Poolside

Poolside closed a $500 million Series B in October 2024 at approximately $3 billion valuation. The round was led by Bain Capital Ventures with participation from Nvidia, eBay Ventures, DST Global, StepStone Group, Citi Ventures, Felicis Ventures, and Redpoint Ventures.

As of October 2025, Poolside is reportedly raising $2B at a $12B valuation, with more than $1B already committed, including $700M from existing investors and $500M to $1B from Nvidia.

Poolside has raised approximately $626 million in total funding across its funding history.

Poolside builds specialized large language models that function as AI software engineers deployed within customer environments. The product consists of four integrated layers that work together to automate coding tasks.

The foundation includes two core models: Malibu, a high-capacity model optimized for complex tasks like multi-file code generation, test creation, and refactoring, and Point, a smaller quantized model designed for sub-200 millisecond code completion in IDEs. Both models share the same tokenizer, allowing seamless context switching between heavy-duty reasoning and real-time assistance.

The Poolside Assistant provides the user interface through IDE plugins for VS Code, JetBrains, Visual Studio, and Coder, plus a CLI tool. Developers can chat with the AI, request inline code edits, generate tests, and search across repositories using natural language. The assistant automatically routes requests to the appropriate model based on task complexity.

Multi-agent orchestration enables spawning single or cooperating agents that can plan multi-step projects, execute code in sandboxes, run tests, and open pull requests automatically. Agents can call internal tools like test runners and static analyzers, as well as external APIs for project management systems.

Everything deploys within customer boundaries—on-premises, in private VPCs, or through managed cloud options like AWS Bedrock. The system includes role-based access controls, audit trails, and license-safe training that excludes GPL/AGPL code. Forward-deployed research engineers embed with customer teams to configure integrations and establish evaluation frameworks before full deployment.

In October 2025, Poolside announced Project Horizon, a 2 gigawatt AI campus in West Texas. The agreement with CoreWeave includes more than 40,000 NVIDIA GB300 NVL72 GPUs coming online starting December 2025 and CoreWeave serving as anchor tenant for the first 250MW phase under a 15‑year lease, with 500MW reserved for expansion. The move advances Poolside’s vertical integration strategy, securing long‑term energy and compute capacity for frontier‑scale training.

Poolside operates as a B2B enterprise software company selling AI coding automation to large organizations. The go-to-market approach focuses on enterprises with 5,000+ developers, including major banks and defense contractors.

The business model centers on custom model deployment rather than generic API access. For each customer, Poolside spins up dedicated training environments within the client's infrastructure, ingests their existing codebase and development history, and fine-tunes models specifically for that organization's coding patterns and requirements.

Revenue comes from enterprise contracts that include both the AI models and professional services. Forward-deployed research engineers work directly with customer teams to integrate the platform with existing development workflows, source control systems, and CI/CD pipelines.

The company uses Reinforcement Learning from Code Execution Feedback to generate synthetic training data, creating billions of coding tasks that are executed and evaluated for correctness. This approach allows continuous model improvement while maintaining data privacy within customer environments.

Poolside's partnership with AWS through Bedrock provides a managed deployment option, expanding market reach beyond organizations that want to manage AI infrastructure themselves. The CoreWeave partnership secures GPU compute resources essential for model training and inference at scale.

GitHub Copilot dominates with 20 million users and 90% Fortune 100 adoption, leveraging exclusive OpenAI models and tight integration with VS Code and GitHub workflows. Amazon CodeWhisperer offers a free individual tier and $19 per seat professional plans, positioning as complementary to AWS cloud services with strength in AWS API code generation.

Google's Gemini Code provides early-access CLI tools and 1 million token context windows, bundled into Gemini Enterprise suites. Anthropic's Claude Code integrates with Claude Enterprise plans, emphasizing safety and compliance with 500,000+ token context for repository-scale editing.

These platforms benefit from massive distribution through existing developer tools and cloud relationships, but face perception challenges around vendor lock-in and closed ecosystems.

Replit raised $250 million in September 2025 at $3 billion valuation, offering browser-based IDEs with autonomous agents targeting SMB and educational markets. Cursor focuses on AI-first code editing with strong developer community adoption.

Sourcegraph provides code search and intelligence platforms, while Tabnine and Codeium offer code completion tools. These companies compete on specialized developer experiences and vendor-neutral positioning but often lack the enterprise security and custom model capabilities that large organizations require.

Meta's Code Llama, DeepSeek-Coder, and StarCoder2 enable self-hosted deployments, appealing to cost-sensitive organizations and those with strict data governance requirements. These solutions require significant internal AI expertise but eliminate vendor dependencies and data sharing concerns.

The open source ecosystem creates pricing pressure on commercial offerings while providing fallback options for organizations hesitant about proprietary AI coding tools.

Poolside can extend its agent capabilities beyond code generation into higher-margin DevOps automation including security patching, CI/CD pipeline management, and legacy system refactoring. This expansion would capture 2-3x more spend per developer seat by pulling budget from QA, application security, and DevOps teams.

The platform's existing sandbox execution and tool integration capabilities provide the foundation for automating entire software development lifecycle workflows rather than just individual coding tasks.

The AWS Bedrock partnership enables Poolside to offer fully managed AI coding services within customer VPCs, fine-tuned on proprietary repositories. This addresses highly regulated sectors like defense, healthcare, and financial services that cannot send code to public APIs.

Packaging this as managed services similar to enterprise-grade alternatives to GitHub Copilot creates opportunities for significant contract values in sectors with strict data governance requirements.

Forward-deployed research engineers currently embed with clients to measure ROI and maintain models in production. Expanding this into subscription-based AI-as-a-team bundles that include models, agents, and expert services could grow average contract values well beyond seat-based pricing.

This services-plus-software approach positions Poolside more like specialized consulting firms, capturing value from implementation expertise and ongoing optimization rather than just software licensing.

Model commoditization: As foundation models become increasingly capable and open source alternatives improve, Poolside's custom model training advantage could erode. If general-purpose coding models achieve similar performance to specialized ones, the company's differentiation and pricing power would decline significantly.

Enterprise sales cycles: Poolside's focus on large enterprise customers with complex security and compliance requirements creates long sales cycles and high customer acquisition costs. Delays in enterprise AI adoption or budget cuts could significantly impact revenue growth and cash burn rates.

Compute dependency: The business model requires substantial GPU resources for model training and inference, creating exposure to compute cost inflation and supply constraints. Competition for scarce H100 chips and rising cloud infrastructure costs could compress margins and limit scaling ability.