$40 billion. That is the rough amount Nvidia has committed to AI companies so far in 2026, not as chip revenue, but as money flowing back into the ecosystem that buys, powers and connects its GPUs.
According to figures cited by TechCrunch on May 9, OpenAI alone accounts for $30 billion of that total through a deal completed in late February. The rest is spread across several billion-dollar public-market investments and roughly two dozen private rounds. Wedbush analyst Matt Bryson described the logic to CNBC as a way to deepen Nvidia's moat.
The market understands both sides of the story. The structure looks uncomfortable, but as long as the cycle keeps turning, few participants want to stop it.
What sits inside the $40 billion
The disclosed deals point to a pattern. OpenAI takes capital and remains Nvidia's largest GPU customer. Corning can use up to $3.2 billion to expand optical fiber capacity for AI data centers. IREN can redirect mining infrastructure toward GPU data centers. In each case, the investment strengthens a part of the chain that ultimately needs Nvidia chips.
Nvidia made 67 venture deals in all of 2025. In the first five months of 2026, the pace has already moved close to 25 private rounds, alongside major public-market bets and the OpenAI commitment. This is no longer routine corporate venture activity.
In plain terms, Jensen Huang is using money earned from the AI boom to support the balance sheets of customers and suppliers that can buy, house or connect more Nvidia hardware.
Why circular investment worries investors
Circular financing is not new, but it is usually treated as a warning sign. The concern is that the same money can appear to move through several companies, making revenue and growth look stronger even when external demand has not expanded by the same amount.
The historic comparison is the telecom bubble around 2000, when equipment vendors financed carriers that then bought the vendors' equipment. Revenue looked strong until end demand failed to keep up.
Nvidia's version is not identical: it is buying equity rather than simply lending to customers. But the core risk is similar. If real demand from consumers and enterprises trails the speed of AI infrastructure expansion, one weak link can pressure the whole chain.
The OpenAI investment is the most sensitive example. Nvidia invested $30 billion in late February; OpenAI then announced larger GPU procurement plans in March and April. Outsiders cannot prove causality, but financially the shape resembles a loop.
What Huang is really betting on
Nvidia's argument is that the constraint is no longer only GPU supply. The bottleneck is the surrounding infrastructure: power, cooling, optical links, data-center sites, land and grid access.
If any of those pieces fail, shipped GPUs cannot run at full value. So Nvidia is choosing to finance the missing pieces directly: fiber capacity at Corning, data-center capacity at IREN, and demand visibility at model companies such as OpenAI.
That is the moat Bryson was describing. Nvidia is not just selling chips; it is assembling a compute system that works best inside Nvidia's own ecosystem.
If the loop holds, rivals such as AMD, Broadcom, Google's TPU program and Amazon Trainium face a tougher path, even with competitive silicon, because the supply chain has already been capitalized around Nvidia.
The other side of the trade
The awkward question is valuation. Nvidia's multiple, return on capital and cash-flow story are built on the idea that it is a chip company. If more earnings are recycled into customers and suppliers, investors may ask whether Nvidia is becoming a private-equity layer for the AI compute chain.
Those two models are priced differently. A high-growth chip company can command a very different multiple from an investment vehicle. If the market ever shifts the category, even a $3.5 trillion valuation could come under pressure.
For now, the stock market has largely looked past the issue. Nvidia shares reached a new high in early May. But the fact that analysts are now using the phrase circular investment in research notes shows that the mainstream narrative is starting to loosen.
How far can it go?
Huang's 2026 move is clear: use $40 billion to shape both demand and supply for AI compute, keeping future buyers and capacity within Nvidia's orbit.
The structure works as long as outside growth outruns the loop itself. If growth slows, investors will revisit how much of the demand was organic and how much was financed by the same ecosystem.
The next test is paid adoption at OpenAI, Anthropic and other leading model companies. If enterprise usage grows as expected, the cycle can continue. If it does not, Nvidia may have to confront a harder question: how much of the money it invested is returning as durable chip demand?
Huang is betting on the first outcome. The real answer may arrive in 2027.
Sources: CocoLoop, TechCrunch, CNBC