Nvidia has agreed to pay $6 billion for a license to Poolside’s Model Factory, the software platform the AI startup used to build its coding models, and to hire 109 of its employees, according to a letter Poolside sent to its investors that was reported by The Information on Wednesday. Poolside is simultaneously raising $1 billion in new funding at a $12 billion valuation for the business that remains.
The deal is one of the most unusual in the recent AI industry. It is not an acquisition and not an acqui-hire, the letter says, in language that Poolside’s founders emphasized to their investors. Nvidia is buying a non-exclusive license to the Model Factory, the internal system of tools and pipelines that Poolside built to train and refine its Laguna family of coding models, and it is extending job offers to the roughly 109 people who worked on that system. The three founders are staying with Poolside.
The structure reflects a pattern Nvidia has been developing. Rather than buying AI companies outright, the chip maker has been licensing technology, hiring portions of teams, and taking stakes, leaving the original companies to continue operating. It used a similar arrangement with the chip startup Enfabrica and, on a larger scale, with the inference company Groq, where it paid for technology and staff while Groq remained independent. The Poolside deal is the largest yet in the series.
For Poolside, the arrangement resolves a question that has hung over the company since its founding. The startup, led by chief executive Eiso Kant, built a reputation for training models on its own infrastructure, running tens of thousands of experiments a month and producing coding models that compete with the best in the field. But the economics of frontier AI favor scale, and Poolside’s investors wanted a path to liquidity. The $6 billion license fee will be distributed to those investors, according to the letter, while the $1 billion new investment keeps the remaining company funded at a $12 billion valuation.
What Nvidia gets is the manufacturing process, not just the product. The Model Factory is the platform Poolside used to build Laguna, its flagship family of models, and Nvidia’s interest is in owning that capability: the tools, the pipelines, and the people who know how to run them. The company has said it wants to build its own models, and the license gives it the infrastructure know-how without the distraction of integrating an entire company.
The deal also makes strategic sense for Nvidia’s hardware business. Poolside trained its models on Nvidia chips, and the Model Factory is built around Nvidia’s platforms, meaning the license reinforces the ecosystem that sells Nvidia’s processors. The more companies that build model factories like Poolside’s, the more demand there is for the chips that run them, and Nvidia is effectively investing in the spread of that model.
The timing of the announcement was notable for its density. On the same day, The Wall Street Journal reported that Nvidia is in talks to invest several hundred million dollars in Cloverleaf Infrastructure, a developer of power infrastructure for data centers that holds land reserved for large-scale electricity generation. The two reports, taken together, show Nvidia extending its reach to both ends of the AI supply chain: the software that trains models and the electricity that powers the data centers where they run.
Cloverleaf’s business is emblematic of the industry’s newest bottleneck. Data centers for AI consume power at a scale that has overwhelmed local grids, and companies that can secure land, transmission rights, and power purchase agreements have become valuable in their own right. Cloverleaf has assembled a portfolio of such sites, according to people familiar with the company, and Nvidia’s investment would give it a stake in ensuring that the data centers using its chips have electricity to run them.
The two moves reflect a broader shift in how Nvidia thinks about its position. The company has historically sold chips and let others worry about the rest, but the AI boom has made it the de facto coordinator of a vast supply chain, from foundries to power plants. By licensing model factories and investing in power infrastructure, Nvidia is placing bets on the pieces of the ecosystem that could otherwise become bottlenecks, protecting the demand for its core product.
The Poolside deal has drawn scrutiny for its price. Analysts noted that $6 billion for a license is a large sum, and Nvidia has not explained what makes the Model Factory worth that much. The answer may lie in the talent: the 109 people receiving job offers include the engineers who built one of the most efficient model training operations in the world, and hiring them gives Nvidia a ready-made team for its own model efforts. The license itself may be the legal wrapper that makes the hiring acceptable.
For Poolside’s remaining employees and its new investors, the future is less certain. The company will continue to develop its coding models, and the founders have said the Model Factory license is non-exclusive, meaning Poolside can sell access to the platform to other customers. The $1 billion raise values the slimmed-down company at $12 billion, and its investors are betting that a company that just sold its most valuable asset can still compete. The answer will come in the models it ships over the next year.
For the industry, the deal is a signal about the value of AI infrastructure. Nvidia’s willingness to spend $6 billion on a model-building platform, and its parallel interest in power infrastructure, shows where the money in AI is moving: not just to the models themselves, but to the systems and energy that produce them. The companies that control those systems, and the land and power that feed them, are becoming the quiet winners of the AI boom.
