POOLSIDE

Jason Warner spent years as the CTO of GitHub. He had a front-row seat to Copilot, the tool that first showed the world that AI could write real code.

Eiso Kant had already built and sold a code-analysis company called Athenian. The two met and agreed on one thing.

Most AI coding tools just rented models from OpenAI. They wanted to own the model.

They founded Poolside in early 2023 and raised a $26 million seed round in May, then extended it to $126 million by August. That is a lot of money for a company with no product.

The pitch was simple. Software is the one field where you can automatically check if the AI got it right, because code either runs or it doesn't.

Poolside is building a large AI model trained specifically to write software. The trick is something they call reinforcement learning from code execution.

In plain English, the model writes code, the code gets run, and the model learns from whether it actually worked. Most models just predict the next word and hope.

Poolside grades itself against reality. The plan is to sell this to big enterprises and governments that want a private coding AI inside their own walls, not a public chatbot.

They also announced plans for their own data centers to train these models. That is where the billions go.

Poolside's main product is its foundation model built for software development, sold as an enterprise coding assistant that plugs into a company's own code base. It handles code completion, code generation, and answering questions about a specific project.

They pitch it as a private alternative to GitHub Copilot, one that runs inside the customer's own infrastructure instead of a public cloud. The bet is that serious enterprises will pay a premium to keep their code in-house.

Poolside skipped the usual startup playbook of launch fast and iterate in public. Instead they raised enormous rounds and went quiet to build.

The strategy is to sell to large enterprises and the public sector, where a private, self-hosted coding model is worth a fortune. They courted regulated industries that cannot send their code to a public API.

The 2024 round pulled in NVIDIA, eBay, and a wall of institutional money, which bought them both cash and credibility. Then came reports of a $12 billion valuation in 2025 and a massive data center project in Texas.

The burn is terrifying. Poolside is trying to train frontier models from scratch, which costs hundreds of millions in compute alone.

Meanwhile giants like OpenAI, Anthropic, and Google pour billions into the same problem. Their model has to be genuinely better than Copilot or Cursor to justify the price.

Those competitors already have millions of paying users. Reports in 2026 pointed to trouble with the huge Texas data center plan.

The core question is brutal. Can a startup out-train companies with ten times the money and a head start?

Poolside's backers read like a who's who of deep pockets. The $126 million seed in 2023 started with Redpoint Ventures, then got a $100 million extension led by French telecom billionaire Xavier Niel and Felicis Ventures.

The $500 million Series B in October 2024 was led by Bain Capital Ventures. DST Global, eBay, and NVIDIA all wrote checks at a $3 billion valuation.

NVIDIA's role matters twice over. They are both an investor and the company Poolside buys its chips from.

By late 2025, reports had NVIDIA committing up to another $1 billion as the valuation climbed toward $12 billion.