Nvidia is paying $6 billion (about 950 billion yen) to license the model-development platform built by AI coding startup Poolside, and it has extended job offers to 109 people who worked on it. Nvidia is also investing $1 billion at a $12 billion pre-money valuation. An investor letter, however, insists the arrangement is neither an acquisition nor an acquihire. The playbook of obtaining technology and talent without buying the company has just been used at a much larger scale.
What $6 Billion Actually Buys
The license covers what Poolside calls its Model Factory, the system the startup used to build Laguna, its family of coding models. Nvidia is therefore paying for the machinery that produces models rather than for a finished model. The agreement is non-exclusive, so Poolside can continue using the platform itself.
Alongside the license, Nvidia is investing $1 billion (about 159 billion yen) at a pre-money valuation of $12 billion (about 1.9 trillion yen). The 109 engineers who worked on Laguna have received offers from Nvidia, while the three co-founders are staying and Poolside continues to operate as an independent company.
Exchange rate: 1 USD = 159 JPY (as of August 21, 2026)
Why the Letter Insists It Is Not an Acquisition
The deal surfaced when Newcomer reported on a letter sent to Poolside investors. That letter states plainly that the transaction is not an acquisition and not an acquihire, meaning a purchase driven primarily by the desire to hire a team.
The denial is deliberate. Buying a company outright invites review by competition regulators. Split the same outcome into three parts, licensing for the technology, individual hiring for the people, and a minority stake for the capital, and the corporate entity never changes hands. The acquiring side ends up with much of what it wanted while the procedural burden shrinks considerably.
Poolside plans to distribute the $6 billion to its investors by the end of 2027. Shareholders get their money back without an IPO or a sale.
The Third Deal After Groq and Enfabrica
This is not the first time Nvidia has used the structure. It reached a roughly $20 billion (about 3.18 trillion yen) arrangement with AI inference chip company Groq and a deal worth more than $900 million (about 140 billion yen) with networking silicon startup Enfabrica, both combining technology licensing with the absorption of staff. Poolside makes three.
Lined up together, the figures show how the cash Nvidia has accumulated from GPU sales is being redirected into acquiring things without acquiring companies. A chip maker is using its balance sheet to lock down everything that sits on top of the chip.
Nvidia's Own Models Now Sit Beside Its Customers
The other detail worth noting is that Nvidia develops its own open model family, Nemotron. Bringing in an external platform for building models reads naturally as an effort to strengthen that line.
The consequence is that Nvidia ends up competing on the same ground as companies buying its GPUs. It sells the chips and also competes with the models that run on them. How the company manages that distance from its customers will be tested from here.
Coding models have been treated as one of the areas where AI spending translates most directly into measurable results. Owning a strong platform for producing them also supports the underlying business of selling more GPUs. The $6 billion can be read as the price of buying that capability in one move.
Summary
Nvidia is paying Poolside $6 billion to license its model-development platform, offering jobs to 109 engineers, and investing another $1 billion. It takes the technology, the people, and a stake, but not the company. As the third deal after Groq and Enfabrica, it shows that the acquisition that is not an acquisition has become a standard instrument in AI.
