Nvidia has agreed to pay Poolside $6bn to license the software the startup used to build AI models, and to offer jobs to 109 of its staff. It is also investing $1bn in what remains, at a $12bn pre-money valuation.

The terms come from a letter Poolside sent its investors, first reported by Newcomer. The licence is non-exclusive. The three founders are staying.

The deal is “not an acquisition and it is not an acquihire”, the letter says. Poolside intends to distribute the $6bn to its investors by the end of next year.

The third deal of this shape

Nvidia has used the same structure twice before. It buys a licence, hires the staff, takes a stake, and the company continues. The licences have been non-exclusive in each case, and none of the three companies has been bought outright.

TNW reported the template in May, when Nvidia paid Groq $20bn for its inference technology and took its top engineers. Groq stayed independent and raised $650m for what was left. By this month it had closed a $350m round at $3.5bn, with Nvidia participating.

Nvidia paid $20bn for the technology and staff in December. Groq as a continuing company was valued at $3.5bn this month, in a round Nvidia joined.

Nvidia struck a similar arrangement with the hardware startup Enfabrica for about $900m, according to The Decoder. Such deals let big technology companies secure staff and know-how without buying a company outright, it notes. They also avoid the regulatory review a purchase can bring.

Who leaves and who stays

In the Groq deal, Nvidia took the company’s top engineers and Groq appointed new leadership. At Poolside the three founders stay and 109 employees receive Nvidia’s offers.

Poolside’s chief executive Eiso Kant described the size of the team on the Latent Space podcast last month. “Less than 70 people built this model,” he said. “Less than 115 between engineering and researchers, like, together did this effort.”

Latent Space called it a reverse-execuhire. The arithmetic behind that label is stark. Poolside’s chief executive Eiso Kant said on the same publication’s podcast last month that fewer than 70 people built the company’s model. Fewer than 115 worked across engineering and research in total.

The software Nvidia is licensing is called the Model Factory. It is the system Poolside used to build its models.

Poolside began as a developer of a coding AI agent and moved into data centres before releasing its own open-source model on Nvidia server chips, according to The Information.

Nvidia was already an investor, having committed up to $1bn to Poolside last October. The licence announced on Friday is non-exclusive, which leaves Poolside able to license the same software elsewhere.

What the letter says about the cluster

The letter sets out why the company stopped building frontier models.

“For the last 3 1/2 years we’ve been directionally correct in a race where capital requirements went vertical,” it says. Then the specific: “At the end of last year, we had a 6 week window in which to raise $2 billion dollars to pay for a 40,000 GB300 cluster coming online in January. We didn’t close it in time, and we lost the cluster.”

The letter goes on to say Poolside could have built a frontier-rivalling model with 10,000 to 20,000 of those chips. Next year’s frontier, it argues, needs “far more than an order of magnitude larger cluster”. The constraint “is not only capital, it is physical data center space and contracted compute”.

The letter also says the company had been “directionally correct” for three and a half years, and that the world “has not yet reached 0.1%” of the transition to AI.

Nvidia’s own open models

Nvidia builds open models in the Nemotron line, and TNW reported this month that it is working towards a trillion-parameter open model. Poolside’s Laguna, built by the same team, was pitched as the West’s answer to DeepSeek and Qwen. Poolside trained it on Nvidia server chips.

Nvidia halved a $250bn commitment to OpenAI this month.

Nvidia has not said what makes the licence worth $6bn. The Information’s Amir Efrati wrote that it is not clear why Nvidia is paying such a large licensing fee. Nvidia has not commented publicly on the deal, and Poolside has not published the letter.

What is left of Poolside

The founders say they are “not ready to share the updated vision”.

Poolside Infrastructure Company, spun out in January, is building a 1.2GW data centre in Texas. It appointed a chief executive two months ago and a chief financial officer this week. Both appointments predate Friday’s announcement.

The letter’s stated thesis points somewhere else again. Human-level capability will be “fully commoditized by open source models”, it argues, while superintelligence will not. It then divides valuable problems into two kinds. Some are intelligence bound, and some are experiment bound. Software and accounting fall into the first and become “a low margin commodity”. Curing cancer falls into the second, because no amount of intelligence substitutes for real-world experiments. The letter says the second is where the value will be.

“We could put 100,000 of the world’s brightest minds together to solve cancer but without a real world experimental feedback loop, they likely never will,” the letter says.

“AI will become the world’s most valuable scientific discovery engine,” it concludes.

Across the three deals, Nvidia has committed roughly $27bn. Each was structured as a licence and a hiring round rather than a purchase, and in each case the company continued to operate.

Get the TNW newsletter

Get the most important tech news in your inbox each week.