Nvidia has agreed to pay $6 billion to license Poolside’s Model Factory, the AI startup’s proprietary system for building and training advanced AI models, while also offering jobs to 109 Poolside employees. The arrangement gives Nvidia access to technology developed around Poolside’s Laguna model and strengthens its ability to develop its own AI models, without formally acquiring the startup. The terms were disclosed in an investor letter first reported by Newcomer.
The deal is structured as a non-exclusive licensing agreement rather than a conventional acquisition. Nvidia is also investing $1 billion in the remaining Poolside business at a $12 billion pre-money valuation, while Poolside’s three founders will remain with the company. The unusually large licensing payment highlights the increasing strategic value of AI model-development infrastructure as companies compete to build increasingly capable models.
Nvidia To License Poolside’s Model Factory For $6 Billion
The centerpiece of the agreement is Poolside’s Model Factory, an internal system used to build, train and evaluate AI models. Nvidia will pay $6 billion for a non-exclusive license to the technology, meaning Poolside will retain ownership and can potentially license the system to other companies.
Poolside developed the Model Factory as part of its work on AI models for software engineering. The system is designed to automate significant portions of the model-development process and was used in the development of the company’s Laguna model.
The deal gives Nvidia access to Poolside’s model-building methodology while allowing Poolside to continue operating as an independent company.
Key Numbers In The Nvidia-Poolside Deal
| Deal Component | Reported Figure |
|---|---|
| Nvidia licensing payment | $6 billion |
| Nvidia investment in remaining Poolside | $1 billion |
| Poolside pre-money valuation | $12 billion |
| Poolside employees joining Nvidia | 109 |
| Poolside founders remaining | 3 |
| License type | Non-exclusive |
| Main technology licensed | Model Factory |
| Model associated with Poolside’s development work | Laguna |
The structure means Nvidia is spending $7 billion across the licensing and investment components, although the two payments serve different purposes. The $6 billion goes toward the technology license, while the $1 billion is an investment into Poolside itself.
109 Poolside Employees Will Move To Nvidia
Nvidia will also offer positions to 109 Poolside employees who worked on the company’s Laguna model and related AI development efforts.
The talent transfer is an important part of the transaction because advanced AI development depends heavily on researchers and engineers who understand the training infrastructure behind frontier models.
Poolside’s three founders are expected to remain with the startup rather than joining Nvidia. This leaves the company with its founding leadership while a large portion of the technical workforce moves to Nvidia.
Workforce Structure After The Deal
| Group | Reported Status |
|---|---|
| Employees joining Nvidia | 109 |
| Poolside founders | 3 remaining |
| Core Model Factory technology | Licensed to Nvidia |
| Poolside company | Continues operating |
| Nvidia access to Model Factory | Non-exclusive |
The arrangement gives Nvidia both the underlying technology and a significant portion of the team that helped develop it. That combination could allow Nvidia to integrate the Model Factory into its own AI research and model-development operations more quickly than if it licensed the software alone.
Nvidia Is Not Formally Acquiring Poolside
Despite the $6 billion price tag and the movement of 109 employees, the transaction is not being described as an acquisition.
Poolside’s investor letter reportedly states that the deal is “not an acquisition and it is not an acquihire.” Instead, the startup is entering into a licensing agreement with Nvidia while continuing as an independent company.
This distinction is important because Nvidia is not obtaining ownership of the entire Poolside business. The startup retains its founders and remaining operations, while Nvidia receives rights to use the Model Factory.
Traditional Acquisition Vs Poolside Structure
| Area | Traditional Acquisition | Nvidia-Poolside Arrangement |
|---|---|---|
| Company ownership | Buyer acquires company | Poolside remains independent |
| Technology | Usually transferred to buyer | Model Factory licensed |
| Employees | Typically transfer with company | 109 offered jobs |
| Founders | May join buyer | Three founders remain |
| License | Not normally the core structure | Non-exclusive |
| Investment in target | Not applicable in same form | $1 billion |
| Poolside valuation | Acquisition price would determine value | $12 billion pre-money |
The structure could allow Nvidia to obtain important AI capabilities while avoiding the full corporate integration associated with buying another company.
Why Nvidia Wants Poolside’s Model Factory
Nvidia already develops its own AI models through its Nemotron family. The company therefore has a direct interest in improving the infrastructure used to train and evaluate models.
Poolside’s Model Factory could provide Nvidia with another approach to model development, particularly around training models for coding and reasoning tasks.
The value of the technology lies not only in an individual AI model. A system that improves how models are repeatedly trained, evaluated and refined can potentially become a reusable infrastructure layer for developing multiple generations of models.
What Model-Development Infrastructure Can Provide
| Capability | Potential Value |
|---|---|
| Model training | Automates and scales training workflows |
| Evaluation | Tests model performance across tasks |
| Reinforcement learning | Helps models improve through feedback |
| Code execution | Allows coding models to evaluate actual results |
| Experiment automation | Speeds up model-development cycles |
| Model iteration | Enables faster testing of new approaches |
Poolside’s approach has focused heavily on software engineering, where models can be trained against coding tasks and evaluated according to whether the resulting code actually works. This creates a more objective feedback loop than relying exclusively on human ratings.
Poolside Says Frontier Model Training Is Becoming A Compute Problem
The investor letter also reportedly argues that the amount of computing power required to build the next generation of frontier AI models is rising dramatically.
Poolside said it could have built a frontier-level model using between 10,000 and 20,000 chips. However, it expects next year’s frontier models to require clusters more than an order of magnitude larger.
The company identified physical data-centre capacity and contracted computing resources as constraints alongside capital.
Poolside’s Reported Compute Argument
| Factor | Poolside’s View |
|---|---|
| Chips needed for a frontier-rivalling model | 10,000–20,000 |
| Future frontier-model requirement | More than 10x larger cluster |
| Key constraint | Compute |
| Other constraints | Data-centre space and contracted capacity |
| Capital | Important but not the only limitation |
The comments underline the changing economics of frontier AI. As models become more capable, simply having enough money to train them may not be sufficient. Companies also need access to huge quantities of accelerators, data-centre space, electricity, networking and cooling infrastructure.
Nvidia Is Becoming More Than A Chip Company
The Poolside transaction also fits into Nvidia’s broader strategy of expanding beyond GPUs.
Nvidia’s dominance in AI computing has been built around GPUs and its CUDA software ecosystem. But the company has increasingly moved into networking, complete AI systems, software, inference and model development.
Owning or licensing technologies that improve model creation could strengthen Nvidia’s position across the AI stack.
Nvidia’s Expanding AI Stack
| Layer | Nvidia’s Role |
|---|---|
| AI accelerators | GPUs and AI computing platforms |
| Networking | High-speed AI data-centre connectivity |
| Software | CUDA and AI development tools |
| AI models | Nemotron family |
| Model infrastructure | Poolside Model Factory license |
| AI systems | Integrated data-centre platforms |
| Developer ecosystem | Tools and services around Nvidia hardware |
This vertical expansion could create a feedback loop: Nvidia sells the hardware needed to train models while also developing the software and models that run on that hardware.
Poolside Gets $1 Billion More Capital
While Nvidia gains access to Poolside’s technology and employees, the remaining Poolside business is also receiving substantial financial support.
Nvidia is investing $1 billion at a $12 billion pre-money valuation. That investment provides capital for the company to continue operating after transferring a large part of its technical workforce and licensing its core Model Factory technology.
Poolside’s investor letter reportedly says it intends to distribute the $6 billion licensing proceeds to investors by the end of next year.
Poolside’s Financial Position Under The Deal
| Financial Item | Amount |
|---|---|
| Nvidia Model Factory license | $6 billion |
| Nvidia equity investment | $1 billion |
| Pre-money valuation | $12 billion |
| Total Nvidia-related commitment | $7 billion |
| Planned distribution of license proceeds | By end of 2027 |
The arrangement gives Poolside substantial resources while preserving the company as an independent entity.
Why The Non-Exclusive License Matters
One of the most notable details is that Nvidia’s license is non-exclusive.
That means Nvidia does not receive sole access to the Model Factory. Poolside can potentially license the technology to other customers, creating an unusual situation in which a $6 billion payment does not give the buyer exclusive control over the underlying system.
The structure also means Poolside could theoretically continue building relationships with other major AI companies.
For Nvidia, the value therefore appears to come from gaining access to the technology and associated talent now rather than preventing competitors from ever using the same system. The companies have not publicly explained why Nvidia considers a non-exclusive license worth $6 billion.
Nvidia’s Recent Deal-Making Strategy
The Poolside arrangement is also notable because Nvidia has recently used structures that combine technology licensing, investment and employee hiring instead of straightforward acquisitions.
The approach can give Nvidia access to specialized technology and experienced teams while leaving the original companies operating independently.
For the AI industry, such arrangements could become increasingly important as regulators pay closer attention to acquisitions involving major technology companies and emerging AI startups.
The Bigger Picture
Nvidia’s $6 billion Poolside deal shows that the competitive race in AI is moving beyond access to GPUs. The infrastructure used to build, train, evaluate and improve models is becoming a strategic asset in its own right. By licensing Poolside’s Model Factory and bringing 109 employees into Nvidia, the chipmaker is strengthening its position in the model-development layer while continuing to build its own Nemotron family.
The unusual structure is equally significant. Nvidia is not buying Poolside outright, and the Model Factory license is non-exclusive. Instead, Nvidia is combining a large technology payment with a targeted talent transfer and a separate $1 billion investment. The deal demonstrates how valuable specialized AI infrastructure and expertise have become as frontier-model development requires increasingly large amounts of computing capacity.
Looking Ahead
The immediate question will be how Nvidia integrates Poolside’s Model Factory and the incoming engineers into its existing AI research and Nemotron efforts. If the technology can improve model training, evaluation or coding performance, it could strengthen Nvidia’s ability to develop models optimized for its own hardware ecosystem. At the same time, Poolside will need to demonstrate that the remaining company can build a sustainable business after transferring much of its technical workforce to Nvidia.
The transaction could also influence how other major technology companies acquire AI capabilities. Instead of buying entire startups, companies may increasingly pursue combinations of licensing agreements, strategic investments and talent transfers. As AI development becomes more expensive and specialized, control over model-building infrastructure may become almost as strategically important as control over the chips used to train the models.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
