Buying the Team Without Buying the Company
NVIDIA is paying $6 billion to license Poolside's "Model Factory" software and hire more than a hundred of its staff. Poolside, the company, keeps its name, its cap table, and its remaining investors. NVIDIA gets everything else.
That distinction sounds like a technicality until you notice it is the same technicality Microsoft used on Inflection and Amazon used on Adept. Three different buyers, three different targets, one identical structure: license the technology, hire the people who built it, leave the corporate shell standing. The Mixture of Experts panel named the pattern directly on the August 28 episode, and once you see three instances you stop calling it a coincidence and start calling it a playbook.
The playbook exists because of what it avoids. A straight acquisition above a certain size triggers a pre-merger antitrust filing in the US, a waiting period, and a review that can block the deal or force concessions. A licensing agreement paired with individual job offers is a different transaction under that law. NVIDIA gets the software. NVIDIA gets the researchers who wrote it. NVIDIA gets none of the scrutiny a $6 billion purchase of a direct AI competitor would draw. The outcome is identical to an acquisition. The paperwork is not.
What gets left behind is worth sitting with. Inflection kept its name and its remaining investors after most of its research team moved to Microsoft in 2024. Adept did the same after Amazon hired its founders and key researchers later that year. Neither company disappeared. Neither company, in any way that mattered to the people who had built it or the customers depending on it, survived either. A cap table and a logo remained. The team and the technology that made the name worth anything left for the buyer that structured the deal to avoid buying the company outright.
NVIDIA running this play is the detail that should get more attention than it has. Microsoft and Amazon absorbing AI teams through licensing deals fits a pattern of hyperscalers extending their reach into the model layer. NVIDIA is the chip supplier those same hyperscalers depend on, moving up the stack using the identical mechanism, at a moment when NVIDIA's dominance in AI infrastructure already draws more antitrust attention than either Microsoft's or Amazon's cloud businesses do on their own. If any company in this market had the most reason to avoid a deal that reads as a straight acquisition, it is the one already sitting closest to a monopoly finding.
Antitrust regulators noticed the Microsoft-Inflection version of this well enough to start asking whether AI partnerships structured this way function as acquisitions in every way that matters except the filing. That scrutiny hasn't stopped a third version of the same structure from closing. It has confirmed the pattern is recognized, and still working.
Here is where this stops being a story about three specific deals and starts being a question every enterprise evaluating an AI vendor should be asking. A partnership with a smaller, differentiated AI company carries the usual risks: runway, roadmap, execution. It now carries one more. The team and the technology an enterprise is depending on can be extracted by a larger buyer through a transaction structure specifically designed not to surface as an acquisition until it is done. No merger announcement. No regulatory filing to search for in advance. No customer notification requirement, because on paper nothing was bought. The first sign a vendor relationship is about to change may be a press release confirming it already has.
The practical tell worth watching is proximity. A small AI vendor already running deep compute or distribution ties with one hyperscaler is a company whose most valuable asset, its own research team, sits one licensing conversation away from working somewhere else. That risk sits upstream of every other diligence question a technology leader normally asks about a vendor, and it is the one due diligence checklist rarely accounts for, because it was built for acquisitions that show up on a filing before they show up on a departure announcement.