Nvidia Pays $6B to License Poolside’s Model-Building Factory

Nvidia has struck an unusually structured deal with AI coding startup Poolside that pushes the chipmaker deeper into the software and model-development layer. According to a letter to Poolside investors obtained by Newcomer and independently summarized by The Information, Nvidia agreed to pay $6 billion for a non-exclusive license to Poolside’s Model Factory technology, invest another $1 billion in the company at a $12 billion pre-money valuation, and offer jobs to 109 employees who worked on Poolside’s model-building systems. Poolside’s three founders are expected to remain with the company, which says the transaction is not an acquisition or an acquihire.

The transaction matters because the asset Nvidia is licensing is not simply a coding model. Poolside describes its Model Factory as the internal platform it uses to train, evaluate and improve foundation models, including orchestration for reinforcement learning from code execution, synthetic-data generation, architecture experiments and automated evaluation across GPU clusters. Poolside’s public materials say experiments that once took weeks to schedule can be run in under an hour through the system. If the reported agreement is executed as described, Nvidia is paying for access to a model-development production system rather than merely buying another finished model.

That distinction fits Nvidia’s broader move up the AI stack. The company remains the dominant supplier of accelerated computing hardware, but it has steadily expanded into model software, inference frameworks, enterprise tooling and developer platforms. A non-exclusive license to a mature model-training system could give Nvidia additional internal capability for building and refining its own open and proprietary models while also improving the software it offers to customers that train on Nvidia infrastructure. The deal therefore looks less like a conventional startup investment and more like an attempt to absorb a repeatable model-development capability without fully absorbing the company that built it.

The reported staffing arrangement is equally important. Nvidia is expected to make job offers to 109 Poolside employees involved in Laguna and the Model Factory, while the founders remain to run an independent Poolside. That structure preserves Poolside as a company but transfers a substantial portion of the technical team to Nvidia. Newcomer reported that Poolside plans to distribute the $6 billion licensing proceeds to investors and employees by the end of 2027. The result is economically similar to parts of an acquisition, but legally and operationally different because the intellectual property license is non-exclusive and Poolside continues operating.

Poolside’s technology is relevant because the startup has been developing open-weight agentic coding models rather than only a hosted coding assistant. Its current Laguna S 2.1 model is described by Poolside as a 118-billion-parameter mixture-of-experts model with 8 billion active parameters and a one-million-token context window. The smaller Laguna XS 2.1 has 33 billion total parameters with 3 billion active. Poolside says its models are trained from scratch using its own infrastructure and reinforcement learning inside its agent harness, making the Model Factory the operational system behind both the models and the experiments used to improve them.

For practitioners, the most significant implication is that the competitive bottleneck in frontier AI is shifting from access to GPUs alone toward the systems that can turn compute into reliable model progress. Training orchestration, experiment scheduling, data generation, evaluation and reinforcement-learning infrastructure increasingly determine how quickly a lab can iterate. Nvidia already sells the hardware on which much of that work runs. Owning or licensing stronger model-development machinery could let it capture more value from the process that happens after customers acquire compute, while also reducing its dependence on outside model labs for software expertise.

The deal also raises a strategic question for Nvidia’s customers. Many leading AI companies depend heavily on Nvidia hardware while simultaneously competing with Nvidia in models, inference software or developer platforms. If Nvidia becomes a stronger model builder and model-training platform operator, those relationships become more complicated. Customers may benefit from better software and integrated tooling, but they may also have to consider whether a critical infrastructure supplier is moving closer to their own product layer. The non-exclusive nature of the Poolside license reduces that concern somewhat because Poolside can continue using and developing its technology.

There are still important uncertainties. Neither Nvidia nor Poolside had published a detailed official transaction announcement in the sources reviewed for this run, so the central financial and staffing terms rest primarily on reporting based on Poolside’s investor letter. The Information separately reported the same core structure, which strengthens confidence, but the agreement’s exact licensing scope, duration, exclusivity boundaries and technical integration plan are not public. It is also unclear how many of the 109 employees will accept Nvidia’s offers or what operational capacity will remain at Poolside after the transfer.

The structure may also attract scrutiny because large technology companies have increasingly used licensing, investment and targeted hiring arrangements that can deliver many of the benefits of an acquisition without buying the entire company. That does not make the transaction improper, but it changes how competition authorities, investors and employees may evaluate it. For Poolside, the arrangement provides major liquidity and fresh capital while keeping the company alive. For Nvidia, it potentially adds a large technical team and model-building infrastructure without taking ownership of all Poolside liabilities and operations.

The next evidence to watch is straightforward. Nvidia or Poolside could publish formal terms, clarify which Model Factory components Nvidia can use, and explain how the transferred team will be organized. Poolside’s product roadmap will show whether the company can continue developing Laguna and its open-weight strategy with a smaller technical organization. Nvidia’s future model releases and developer tooling may reveal where the licensed technology is actually used. Until those details emerge, the material development is the reported scale and structure of the transaction: a $6 billion non-exclusive technology license, a $1 billion investment and a large talent transfer that moves Nvidia further from being only the supplier of AI compute.