Jensen Huang isn’t building a semiconductor company. He’s building the Federal Reserve of AI, a system that keeps the AI economy afloat, both on the supply and the demand side.
Nvidia is not just building GPUs. They are engineering demand for their core product at an unprecedented scale. And they are doing it by propping up every part of the AI ecosystem, infra providers, model builders, tool providers, app builders, and even the next frontier.
Consider the bets Nvidia has placed in the last three years.
On the model side:
OpenAI (participated in the $6.6B round, then committed $100B in a joint deployment deal).
Anthropic ($10B strategic investment).
xAI ($2B in a $20B round).
Mistral AI (Series B and Series C, followed by an 18,000 Blackwell GPU deployment deal in France).
Cohere (Series D).
On the cloud and infrastructure side:
CoreWeave (backed at Series A in 2023, now public, with Nvidia still a major shareholder, plus a $6.3B capacity backstop through 2032).
Lambda (Series D).
Together AI (Series B).
Crusoe (growth round).
And just this month, Helix Digital Infrastructure, a new KKR-led venture with $10B+ in committed capital to build hyperscale data centers, power generation, and connectivity infrastructure for AI workloads. Nvidia is both an anchor investor and strategic partner for Helix.
On the networking and silicon side:
Enfabrica, a next-gen interconnect chip startup. Nvidia invested at Series B, then acquired the team and technology for $900M in 2025. Because at hundred-thousand-GPU scale, the chip is only as fast as the network around it.
On the physical AI side:
Figure (Series B, then Series C).
Wayve (participated in a $1.05B round, with a reported $500M strategic follow-on).
Nuro (growth round).
Waabi (Series B).
On applications:
Perplexity (backed across multiple rounds from Series B onward).
Hippocratic AI (Series B).
Cursor (Series D).
Poolside (Series B, with up to $1B follow-on reported).
Kore.ai (growth round).
Scale AI (participated in a $1B round).
Runway (participated in a $308M round).
In total: 16 investments in 2022. 54 in 2024. 67+ deals in 2025 (not counting the $100B OpenAI commitment). Over $40B in equity deployed in the first five months of 2026 alone.
Every single category above is a node in the same network, one where the underlying resource being consumed is Nvidia silicon. This investment portfolio is not about diversification. It is a demand manufacturing machine.
The CoreWeave deal is the clearest illustration. Nvidia backstopped CoreWeave with a $6.3B guarantee through 2032, committing to purchase any cloud capacity CoreWeave cannot sell to other customers. Nvidia is guaranteeing the revenue floor of an infrastructure company that exists primarily to rent Nvidia GPUs. That backstop allows CoreWeave to access debt financing at terms it could never obtain independently. It then uses this facility to buy more Nvidia chips. Nvidia records that as revenue.
Nvidia underwrites the economics of its own customers, allocates cloud credits through that infrastructure, and receives equity in return.
For every $10B Nvidia invests in a company like OpenAI, it is estimated to generate approximately $35B in GPU purchases or lease payments. A 3.5x return on deployed capital, denominated in chip revenue rather than financial returns. That reframes what “investment” means for Nvidia entirely.
The physical AI bet is the longest and least appreciated.
LLM training is a known demand driver. Physical AI, robots that perceive, reason, and act in the real world, requires orders of magnitude more compute per inference cycle than language models. Nvidia is capitalizing Figure, Wayve, Nuro, and Waabi now, before this space is even understood by most people.
Nvidia generated $97B in free cash flow last fiscal year. It is deploying a portion of that to shape which AI architectures win, which infrastructure providers scale, and which physical AI bets get funded.
Every company in AI either buys from Nvidia or is funded by Nvidia, in some way. Sometimes both. That’s not a vendor. That’s a sovereign.