, confirming the market has already reached multi-billion scale. The SAM for Cursor — AI coding assistants and AI-native IDEs specifically, excluding broader DevSecOps and infrastructure AI — is estimated at $5–10B by 2026, growing to $15–20B by 2029 at a 27–30% CAGR. Cursor's SOM, defined as the market accessible given its current go-to-market (primarily individual developers and engineering teams at tech-forward companies globally), is estimated at $3–5B in near-term, with Cursor already achieving $2B ARR — a significant SOM capture.

2.3 Buyer Segmentation and Adoption Drivers

The AI coding tool buyer landscape spans four primary segments, with distinct decision dynamics in each. **Individual developers (self-serve PLG)** represent Cursor's founding cohort and remain its growth engine. These buyers self-select based on word-of-mouth and viral communities (Twitter/X, Reddit, Hacker News). They pay $20–$200/month personally or expense it without enterprise approval. Stack Overflow's 2024 survey found 81% identify increased productivity as their top motivation for adopting AI tools, and 72% hold a favorable view of AI tools overall (though down from 77% in 2023, flagging some disappointment with early experiences). GitHub's survey data confirms that 62% of professional developers now use AI tools in their workflow, up from 44% in 2023. **Engineering teams / SMBs (Teams plan)** are the next adoption layer. Team leads and CTOs at startups and mid-size companies centralize billing and enforce standardization. This segment drives the bulk of Cursor's 40,000+ enterprise teams. The decision-maker is typically an engineering VP or CTO with discretionary budget, often $40–$100K/year for a team. **Large enterprise (IT-mediated)** is the high-growth segment for the next 2–3 years. Fortune 500 companies face additional procurement hurdles: information security review, data privacy compliance (GDPR/CCPA), software license management, and often executive approval above $100K. Cursor's SOC 2 Type II certification is specifically targeted at this segment. Microsoft's 2024 Annual Report notes that employees at nearly 60% of Fortune 500 companies use Copilot in some form, indicating the enterprise category is already normalized around AI tools. **Regulated industries (FinServ, Healthcare, Gov)** are an emerging, high-value segment with significant friction. These buyers require data-residency guarantees, on-premise deployment options, or air-gapped environments. HIPAA-covered entities in healthcare and GLBA-covered firms in financial services face strict code data handling requirements; the existence of robust vendor security programs (SOC 2, FedRAMP for gov) is a threshold requirement. A key structural feature of the market is the **PLG → enterprise motion**: Cursor and GitHub Copilot both rely on individual developers falling in love with the product before it surfaces as an enterprise procurement conversation. This bottom-up motion dramatically reduces CAC versus traditional enterprise sales.

2.4 Growth Drivers, Constraints, and Adverse Signals

**Growth drivers** are well-documented. First, the productivity evidence is compelling: GitHub's 2023 controlled experiment found developers using Copilot completed coding tasks 55% faster; Microsoft's 2024 Annual Report cites productivity gains of up to 55%. Second, adoption is accelerating on its own momentum — 62% of developers already use AI tools (SO 2024), creating social pressure on holdouts. Third, enterprise willingness to pay is being established rapidly: GitHub Copilot at $19–$39/user/month (Business/Enterprise tiers) has proven the enterprise budget exists; Cursor's Teams plan at $40/user/month is slightly above Copilot's pricing, which signals increasing pricing power as products differentiate. Fourth, model quality improvements (from GPT-4 to Claude 3.5/3.7 to Gemini 2.0) translate directly into product improvement without Cursor needing to invest in model training. Fifth, the shift to multi-agent workflows (up to 8 parallel coding agents in Cursor) is fundamentally changing the value proposition from "autocomplete help" to "autonomous software delivery," which justifies dramatically higher pricing. **Market constraints** create friction against the growth narrative. Accuracy and trust are the primary headwinds: Stack Overflow 2024 found that 31% of developers are skeptical of AI accuracy, and developer favorability fell from 77% to 72% YoY, likely reflecting disappointed expectations from early adopters. Data privacy concerns — especially around code leaving corporate infrastructure — create enterprise sales friction. Copyright liability risk is non-trivial: the GitHub Copilot class action lawsuits (still unresolved as of 2025) allege that AI models trained on open-source code produce infringing outputs; similar exposure could affect Cursor. In regulated industries (healthcare, finance, defense), procurement cycles extend to 6–18 months, limiting near-term SAM capture. Finally, market consolidation pressure is high: OpenAI's ~$3B acquisition of Windsurf/Codeium (reported May 2025) signals that hyperscalers are willing to acquire at scale, and Microsoft/GitHub already bundles Copilot into existing enterprise agreements, creating a bundling disadvantage for standalone tools like Cursor.

2.5 Exhibits

3.1 Competitive Landscape Overview

Cursor operates in the fast-moving AI developer tools category against a multi-tier set of competitors: direct IDE-native peers (Windsurf/Codeium, soon under OpenAI ownership), the dominant incumbent plugin (GitHub Copilot, backed by Microsoft), hyperscaler-native tools (Amazon Q Developer, Google Duet AI), enterprise-privacy specialists (Tabnine), and IDE-bundled assistants (JetBrains AI). Two structural forces define competition: (1) whether a tool is a standalone IDE replacement or a plugin that bolts onto existing editors; and (2) whether the vendor owns its underlying AI model or relies on third-party APIs. Cursor is plugin-free (standalone VS Code fork) and currently model-agnostic (relying on Anthropic Claude, OpenAI GPT-4, and others via API), differentiating on UX depth and multi-agent orchestration rather than model exclusivity. The market is experiencing rapid consolidation: OpenAI's announced ~$3 billion acquisition of Windsurf (formerly Codeium) in May 2025 [CP001] demonstrates that hyperscalers view AI coding tools as strategic moats worth vertical integration. Cursor's lack of a proprietary model is therefore an evolving competitive risk as peers gain tighter model-tool integration.

3.2 Competitor Profiles and Positioning

GitHub Copilot is the market incumbent with the largest installed base: 1.8 million paid subscribers and 77,000 enterprise customers as of fiscal year 2024, implying an estimated $300–350 million in ARR based on subscriber counts and blended ARPU [CP002]. Copilot is distributed as a plugin across VS Code, JetBrains, Neovim, and other editors, leveraging Microsoft's enterprise sales force and Azure bundling. Its key advantages are brand trust, existing enterprise procurement channels, and deep GitHub integration; its structural weakness is that plugin architecture imposes context limitations that a standalone IDE like Cursor can escape. GitHub itself has begun imposing agentic session limits in 2025 as parallelized AI workflows overwhelm compute infrastructure—a sign that plugin architecture struggles to scale agentic use cases [CP008]. Windsurf (formerly Codeium, rebranded 2024) is Cursor's most direct peer—a VS Code-fork standalone IDE targeting the same professional developer segment with an agentic core model called Cascade. OpenAI agreed to acquire Windsurf for approximately $3 billion in May 2025, signaling OpenAI's intent to compete at the IDE layer through vertical integration of model and tool [CP001]. Post-acquisition, Windsurf would gain exclusive access to OpenAI's frontier models and potential integration with ChatGPT's 300 million+ user base, creating a distribution advantage Cursor lacks [CP006]. Amazon Q Developer (formerly CodeWhisperer) targets AWS-native developers as a plugin at $19/user/month for its Pro tier with a free tier for individuals [CP004]. Q Developer's differentiation lies in deep AWS service awareness (code suggestions for S3, Lambda, EC2, etc.) and Amazon's enterprise distribution. Its limitation is that it is primarily an AWS ecosystem play, limiting appeal to multi-cloud or cloud-agnostic enterprises. Tabnine competes on enterprise privacy with cloud, on-premises, and air-gapped deployment options, positioning itself for regulated industries where cloud-only tools like Cursor are blocked by data governance requirements [CP010]. JetBrains AI is bundled with JetBrains subscriptions across its IDE portfolio (IntelliJ, PyCharm, WebStorm, etc.), reaching an installed base of approximately 10 million+ developers without a separate license requirement [CP011]. Replit is an adjacent competitor targeting browser-based cloud development with AI-native tooling, predominantly reaching students and rapid prototypers rather than professional enterprise developers [CP012].

3.3 Moat Durability, Switching Costs, and Competitive Risk

Cursor's moat rests on three reinforcing factors: (1) workflow integration depth — developers who embed Cursor agents, Rules files, and custom MCP servers into daily workflows face meaningful relearning costs when switching, even if the underlying model is portable; (2) codebase indexing — Cursor's proprietary codebase index provides context-aware completions that improve with project history, creating a soft data lock-in; and (3) network effects via team collaboration features — shared rules and audit trails create organizational stickiness as teams standardize on Cursor. Against these moats, the primary risks are multi-homing and model commoditization. Surveys indicate developers commonly use two or more AI tools simultaneously, with Cursor often used for complex tasks while GitHub Copilot remains active in their editor [CP020]. This multi-homing behavior limits Cursor's net revenue retention ceiling. The most significant structural risk is model access concentration: Cursor relies on Anthropic, OpenAI, Google, and xAI APIs for model inference, and while this provides flexibility, it also means competitors like Windsurf (post-OpenAI acquisition) could receive preferential access or pricing [CP022]. Cursor's April 2026 acquisition option from xAI (valued at $60B) suggests a possible future infrastructure tie-up, but the strategic direction remains uncertain [CP023]. Switching costs for enterprise customers are moderate-to-high: enterprise teams that configure Role permissions, whitelist repositories, set up Audit Log pipelines, and integrate Cursor into CI/CD workflows face non-trivial migration effort [CP015]. However, enterprise lock-in remains weaker than incumbents with deep platform integration (e.g., GitHub's native code review and CI/CD pipeline integration).

3.4 Exhibits

4.1 Revenue Model and Pricing Architecture

Cursor (Anysphere, Inc.) monetizes through a tiered subscription model targeting individual developers and enterprise teams, supplemented by usage-based overage and API token pricing for heavy agentic workloads. The product is freemium: the Hobby tier is free with limited AI usage (2,000 monthly code completions), designed to drive top-of-funnel developer adoption and conversion. The Pro tier at $20/month unlocks extended agent limits and access to frontier models. Pro+ at $60/month (3× usage on OpenAI, Claude, and Gemini models) serves daily agent users. Ultra at $200/month provides 20× usage vs. Pro, targeting power users running multi-file agentic workflows. Teams at $40/user/month adds enterprise controls — centralized billing, admin dashboards, RBAC, and SAML/OIDC SSO — and is the primary enterprise acquisition vehicle alongside custom Enterprise contracts [CI001]. Cursor also introduced its own AI models: Composer 2 (March 2026) is available at API-accessible token pricing — standard tier at $0.50/M input and $2.50/M output tokens, fast tier at $1.50/M input and $7.50/M output tokens — enabling usage-based revenue that scales with agentic workload complexity rather than fixed subscription limits [CI002]. This billing architecture has shifted from purely per-seat SaaS to a hybrid model where heavier workloads generate proportionally more revenue, reducing the risk of underpricing heavy-usage customers that was flagged in Cursor's July 2025 pricing controversy. As of early 2026, enterprise and corporate buyers account for approximately 60% of total revenue, reflecting a rapid shift from a developer-led self-serve motion to enterprise-led contracts since the Koala acquisition in July 2025 [CI003].

4.2 Revenue Traction, Unit Economics, and Cost Structure

Cursor's revenue trajectory is exceptional by any SaaS benchmark. Starting from $1M ARR in 2023, the company reached $100M ARR in January 2025 (approximately 12 months from first material revenue), $500M ARR in June 2025, $1B ARR in November 2025, and $2B ARR in February 2026 [CI007]. This $1M-to- $2B trajectory took approximately 36 months, with revenue doubling roughly every two months through 2025. SaaStr identified Cursor as the fastest-growing B2B SaaS company from $1M to $500M ARR on record, surpassing prior records set by Wiz, Deel, and Ramp [CI008]. Sacra estimates that Cursor reached slight gross-margin profitability in April 2026 and was forecasting more than $6B in annualized revenue run rate by end-2026 [CI009]. The primary cost of revenue is AI model inference — Cursor routes queries through Anthropic Claude, OpenAI GPT, Google Gemini, and its own Composer models, paying per-token API fees that represent the largest single cost component. For latency-sensitive operations like inline code completion, Cursor uses fine-tuned Mixture-of-Experts (MoE) models to reduce inference cost per call while maintaining quality [CI010]. Cursor launched proprietary models Composer (October 2025) and Composer 2 (March 2026), aiming to reduce dependency on third-party API costs over time and improve gross margins [CI011]. Sales and marketing expenditure has historically been minimal relative to revenue because Cursor's growth is overwhelmingly product- led: the free tier drives organic developer adoption, with conversion to paid tiers occurring inside the product. Enterprise sales is supplemented by a growing direct sales team post-Koala acquisition, but CAC remains relatively low compared to traditional enterprise SaaS, with customer acquisition anchored by word-of-mouth and developer community referrals [CI012]. The headcount of approximately 150 employees (as of early 2026) implies very high revenue per employee — over $13M ARR per employee at $2B ARR — a figure that reflects the capital-light nature of software distribution combined with unusually high revenue growth [CI013]. Infrastructure costs are primarily AWS-based with compute supplemented by the SpaceX/xAI Colossus partnership for model training [CI014].

4.3 Capital Adequacy, Financing History, and Financial Verdict

Cursor has raised approximately $3.47 billion in total venture financing across five rounds: a Seed round of approximately $8M, a Series A of $60M at a $400M valuation, a Series B of $105M at a $2.6B valuation (December 2024), a Series C of $900M at a $9.9B valuation (November 2025), and a Series D of $2.3B at a $29.3B valuation (March 2026) [CI018]. The scale of capital raised far exceeds the company's operational cash needs given its rapid progression to gross-margin profitability — the capital is primarily strategic: building a moat in model partnerships and compute infrastructure, funding the xAI Colossus training partnership, and supporting enterprise sales buildout [CI019]. Monthly cash burn is not publicly disclosed; however, at ~$2B ARR with approaching gross-margin breakeven and approximately $3.47B raised, Cursor's implied runway is multiple years even under pessimistic burn assumptions. The Series D at a $29.3B valuation implies an ~15× forward ARR multiple if 2026 ARR reaches $2B, or ~5× if the $6B+ forecast is met. Key financing risks include: model API cost inflation if Anthropic or OpenAI raises API prices (Cursor has multi-year partnerships but terms are not disclosed), compute cost escalation for Composer model training, and the xAI/SpaceX compute partnership which creates dependency on a politically-sensitive counterparty [CI020][CI021]. A $60B xAI acquisition option received by Cursor in April 2026 adds an unusual strategic optionality dimension to the capital structure but does not constitute committed financing [CI022]. The overall financial verdict is favorable on revenue quality (strong subscription NRR proxy, enterprise pivot, rapid growth), capital adequacy (substantial runway), and trajectory (approaching profitability); the primary financial diligence blockers are gross margin confirmation (not publicly disclosed), model API cost structure (private), and enterprise contract ACV and NRR data (not publicly available) [CI023].

4.4 Exhibits

5.1 Product Architecture and VS Code Foundation

Cursor is a fork of Visual Studio Code (VS Code), built on Electron and deliberately inheriting VS Code's extension ecosystem, Language Server Protocol (LSP) infrastructure, and TextModel object system. This architectural choice dramatically reduces onboarding friction: developers keep their settings, themes, keybindings, and extensions while gaining Cursor's AI layer. The fork is maintained by Anysphere's small but prolific engineering team, which monitors upstream VS Code security patches and merges them based on risk assessment. Central to Cursor's AI architecture is the shadow workspace, introduced in September 2024. The shadow workspace runs as a hidden Electron window that mirrors the user's file system without altering their active coding session, giving background AI agents full LSP access (lints, type errors, go-to-definition) while preserving independence, privacy, concurrency, universality, maintainability, and speed. This enables agents to iterate on code in the background, detect lint errors after each edit, and self-correct before surfacing results. The platform's codebase indexing engine builds vector embeddings over entire repositories — handling millions of lines across hundreds of thousands of files — enabling semantic code search over the indexed codebase. Cursor Rules allow teams to encode project-specific instructions, conventions, and context engineering logic (e.g., NAB's CEL library built on Cursor's rules, skills, and hooks primitives).

5.2 Proprietary AI Models — Cursor Tab and the Composer Series

Unlike competitors that primarily route requests to third-party frontier models, Cursor has built and maintains two classes of proprietary AI: Cursor Tab (completion model) and the Composer series (agentic reasoning model). Cursor Tab is a fine-tuned fill-in-the-middle (FIM) model trained with reinforcement learning to excel at multi-line, multi-file completions. Tab predicts the next edit location after each accepted suggestion, auto-imports unresolved symbols in TypeScript/Python, and suggests complete diffs rather than individual tokens. The Composer series is a mixture-of-experts (MoE) language model series trained at scale using RL in production coding environments. Composer is given access to editing, terminal, grep, and semantic search tools and optimized via a reward model reflecting real software engineering quality. Training infrastructure uses custom MXFP8 kernels with expert parallelism and hybrid sharded data parallelism across thousands of NVIDIA GPUs (PyTorch + Ray). The warp decode kernel (April 2026) further achieves 1.8x faster MoE decode on Blackwell GPUs by reorganizing computation around outputs rather than experts. Composer 2 (March 2026) achieves 61.3 on CursorBench, 61.7 on Terminal-Bench 2.0, and 73.7 on SWE-bench Multilingual — frontier-level scores at $0.50/M input and $2.50/M output tokens (standard) or $1.50/$7.50 (fast variant, 4× faster generation). A companion Cursor Bench internal evaluation based on real-world engineering requests from Cursor engineers sets the quality bar for ongoing RL training runs. Cursor also exposes all major frontier models (Claude Sonnet, GPT-5, Gemini Flash 2.5, Qwen Coder) within the same agent harness for users to choose.

5.3 Agent Infrastructure, Cloud Platform, and MCP Ecosystem

The Cursor agent harness is the central orchestration layer that manages the context window (system prompt + tool descriptions + conversation history + user request) for every AI interaction. The harness is continuously tuned per model: when Anysphere gains early access to a new frontier model, engineers spend weeks customizing prompt structure, context injection, tool descriptions, and guardrails until that model's performance inside the harness exceeds its stock benchmark results. Mid-chat model switching (introduced April 2026) lets developers change models without losing conversation context. Cursor 3 (April 2026) introduces the unified workspace combining local and cloud agents. Cloud agents run on dedicated isolated virtual machines, removing the local resource constraints that cap parallelism. Amplitude runs 1,000+ automated agent runs per week without manual prompting; NAB's engineers kick off multiple agent branches on legacy Assembly code, collecting structured flowcharts and migration reports. As of February 2026, 35% of PRs merged internally at Cursor are created by autonomous cloud agents. Agent usage company-wide has grown approximately 15× over the past year; Cursor now has 2× more agent users than Tab users (reversed from a 2.5:1 Tab-to-agent ratio in March 2025). Cursor Automations enable always-on cron-based and trigger-based cloud agent workflows (Amplitude uses them to convert Slack bug reports into PRs automatically). The Cursor SDK (April 2026, npm install @cursor/sdk) exposes the same agent runtime, harness, and models programmatically via a TypeScript API, opening a developer ecosystem around the platform. MCP (Model Context Protocol) support makes Cursor a first-class integration point for external tools, databases, and workflows. Cursor Marketplace distributes plugins — bundles of MCP servers, skills, subagents, rules, and hooks — with admin-controlled distribution modes (Default Off, Default On, Required) for enterprise governance.

5.4 Security, Compliance, and Enterprise Controls

Cursor maintains a SOC 2 Type II attestation (available via trust.cursor.com) with annual third-party penetration testing and an executive summary available on request. Data is encrypted with AES-256 at rest and TLS 1.2+ in transit. Cursor operates exclusively on AWS infrastructure and explicitly maintains no infrastructure in China, uses no China-headquartered subprocessors, and commits to annual subprocessor reviews under its vendor risk management program. Privacy Mode — enabled by default for all team members — implements Zero Data Retention (ZDR) contractual controls with all model providers, ensuring that code is not stored by providers or used for training. GDPR and CCPA compliance is maintained; users can delete accounts via the Settings dashboard. Enterprise controls include SSO (Okta, Azure AD, Google Workspace) and SCIM for automated user provisioning and deprovisioning; Customer-Managed Encryption Keys (CMEK); compliance logging; MDM deployment; and model-level access controls that let admins allow/blocklist specific model providers, model configurations (speed/context window), or new providers by default. Soft spend limits (rather than hard cutoffs) and automatic alerts at 50/80/100% thresholds balance cost governance with developer productivity. Two CVEs were disclosed in 2025: CVE-2025-54135 and CVE-2025-54136 (CurxExecute/MCPoison) relating to MCP handling in Cursor. Tenable published a detailed FAQ. The disclosures highlight the emerging attack surface from integrating untrusted MCP servers, an area Cursor has since addressed with documented security considerations for MCP and agent security best practices in its developer docs. No on-premises deployment is currently offered, which may limit adoption in certain highly regulated environments.

5.5 Technical Differentiation and Platform Moat

Cursor's competitive differentiation operates at three reinforcing levels. First, proprietary models (Tab and Composer) are trained on real-world coding environments at Anysphere's infrastructure cost — an investment that pure API-routing competitors cannot replicate without building their own training infrastructure. Second, the agent harness compounding effect: each frontier model integration is paired with weeks of harness tuning, yielding measurably better in-product performance than stock model APIs deliver. Third, the data flywheel: with over 1 million paying users generating structured coding tasks and evaluation feedback, Anysphere continuously retrains Composer using RL against real production signals. Enterprise adoption deepens the moat: NAB built internal tooling (NAB CEL) directly on Cursor's context engineering primitives (rules, skills, hooks). Amplitude built autonomous development pipelines on Cursor Automations. These organizational integrations create switching costs analogous to those from ERP or DevOps platform lock-in. The developer ecosystem (JetBrains ACP integration, Cursor Marketplace, Cursor SDK) extends platform reach beyond the standalone IDE, positioning Cursor as an agentic development runtime rather than an editor. SpaceX and other customers have entered deep model training partnerships with Anysphere (announced April 21, 2026), further advancing Composer's training corpus and quality.

5.6 Exhibits

6.1 Customer Segmentation — PLG Base and Enterprise Motion

Cursor operates a product-led growth (PLG) architecture with four distinct buyer tiers. Individual developers on the Hobby (free) tier access limited AI features and usage quotas; Pro ($20/month) and Pro+/Ultra ($60-$200/month) users access full AI completions and higher model usage limits. The Teams tier ($40/user/month) serves engineering squads of 2-100+ developers with shared billing, admin controls, Privacy Mode by default, and model-level access governance. Enterprise (custom pricing) targets large organizations (typically 1,000+ engineers) needing SSO/SCIM, CMEK, audit logs, compliance logging, and custom security reviews. As of late 2025, Cursor had approximately 1M+ paying users across all tiers and 40,000-50,000 team accounts, with the company claiming 64-70% adoption across the Fortune 500/1000. Enterprise accounts contribute approximately 60% of ARR despite representing a smaller fraction of total users, reflecting the ~20× ARPU differential between enterprise contracts and individual Pro subscriptions. Geographic reach spans North America, Europe, and Asia-Pacific, with cursor.com localized in 9 languages (English, simplified Chinese, Japanese, traditional Chinese, Spanish, French, Portuguese, Korean, German), indicating deliberate international expansion. Enterprise customers include global financial institutions (NAB in APAC, Optiver globally), US tech companies (NVIDIA, Stripe, Coinbase), fintech (Rippling, OnePay), and LATAM tech (Mercado Libre).

6.2 Adoption Trajectory and Growth Evidence

Cursor's ARR trajectory is among the fastest documented in B2B software history. Anysphere's ARR grew from approximately $4M (April 2024) to $48M (October 2024) — a 12× expansion in six months — then to $100M ARR by January 2025, $200M by April 2025 (doubling in three months), $500M by June 2025, $1B by November 2025, and $2B by February 2026. From $100M to $2B ARR in approximately 13 months represents a roughly 20× expansion. The number of paying individual users reached approximately 720,000 at $200M ARR (April 2025), implying an average revenue per user (ARPU) of roughly $277 annually at that stage. By late 2025, Anysphere reported passing 1M paying users. Enterprise expansion accelerated significantly through 2025: Anysphere hired its first dedicated sales representative in late 2024, and by April 2025 had grown the enterprise GTM team to 7 reps (per Sacra), compared to competitor Windsurf/Codeium's 75-person sales team. The lean GTM structure reflects Cursor's reliance on bottom-up developer adoption and word-of-mouth rather than traditional enterprise sales motion. By December 2024, Cursor had already reached notable enterprises including OpenAI, Midjourney, Perplexity, Replicate, Shopify, and Instacart. By 2025, the customer base had expanded to include NVIDIA (40,000 engineers company-wide), Coinbase (all engineers by Feb 2025), NAB (6,000+ developers), Stripe (hundreds to thousands of engineers), Rippling (150 → 500+ engineers in weeks), and dozens of the largest global technology companies.

6.3 Named Customer Proof — Production Deployments and Reference Quality

Cursor's customer page and case studies provide direct executive testimonials from production deployments across a range of industries and company sizes. At the large-enterprise end, NVIDIA deployed Cursor company-wide to over 40,000 engineers. Coinbase CEO Brian Armstrong stated that by February 2025, "every Coinbase engineer had utilized Cursor" and that "single engineers are now refactoring, upgrading, or building new codebases in days instead of months." Stripe CEO Patrick Collison noted Cursor "quickly grew from hundreds to thousands of extremely enthusiastic Stripe employees." Datadog CTO and co-founder Alexis Lê-Quôc stated "100% adoption across our engineering team." Optiver expanded "to firm-wide deployment" across a global trading firm. Trimble reports 800+ engineers deployed. In the mid-market, National Australia Bank standardized 6,000+ developers on Cursor, choosing it over Amazon Q and GitHub Copilot, and ran 3× faster legacy migrations. Amplitude achieved a 3× increase in weekly production commits with 1,000+ automated agent runs per week. Brex CTO James Reggio confirmed "more than 70% of our engineers now use Cursor." Rippling CTO Albert Strasheim highlighted growth from 150 to 500+ engineers (~60% of the engineering org) in weeks. Monday.com reports a 2-5× increase in engineering velocity; Mercado Libre VP Technology reports features built in a day versus weeks previously. Fox CTO Melody Hildebrandt noted unprecedented employee gratitude for a tooling change. Across industries, independent investor validation comes from Carlyle (a top-10 global PE firm), whose technology executive stated engineering teams in Carlyle's portfolio are using Cursor to accelerate product roadmaps and ship features faster.

6.4 Retention Dynamics and Expansion Pattern

Cursor does not publish NRR, GRR, cohort retention rates, or contract renewal data. However, the ARR trajectory provides strong indirect evidence of high dollar retention and robust net expansion: growing from $100M ARR (January 2025) to $2B ARR (February 2026) in thirteen months, against a starting base of approximately 720,000 paying users, implies that revenue expansion — via seat additions and tier upgrades — substantially exceeds any churn. This is consistent with NRR well above 150% for the enterprise segment. The land-and-expand pattern is well-documented: initial adoption typically starts with 1-5 individual engineers on the Pro plan, spreads virally through word-of-mouth within an engineering team, triggers a Teams plan upgrade, and ultimately leads to a company-wide enterprise mandate. NAB expanded from a pilot to 6,000+ developers to a planned 10,000+ engineers. Coinbase went from individual use to company-wide mandate. Rippling achieved 60% of engineering org adoption in weeks. Enterprise switching costs are compounded by organizational tooling built on Cursor's primitives: NAB built NAB CEL (context engineering library) directly on Cursor's rules, skills, and hooks APIs; Amplitude built autonomous development pipelines on Cursor Automations. These represent durable organizational integrations that increase renewal probability and deepen expansion potential over time. The April 2025 support bot incident (an AI-generated support response fabricated a fake "one machine" policy, triggering user backlash) is the most notable retention risk event to date and illustrates the danger of deploying LLM-based support at scale without guardrails. The company issued a public apology and updated the support workflow.

6.5 Concentration Risk and Channel Dependency

Revenue concentration risk is modest at the user level (1M+ paying users limits single-account concentration), but moderately elevated at the enterprise account level. Roughly 64-70% of Fortune 500/1000 companies are claimed adopters, suggesting some of the largest accounts (NVIDIA at 40,000+ engineers, NAB at 6,000+ growing to 10,000+) could represent individually material revenue percentages. No top-10 customer concentration data is publicly available. Cursor's go-to-market is entirely direct: no channel resellers, no systems integrator partnerships, and no marketplace distribution partnerships have been publicly announced as of May 2026. This creates both a moat (direct relationships with engineering leaders) and a scaling constraint (GTM headcount must scale to serve enterprise accounts). The Koala acquisition (July 2025) added a PLG optimization team to maintain the bottom-up developer motion alongside the expanding enterprise sales team. Pricing sensitivity is an observed risk: the April 2025 pricing controversy (when Cursor's AI support bot fabricated a one-machine policy and users perceived a stealth price increase) generated significant social media backlash and HN discussion. Future pricing adjustments to the individual Pro tier carry the risk of alienating the developer base that is the primary acquisition channel for enterprise land-and-expand.

6.6 Exhibits

7.1 Regulatory, Legal, and Compliance Risks

Cursor's primary legal exposure is copyright: AI coding models trained on open-source code may have incorporated copyrighted material, and the US Copyright Office's May 2025 Part 3 report on Generative AI Training (pre-publication) explicitly signals unsettled liability for training-data usage. Active litigation like Andersen v. Stability AI establishes that plaintiffs will pursue copyright claims against AI content generators; analogous claims could reach Cursor through its model providers or directly from enterprise customers whose codebases contain licensed IP. The EU AI Act presents a second regulatory wave. GPAI model obligations applied from August 2025; Cursor, as a deployer of systemic-risk GPAI models (GPT-4o, Claude), faces secondary transparency and incident-reporting obligations it has not yet publicly addressed. European enterprise deployments also trigger GDPR Art. 25 data-protection-by-design requirements: unless Privacy Mode (ZDR) is consistently activated, source code containing PII or trade secrets may be transmitted to non-EU providers without adequate transfer mechanisms. The most material near-term compliance gap is the absence of FedRAMP authorization and on-premises deployment. This structurally excludes Cursor from US federal government, defense contractors, HIPAA- covered healthcare systems, and highly classified environments. GitHub Copilot Enterprise and Amazon Q offer FedRAMP paths; Cursor does not. This creates an addressable market ceiling of approximately 15-25% of the enterprise software development TAM that Cursor cannot serve without either FedRAMP investment or an on-premises product.

7.2 Operational, Security, and Technical Risks

Cursor's operational risk profile centers on three compounding factors: AI hallucination in trust-sensitive surfaces, expanding agentic attack surface, and VS Code fork technical debt. The April 2025 support bot incident — in which Cursor's AI fabricated a policy banning competitor tools — demonstrated the company's vulnerability to AI-generated misinformation in customer-facing deployments. While the immediate enterprise impact was limited, the incident revealed that Cursor's operations are exposed to the same AI failure modes it enables for customers. A second, larger incident in an enterprise support or agent context (e.g., the agent taking irreversible actions on a production codebase based on a hallucinated instruction) could trigger significant enterprise liability and procurement pauses. The OWASP LLM Top 10 framework identifies prompt injection (LLM01) and excessive agency (LLM06) as the critical risks for AI coding IDEs. Cursor Automations and Agent mode, which execute multi-step agentic pipelines with filesystem and terminal access, represent elevated OWASP LLM06 exposure. The integration of MCP third-party tool servers further expands the trust boundary: a malicious MCP server could exfiltrate credentials or inject instructions that cause Cursor's agent to modify production infrastructure. Cursor has not disclosed fine-grained MCP permission controls or audit logging for agent actions. The VS Code fork creates compounding technical debt as Microsoft evolves VS Code upstream. Cursor's differentiating features (shadow workspace, Composer multi-file agent) are implemented in divergent forks of VS Code and Electron. If Microsoft adds equivalent agentic features to VS Code natively — as Copilot Chat has progressively done — Cursor's differentiation narrows while the fork maintenance burden grows. The risk is asymmetric: Microsoft controls the upstream, Cursor does not.

7.3 Partner, Dependency, and Competitive Risks

Cursor's most concentrated risk is its three-model API dependency on OpenAI, Anthropic, and Google Gemini. At the $200M ARR stage, model API COGS represented approximately 80-90% of revenue; by the $500M ARR stage Sacra estimated this had declined to ~60-65%. Despite improvement, Cursor lacks a self-hosted inference fallback that would protect it from unilateral pricing increases by its model providers. Both OpenAI and Anthropic are building their own developer tooling (OpenAI Canvas, Anthropic Projects/ Claude.ai), creating a structural incentive to eventually restrict or de-prioritize API access for competing application layers. The SpaceX/xAI acquisition option is the highest-severity partnership risk. The $60 billion option disclosed in April 2026 — with SpaceX holding a bilateral right to acquire Cursor by year-end — creates enterprise procurement paralysis for Cursor's largest competitor accounts. Companies directly competing with SpaceX, Tesla, or xAI ventures may refuse to deploy Cursor enterprise-wide. EU antitrust review of the deal could take 18-24 months and create prolonged uncertainty. If the acquisition completes, Cursor becomes part of the Musk AI empire alongside xAI/Grok, potentially driving away customers who resisted vendor concentration. Microsoft's Copilot bundling is a structural threat: GitHub Copilot is included at no incremental cost in GitHub Enterprise Cloud ($21/user/month) and integrated into Microsoft 365 E3/E5 licenses. This creates pricing pressure that Cursor's $40-200/user pricing cannot overcome in Microsoft-centric enterprise accounts. Amazon Q Developer is FedRAMP-eligible and integrated with AWS developer toolchains, making it the default choice for AWS-heavy regulated sector accounts. Both competitors have structural advantages Cursor cannot easily replicate through product velocity alone.

7.4 Financial, People, and Execution Risks

Cursor's financial risk centers on COGS trajectory and capital adequacy. The company reached only slight gross-margin profitability in April 2026 after raising $3.47 billion; GAAP profitability requires further model cost optimization or own-model inference deployment. If model API prices stabilize or increase (reversing the 2023-2025 trend of declining inference cost), Cursor's path to sustained profitability extends and capital needs grow. At ~150 employees against $2B+ ARR, Cursor's revenue-per-employee ratio (~$13M+) is extraordinary but unsustainable at enterprise scale. Enterprise customers require dedicated customer success, security review, professional services, and compliance teams — none of which are evident in Cursor's current headcount. As deal sizes grow and procurement complexity increases, the absence of enterprise go-to-market infrastructure becomes a retention and expansion risk. Key-person concentration is extreme: four co-founders run the company with no publicly disclosed leadership bench below them. The loss of any founder — through acquisition distraction, departure, or health event — would create organizational uncertainty amplified by Cursor's hyper-growth phase. The xAI acquisition option introduces a founder alignment risk: if any founder is more aligned with the xAI transaction than with independent company building, that tension could manifest in product roadmap decisions or recruiting.

7.5 Mitigations, Thesis-Break Triggers, and Diligence Asks

Cursor has meaningful risk mitigations in place. SOC 2 Type II provides a compliance baseline for commercial enterprise accounts. Privacy Mode with ZDR addresses the data residency concerns of privacy-sensitive enterprises. The three-model API architecture provides redundancy against single- provider outage. Cursor's PLG velocity has demonstrated resilience through both the April 2025 bot incident and broad competitive pressure, with ARR accelerating post-incident. Thesis-break triggers that would materially change the investment thesis include: (1) OpenAI or Anthropic restricting API access for AI coding IDE resellers; (2) a material security breach or supply-chain injection through Cursor's agent pipeline causing enterprise data loss; (3) the xAI acquisition completing and triggering significant enterprise customer churn; (4) Microsoft bundling advanced agentic coding into VS Code natively at zero incremental cost; or (5) adverse copyright precedent requiring Cursor to indemnify enterprise customers for AI-generated code. Priority diligence asks: (1) Cursor's enterprise agreement terms including IP indemnification provisions; (2) actual gross margin and COGS breakdown by model provider; (3) incident response and bug bounty program details; (4) FedRAMP roadmap and timeline; (5) MCP permission control architecture and agent audit logging; (6) founder vesting schedules and acquisition lockup terms under the xAI option.

7.6 Exhibits

8.1 Investment Thesis and Entry Context

Cursor presents one of the most compelling growth stories in enterprise software history: $2B ARR achieved in approximately 3 years from founding, representing the fastest ARR trajectory ever documented in developer SaaS. The Series D at $29.3B values Cursor at ~14.7x trailing ARR, a multiple that sits modestly below Datadog's public market premium (~17-22x) but reflects an implicit bet on two things: continued 150%+ YoY ARR growth and substantial gross margin improvement from the current ~15-20% to enterprise SaaS norms of 70%+. The investment thesis rests on four pillars: (1) Cursor has category-defining market position in AI coding — 1M+ paying users, 64-70% Fortune 500 penetration (company-claimed), $2B ARR; (2) AI coding is a winner-take-most market driven by model quality, context management, and workflow integration — areas where Cursor is leading; (3) enterprise expansion from individual Pro users to Teams to Enterprise has demonstrated 3-6x land-and-expand multipliers at named accounts; and (4) own-model inference (Cursor 3, April 2026) signals a pathway to margin improvement that is critical to valuation justification. The anti-thesis is equally compelling: Microsoft controls the VS Code upstream and GitHub Enterprise, giving it a structural ability to bundle equivalent features at zero incremental cost; Cursor's current economics show near- zero gross margin, meaning investors are paying a 5-7x premium over current-economics value for projected future margin expansion; and the SpaceX/xAI acquisition option effectively caps near-term returns at ~2x.

8.2 Comparable Valuations and Market Benchmarks

The developer tools M&A landscape provides three key data points. GitHub's 2018 acquisition at $7.5B on ~$200M ARR (37.5x multiple) reflected strategic platform value for the world's code repository. Windsurf's 2025 Google acquisition at $2.4B on ~$150-250M ARR (10-16x multiple) establishes a floor for AI coding IDE acquisitions. Cursor's $29.3B at $2B ARR (14.7x) sits in this range, reflecting a financial-strategic blended multiple. Public market comps show high-growth developer platform companies trading at 11-25x ARR: GitLab at 11-12x ($760M ARR, 29% YoY growth, NRR 124%), Datadog at 17-22x ($2.3B ARR, ~25% YoY growth), Cloudflare at 25-30x ($1.6B ARR, strong developer positioning). Cursor's 14.7x at 150%+ YoY growth appears conservatively priced relative to peers with lower growth rates, assuming gross margin improvement materializes. Private AI company benchmarks: Databricks at $62B on ~$1.5B ARR (41x multiple with 75% gross margin) and OpenAI at $300B on ~$4-5B ARR (60-75x, reflective of model layer premium) suggest that as Cursor matures its gross margin, there is precedent for further multiple expansion from 14.7x toward 20-30x, implying significant upside optionality in the bull case.

8.3 Bull / Base / Bear Scenarios and Return Analysis

Bull case: Cursor reaches $8-10B ARR by end-2026, gross margin improves to 40-50% with own-model inference contributing 30-40% of completions, and the xAI acquisition option is not exercised. At 15x ARR, Cursor is valued at $120-150B — implying 4-5x return from Series D entry and a 2027-2028 IPO at a $100B+ market cap. This requires sustained 150%+ YoY ARR growth and no material competitive disruption from Microsoft or API access restriction. Base case: Cursor reaches $5-6B ARR by end-2026, gross margin improves to 30-40%, and is acquired by xAI or IPOs in 2028 at 10-12x ARR ($50-72B). This implies a 1.7-2.5x return from Series D entry over 2-3 years, approximately equivalent to the xAI acquisition option floor. For late-stage VC funds requiring 3x+, the base case does not clear the return threshold without significant multiple expansion at IPO. Bear case: ARR decelerates to $3-4B (Microsoft bundling, API restriction), gross margin stalls at 15-25%, and multiples reset to 8x in a risk-off market correction. Cursor trades at $24-32B, below the $29.3B Series D entry, resulting in a loss for investors at the latest round. In this scenario, earlier-round investors at lower valuations still achieve positive returns while Series D investors experience a 10-20% loss.

8.4 Valuation Stance, Recommendation, and Final Diligence Asks

Valuation stance: PRICED FOR GROWTH WITH LIMITED MARGIN OF SAFETY. At 14.7x ARR on $2B ARR with near- zero gross margins, Cursor is pricing in substantial margin improvement and continued hyper-growth that has not yet been sustained for more than 13 months. The valuation is defensible but requires belief in two unproven assumptions: (1) gross margin will reach 60%+ as own-model inference scales; and (2) Microsoft will not exercise its bundling advantage before Cursor's enterprise flywheel becomes self-sustaining. Recommendation: SELECTIVE BUY WITH CONDITIONS. For late-stage growth funds with 3-5 year horizons and 3x+ return targets, the investment is justified IF: (a) investors obtain governance rights to participate in or block the xAI acquisition decision; (b) pre-investment diligence confirms gross margin improvement to 35%+ is on the 12-month roadmap; (c) ARR cohort data confirms >130% NRR in the enterprise tier. Investors that cannot satisfy these conditions should PASS at the $29.3B valuation. Risk rating: HIGH. The combination of near-zero gross margin, existential competitor in Microsoft/Copilot, xAI acquisition return cap, and model API dependency creates a risk profile that is materially above the median enterprise SaaS investment. The growth trajectory partially compensates, but the margin of safety at $29.3B is thin. Priority diligence asks: (1) GAAP financial model (P&L, gross margin by cohort, COGS breakdown by model provider); (2) Series D governance terms (acquisition veto rights, Board representation); (3) ARR cohort retention and NRR data by customer segment; (4) FedRAMP investment plan and timeline; (5) own-model inference deployment roadmap with ARR% coverage targets; (6) Koala acquisition earnout terms and dilution impact.

8.5 Exhibits