Nvidia‘s (NVDA 0.27%) announcement that it will invest up to $100 billion in OpenAI is being hailed by the company as a massive bet on the future of artificial intelligence (AI). Still, investors should take a closer look at what is really going on here. The money OpenAI receives will ultimately be plowed right back into Nvidia hardware, mostly through Oracle‘s cloud buildout, where the two companies recently signed a massive $300 billion deal.

OpenAI plans to deploy Nvidia systems that need 10 gigawatts of power, which is equal to roughly 4 million to 5 million graphics processing units (GPUs). If that sounds like a lot, it is, as it’s about the same total number of GPUs that Nvidia will ship this year. The first $10 billion of Nvidia’s investment will be deployed as soon as the first gigawatt of capacity is up, and the rest will be rolled out in stages as new data centers come online.

Image source: Getty Images.

Circular financing

On paper, the OpenAI investment helps secure billions of dollars in future demand. But it’s worth remembering that Nvidia is now helping finance one of its biggest customers to keep buying its chips. This is called circular financing.

Nvidia is essentially funding its own demand. This is exactly what Cisco Systems (CSCO -0.93%) did during the internet bubble, when it provided credit to telecoms so they could buy more Cisco routers. Those sales looked great — until the capital dried up and the entire market collapsed.

This is also a defensive move by Nvidia. More and more of Nvidia’s largest customers are designing their own custom AI chips. Alphabet has its TPUs, Amazon has Trainium and Inferentia, and Microsoft is working on its own chip. OpenAI itself has been developing custom chips to bring its costs down, and before this announcement, it placed a $10 billion order with Broadcom for custom chips to be delivered next year.

This is the same threat that Nvidia saw play out in crypto, where ASICs (application specific integrated circuits) displaced GPUs for Bitcoin mining. Nvidia doesn’t want to see that happen again. By investing in OpenAI, it’s trying to keep one of its biggest customers locked into the Nvidia ecosystem.

This also comes at a time when the market is shifting more toward inference, where Nvidia’s moat is much smaller. Training large language models (LLMs) is where Nvidia’s CUDA software platform shines. However, inference isn’t as complex and doesn’t require the same deep software integration. That’s why hyperscalers (owners of massive data centers) are so motivated to build custom chips.

Inference is also a continuous cost, so the economics of cost per inference start to dominate the discussion. That’s why Nvidia also took a $5 billion stake in Intel and announced a collaboration on AI processors, as it’s also trying to stave off Advanced Micro Devices in the inference market and keep its grip on this next phase of AI computing.

Is this a house of cards?

There’s no question that Nvidia is in a dominant position right now, and the OpenAI deal only strengthens its near-term outlook. But its OpenAI investment clearly looks like a defensive move that adds risk. When Cisco used circular financing during the internet boom, it looked brilliant, until the customers it was funding went bust.

Both Nvidia and OpenAI are better positioned, but the principle is the same: Nvidia is using its balance sheet to keep demand high. That works as long as the AI boom keeps running, but it makes the company more exposed if spending slows or if hyperscalers switch to cheaper solutions.

Nvidia remains the key player in AI infrastructure, but this deal is a reminder that its growth isn’t risk-free. A lot of Nvidia’s success is now riding on an unprofitable company that is bleeding massive amounts of cash that it is financing. OpenAI hasn’t actually proven yet that it has a great business model, and if it fails, this becomes a house of cards that collapses onto Nvidia.