Poolside
Foundation-model lab building frontier AI for software development, sold to enterprises.
1. Core Product / Service
Poolside builds its own foundation models specialized for code rather than wrapping third-party LLMs. Its flagship model, malibu, is trained using Reinforcement Learning from Code Execution Feedback (RLCEF) — running generated code and rewarding what actually executes correctly. A separate completion model, point, handles low-latency inline suggestions. Both feed poolside Assistant, an agentic pair programmer that writes and edits code, answers questions, and performs context-aware code review through a natural-language interface, using context drawn from the developer's workspace. The company also built an internal "Model Factory" framework for constructing and evaluating its foundation models. [Wikipedia, 2026-06-29]
The product is delivered as a deployable foundation model rather than a hosted-only SaaS. A December 2024 partnership with AWS lets enterprises run Poolside's models inside their own environment via Amazon Bedrock and EC2, emphasizing security, privacy, and the ability to fine-tune on a company's private codebase. [Wikipedia, 2026-06-29]
2. Target Users & Pain Points
Poolside sells to large enterprises — particularly regulated and security-conscious organizations — that want frontier-grade coding AI but cannot send proprietary source code to a public API. The core pain solved is twofold: (1) generic coding assistants are not tuned to a company's internal codebase, conventions, and private libraries; and (2) compliance and IP constraints rule out third-party cloud inference. By shipping a model that can be deployed in-VPC and customized, Poolside targets buyers for whom in-house deployment and data control are decisive. [Wikipedia, 2026-06-29; AWS, 2026-06-29]
3. Competitive Landscape
| Company | Approach | Differentiation vs Poolside |
|---|---|---|
| cursor | AI-native IDE on top of frontier models | Owns the developer surface; does not train its own frontier code model |
| GitHub Copilot | Assistant layered on OpenAI/Anthropic models | Massive distribution via GitHub; not a self-built foundation model |
| anthropic | General frontier lab; Claude strong at code | Broad model, not code-only; sells API + Claude Code |
| openai | General frontier lab + Codex agents | Scale and capital advantage; general-purpose |
Poolside's bet is vertical: a foundation lab focused solely on code, deployable on customer infrastructure. Its differentiation rests on owning the full stack — model, training method (RLCEF), and (via Project Horizon) compute — rather than competing on the IDE surface. The structural risk is that general frontier labs match or beat code-specific quality while carrying far larger budgets. [TechCrunch, 2026-06-29]
4. Unique Observations
- Poolside is one of the clearest cases of a vertical foundation lab attempting to go full-stack — model plus its own 2-gigawatt compute campus (Project Horizon, Pecos County, Texas, first phase October 2025). That vertical ambition is also what broke it: in April 2026 a planned $2B Series C at a ~$14B valuation collapsed after CoreWeave walked away from a 15-year anchor lease and Nvidia's rumored ~$1B anchor check fell through, reportedly amid investor doubt that Poolside could train models competitive with anthropic, openai, and Google DeepMind at the frontier. [Bloomberg/FT via TechCrunch, 2026-06-29]
- The company quietly relocated its center of gravity toward France (Paris hub, monthly concentrated work weeks) while remaining US-headquartered — a structure tied to its early French backers and capital base. [Wikipedia, 2026-06-29]
- The gap between the $3B post-money (Oct 2024) and the aborted $14B Horizon-era ask illustrates how much of the implied valuation was a bet on compute and frontier-model promise rather than realized revenue; reporting in spring 2026 pointed to a far smaller bridge near the $3B level rather than a markup. [Sacra, 2026-06-29]
- Selling a deployable code model to enterprises positions Poolside against cursor indirectly — Poolside owns the model and lets the IDE layer be neutral, the inverse of Cursor's surface-first strategy.
5. Financials / Funding
- Total raised (primary equity): $0.63B
- Latest valuation: $3.0B
| Date | Round | Amount | Post-money | Lead investor(s) |
|---|---|---|---|---|
| 2023-05 | Seed (initial tranche) | $0.03B | — | Redpoint Ventures |
| 2023-08 | Seed extension (some sources label Series A) | $0.10B | $0.5B | Felicis; Xavier Niel |
| 2024-10 | Series B | $0.50B | $3.0B | Bain Capital Ventures |
| 2025-10 | Series C (FAILED / did not close) | $2.00B | $14.0B | Nvidia (NVentures, anchor, up to $1B) |
6. People & Relationships
- Founders / key people: Jason Warner (CEO, former GitHub CTO) and Eiso Kant (CTO, software entrepreneur), co-founded the company in early 2023. ~150 employees across the US, UK, and France as of December 2025. [Wikipedia, 2026-06-29]
- Notable investors: Redpoint Ventures, Felicis, Xavier Niel (French telecom billionaire), Bain Capital Ventures, DST Global, eBay; Nvidia (NVentures) committed up to $1B as anchor of the failed Series C. [TechCrunch, 2026-06-29]
- Partners / competitors: AWS (Bedrock/EC2 deployment partner, Dec 2024); coreweave (Project Horizon data-center partner that exited the anchor lease in 2026). Competes with cursor, GitHub Copilot, anthropic, and openai in AI-assisted software development. [Bloomberg, 2026-06-29; Wikipedia, 2026-06-29]