Nvidia Strikes $7 Billion Poolside Deal to Build Open-Weight AI Models Against DeepSeek and OpenAI

Nvidia's $7 Billion Bet on Open-Weight AI Models

Nvidia has structured a multibillion-dollar arrangement with AI startup Poolside, paying $6 billion to license Poolside AI model software while investing an additional $1 billion at a $12 billion pre-money valuation1

1

3

3

. The Nvidia Poolside deal marks a four-fold jump from Poolside's 2025 valuation of $3 billion and represents the chipmaker's most aggressive push yet into developing its own AI models rather than merely supplying the infrastructure3

3

.

The strategic investment in Poolside includes hiring 109 of the startup's employees, who will work on Nvidia's Nemotron project to develop open-weight AI models3

3

4

4

. According to Poolside CEO Eiso Kant, fewer than 70 people built the company's model, with fewer than 115 working across engineering and research total1

1

. The three founders are staying with Poolside, which plans to distribute the $6 billion licensing fee to investors by the end of next year1

1

.

Not an Acquisition: Nvidia's Licensing Strategy

Poolside's investor letter emphasizes this is "not an acquisition and it is not an acquihire," though the deal follows a template Nvidia has deployed twice before1

1

. Nvidia previously paid Groq $20 billion for its inference technology and hired its top engineers, with Groq remaining independent and later raising $350 million at a $3.5 billion valuation with Nvidia participating1

1

. The company also structured a similar $900 million arrangement with hardware startup Enfabrica1

1

5

5

.

These deals allow Nvidia to secure staff and expertise without triggering regulatory scrutiny that traditional acquisitions bring1

1

. The non-exclusive nature of the license means Poolside retains the ability to license the same AI model-building software elsewhere1

1

3

3

.

What Nvidia Is Actually Licensing

The $6 billion license covers Poolside's Model Factory, the system the startup used to build its models1

1

. Poolside began as a developer of coding AI agents before moving into data centers and releasing its open-source Laguna model on Nvidia server chips1

1

. The Laguna S model was positioned as a Western alternative to Chinese AI models from DeepSeek and competitors2

2

4

4

.

Nvidia's goal is to use the licensed technology and incoming engineers to build some of the most capable open-weight AI models available, positioning itself to compete with OpenAI, Anthropic, and Chinese models like DeepSeek and Kimi K32

2

4

4

. Open-weight AI models are generally far cheaper to run and allow developers to make customizations to suit their needs3

3

.

Why Poolside Needed This Deal

The arrangement came together after Poolside failed to raise capital for critical computing resources. "At the end of last year, we had a 6 week window in which to raise $2 billion dollars to pay for a 40,000 GB300 cluster coming online in January," the investor letter revealed. "We didn't close it in time, and we lost the cluster"1

1

2

2

.

Poolside's letter explained the startup had been "directionally correct in a race where capital requirements went vertical." The company estimated it could have built a frontier-rivaling model with 10,000 to 20,000 of those chips, but next year's frontier models need "far more than an order of magnitude larger cluster." The constraint "is not only capital, it is physical data center space and contracted compute"1

1

.

Jensen Huang's Vision for Open AI Ecosystem

Nvidia CEO Jensen Huang has emphasized that US AI leadership will come through an open ecosystem rather than frontier dominance. "Our AI leadership will be judged not by one frontier AI model, but by whether the United States builds a strong, open ecosystem that diffuses into every sector," Huang wrote in a recent letter3

3

.

Huang argued that "open weights expand access to the AI economy" and that "relying solely on closed models is not inherently safe: they can be breached, misused, or fail in ways that outsiders cannot detect." He added that "concentrating advanced AI capabilities behind a small number of closed models compounds that risk"3

3

. China currently leads the open-weight space, with Alibaba's Qwen models downloaded more than 3 billion times over the past six months, placing it ahead of Google, OpenAI, Meta and other rivals3

3

.

Nvidia's Strategic Hedge Against GPU Demand Collapse

By earmarking $7 billion for Poolside, Nvidia is establishing itself as a potential buyer of last resort for its own GPUs should demand suddenly collapse5

5

. The deal allows Nvidia to redirect a significant proportion of its GPUs to internal Nemotron-related workloads if necessary, transforming the GPU giant from merely supplying infrastructure to competing directly with companies like OpenAI and Anthropic5

5

.

This hedging strategy appears timely. OpenAI CEO Sam Altman recently conceded he was wrong about his aggressive AI adoption timeline, admitting the economy will incorporate AI at a much slower pace than previously believed5

5

. Nvidia unveiled Nemotron 3.5 Lightning in early August, claiming the model generates tokens at a rate 4x faster than similar-sized models, though it only speeds up actual agentic tasks by 30 percent5

5

. Poolside's technology could help bridge these performance gaps.

What's Next for Both Companies

Poolside employees joining Nvidia will reportedly work on the company's Nemotron project to develop open-weight AI models3

3

. Nvidia launched its Nemotron Coalition in March, bringing together open-model developers to share expertise, data and computing resources4

4

. The company is also working towards a trillion-parameter open model and recently halved a $250 billion commitment to OpenAI1

1

.

Poolside's founders say they are "not ready to share the updated vision" for what remains of the company1

1

. Poolside Infrastructure Company, spun out in January, is building a 1.2GW data center in Texas and appointed a CEO two months ago and a CFO this week, both before Friday's announcement1

1

. The investor letter suggests human-level AI capability will be "fully commoditized by open source models," pointing toward a future where open-weight AI models become the standard rather than the exception1

1

.