Nvidia has agreed to pay the artificial intelligence coding startup Poolside roughly 6 billion dollars to license its model development software and take on about 109 of its employees, while separately investing 1 billion dollars into what remains of the company at a 12 billion dollar valuation.
Read that structure again, because the structure is the story. Nvidia is not buying Poolside. It is licensing the technology, hiring most of the people who built it, and taking an equity stake in the shell that stays behind. The practical result looks a great deal like an acquisition and does not trigger the review an acquisition would.
What Nvidia Is Getting
The asset is Poolside’s Model Factory, the internal tooling the company built for developing and training models. Software like this is the unglamorous machinery underneath model development, and it is the kind of thing that takes years to build well and cannot be bought off a shelf.
The 109 employees matter as much as the code. In this field the people who built a system are frequently the only ones who can extend it, which is why the hiring half of these deals is rarely incidental.
Why It Is Not Called an Acquisition
A straight purchase of a company at this size draws antitrust review. A license plus a hiring wave plus a minority investment does not, or at least has not so far, because on paper the target company still exists and still has shareholders.
This is not a novel maneuver. Google, Meta, Microsoft and Amazon have each run versions of it over the past two years with smaller artificial intelligence startups. What is new is the scale. Six billion dollars for a license and a hiring round is a different order of magnitude than what came before.
What Happens to Poolside
The company continues, on paper, at a 12 billion dollar valuation with Nvidia holding a stake. What it continues as is the harder question, given that a large share of the engineering talent is leaving with the deal.
Investors in the original company generally do well in these arrangements, which is part of why founders accept them. Employees who stay behind are in a more uncertain position, working at a company whose most valuable asset has just been licensed to its largest shareholder.
The Regulatory Question
The obvious question is whether regulators will treat the pattern as what it functionally is. Antitrust law is built around transactions, and this is deliberately structured to be several transactions that individually look ordinary.
There has been academic and congressional interest in whether that framing should hold, but no enforcement action has tested it at this scale. Until one does, the structure works, and every deal that closes makes the next one easier to justify.
Why Nvidia Wants Development Tooling
Nvidia sells the hardware that model training runs on, and its long term position depends on the software layer around that hardware staying close to its own ecosystem. Owning the tooling that developers use to build models is a way of making that gravity stronger.
The company has spent the last several years extending upward from chips into software, frameworks and services. Chip access and policy have shaped who can compete at all, a dynamic our coverage of Nvidia H200 chips reaching Chinese buyers examined in detail.
What This Says About Startup Exits
For founders, the license and hire has quietly become a preferred outcome. It closes faster than an acquisition, avoids a regulatory process that can run a year or more, and delivers cash to investors without the integration risk of a full merger.
For the broader market it is a less comfortable trend. If the most capable teams get absorbed into a handful of large firms through a mechanism designed to avoid review, the competitive landscape consolidates without anyone formally approving the consolidation.
What the Numbers Do and Do Not Mean
Some coverage has combined the 6 billion license with the 1 billion investment and reported a 7 billion figure. Those are separate transactions with different structures, and merging them overstates what was paid for the technology.
It is also worth noting that the terms are reported rather than company confirmed. The original scoops came from subscription outlets and were subsequently corroborated, which is strong sourcing but not the same as a filing.
What to Watch Next
Watch whether any regulator opens an inquiry, because the size of this one makes it the most likely candidate yet to draw attention. Watch what Poolside does with the money and the people who stay. And watch whether other chipmakers copy the template.
The deeper signal is about where value is settling in artificial intelligence. It is not settling in models, which depreciate fast. It is settling in the tooling and the people who can build the next one, and Nvidia just paid 6 billion dollars to say so out loud.
Frequently Asked Questions
Did Nvidia buy Poolside?
No. Nvidia licensed Poolside’s Model Factory software for about 6 billion dollars, hired roughly 109 employees, and separately invested 1 billion dollars for a stake. Poolside continues to exist as a company.
Why structure it that way?
A direct acquisition of this size would trigger antitrust review. A license, a hiring wave and a minority investment are separate transactions that individually do not, at least under current enforcement practice.
Is the deal 6 billion or 7 billion dollars?
Six billion for the license, plus a separate 1 billion investment. Some coverage combines them into a 7 billion figure, which overstates what was paid for the technology itself.
What is Model Factory?
Poolside’s internal software for developing and training artificial intelligence models. It is infrastructure tooling rather than a consumer product, and it takes years to build well.
Have other companies done this?
Yes. Google, Meta, Microsoft and Amazon have all used versions of the license and hire structure with smaller startups over the past two years. This is the largest example so far.
Are the terms confirmed?
They are reported rather than company confirmed. The initial scoops came from subscription outlets and were corroborated by Bloomberg, which is strong sourcing but not an official filing.
