Nvidia's $7 billion Poolside deal targets DeepSeek, OpenAI dominance

Nvidia's $7 billion Poolside deal pushes the chipmaker into open-weight AI model development, directly challenging Chinese rivals DeepSeek and Kimi K3 while complicating ties with customers OpenAI and Anthropic.

"The core principle behind the deal is to ensure that future artificial general intelligence does not become a technology controlled behind closed doors by a handful of companies, but can be built openly by a broader community," Poolside co-founders Eiso Kant and Jason Warner wrote in a letter to shareholders.

The arrangement includes a $6 billion technology licensing fee, a $1 billion equity investment at a $12 billion pre-money valuation, and offers to more than 100 Poolside engineers who will join Nvidia's Nemotron open-weight model program. Poolside's management team, including Kant, Warner, and Chief Operating Officer Margarida Garcia, will remain independent and continue undisclosed research.

Nvidia shares fell about 1 percent last Friday to close at $214.72. Analysts' consensus rating remains a strong buy with an average price target of $301.82, implying roughly 40 percent upside. The deal extends Nvidia's dominance from GPU hardware into the AI model layer, a move that could reshape competitive dynamics across the AI stack.

Why Open-Weight Models Became the U.S.-China AI Battleground

Open-weight models allow developers to download and modify model weights free of charge, with usage costs typically far lower than closed models and greater customization flexibility. By contrast, the flagship models of OpenAI, Anthropic, and Google do not publicly release full model weights or source code.

This gap became a critical variable in the U.S.-China AI race after DeepSeek released its first low-cost open-weight model in January 2025, triggering sharp volatility in U.S. tech stocks. Venture capitalist Marc Andreessen described it as AI's "Sputnik moment." Since then, China has seen a wave of open models, including Moonshot AI's Kimi and Z.AI's GLM series, narrowing the gap with the most advanced U.S. AI technology to a matter of months.

Nvidia CEO Jensen Huang has long supported the open-weight approach and accelerated his push over the past six months. In March, the company launched the Nemotron Coalition, bringing together Mistral, Thinking Machines Lab, and Perplexity, with members sharing data, expertise, and computing resources. Last month, Huang published an open letter titled "Open Weights and American AI Leadership," arguing that U.S. AI leadership depends on building open models across the industry rather than a single most-advanced model.

The Poolside deal follows Nvidia's recent $20 billion licensing and hiring arrangement with inference-chip designer Groq. For Poolside, the transaction provides substantial capital and a powerful commercial relationship as it moves from building coding agents toward competing in the open-weight model market. The startup had failed to raise $2 billion in late 2025 to pay for a computing cluster equipped with 40,000 GB300 chips, a setback that prompted the Nvidia deal.

The deal also complicates Nvidia's relationships with major AI labs. OpenAI and Anthropic are not only Nvidia's most important AI chip customers but also the primary developers of closed AI models. Both companies are pursuing custom chip strategies — Anthropic recently recruited a chip expert who led Google's TPU development, while OpenAI is partnering with Broadcom on inference chips with deployment expected by the end of this year.

Memory Shortage Forces Nvidia to Raise Server Prices

Separately, Nvidia has informed major customers it could raise prices for AI server systems scheduled for delivery early next year by more than 15 percent. The potential increases would apply to both its Vera Rubin and Grace Blackwell systems, driven by a shortage in the memory chip market led by Samsung Electronics, SK hynix, and Micron.

Explosive demand for AI data centers has pushed up prices for high-bandwidth memory and server DRAM, directly raising the cost of manufacturing AI servers. Samsung ranked first in the global DRAM market in the second quarter with a 39 percent share, showing its market influence. As next-generation systems require ever larger amounts of memory, the pricing power of memory suppliers has strengthened considerably.

Even Nvidia, known for industry-leading profitability, appears unable to absorb the full impact of rising memory costs. The price increases reinforce the profitability outlook for the memory industry while adding to the cost burden for Big Tech companies operating large-scale data centers. Amazon, Microsoft, and Google continue to buy Nvidia products in large volumes while increasingly pursuing their own chip strategies, creating a complex web of competition and cooperation across the AI supply chain.

Nvidia's expansion into model development and its server price increases together show a company using its dominant position in AI hardware to extend influence across the entire AI technology stack. The question for investors is whether Nemotron can achieve sufficiently broad adoption for Nvidia to gain meaningful influence in another segment of the AI market without creating excessive friction with its largest chip customers.

This article is for informational purposes only and does not constitute investment advice.