On Friday morning, May 9, CNBC published a number that should have surprised people on a normal week and did not even crack the day's top three AI stories. Nvidia has committed more than $40 billion in equity investments to AI companies in the first 18 weeks of 2026.
That is not its revenue. That is not its capital expenditure. That is what Jensen Huang's company has handed back to its own customers and prospective customers, in the form of equity checks, in 134 days.
The single biggest line item is a check to OpenAI. According to CNBC, $30 billion went into a single late-February investment in OpenAI, the customer that put Nvidia at the center of the generative AI economy four years ago.
The rest, roughly ten billion dollars, is spread across seven multi-billion-dollar deals with publicly traded companies and about two dozen private startup rounds. The most recent two, both disclosed in early May, are commitments of up to 3.2 billion dollars in Corning, the New York glassmaker whose specialty optical fiber is now a load-bearing component of AI datacenters, and up to 2.1 billion in IREN, an Australian data-center operator that began life as a Bitcoin miner.
The Math Looks Different When You See It Stacked
A 40 billion dollar deployment in 18 weeks works out to roughly 298 million dollars per day, including weekends. For comparison, that is more than what most Fortune 500 companies generate in monthly revenue. It is more than what Twitter was valued at in 2013. It is roughly eight times what Anthropic raised at its first 4.5 billion dollar valuation, in pure equity, paid in less than 130 trading sessions.
The 67 venture deals Nvidia did in 2025 already made it one of the most active corporate AI investors on the planet. According to FactSet, the company has now participated in about two dozen private startup rounds in 2026 alone, on top of the public-market positions. The pace, in other words, has roughly doubled.
| Investment | Amount | Type | Recipient | Timing |
|---|---|---|---|---|
| OpenAI | $30 billion | Private | AI lab, top customer | Late February 2026 |
| Corning (NYSE: GLW) | Up to $3.2 billion | Public market | Specialty glass / optics | Early May 2026 |
| IREN (NASDAQ: IREN) | Up to $2.1 billion | Public market | AI data centers | Early May 2026 |
| 5 other public-company stakes | Multi-billion (each) | Public market | Various AI infra | 2026 to date |
| ~24 private rounds | Undisclosed totals | Private | AI startups (per FactSet) | 2026 to date |
| Reference: 2025 venture activity | 67 deals | Private | AI startups | All of 2025 |
The pattern matters because Nvidia's most lucrative customers are also its most expensive ones to keep. OpenAI's compute appetite is now measured in gigawatts. Corning's glass goes into the optical interconnects that Nvidia's largest customers, including OpenAI, use to wire together GPU clusters. IREN runs the kind of build-it-and-power-it datacenter that companies racing to deploy Nvidia's Blackwell and Rubin chips need yesterday. Each check buys equity. Each check also tightens the loop between Nvidia's chip sales and the customer's ability to absorb those chips.
The Critics Have a Word for It
The pattern has a name in equity-research circles. They call it the circular deal.
Wedbush Securities analyst Matthew Bryson, quoted by CNBC on May 9, said Nvidia's behavior falls "squarely into the circular investment theme." Bryson's framing, however, was not entirely critical. He suggested that if the investments work, they could help Nvidia build "a competitive moat." Money flows out of Nvidia, into a customer, and back into Nvidia in the form of GPU orders. The customer scales. Nvidia scales with it. The customer becomes harder to dislodge by the time AMD or a custom-silicon competitor arrives.
The circularity question becomes sharpest when you map the flows. Nvidia's 30 billion dollars went into OpenAI in late February. In April, OpenAI raised 122 billion dollars at an 852 billion dollar valuation, with Amazon putting up roughly half. OpenAI uses that money to buy compute. A meaningful share of that compute, the share that runs on Nvidia silicon, returns to Nvidia in revenue.
In September 2025, the two companies had discussed an even larger arrangement: up to 100 billion dollars from Nvidia in exchange for OpenAI deploying 10 gigawatts of Nvidia-powered systems. That deal never landed in its original form. OpenAI pivoted to a partner-led infrastructure strategy with Oracle, Microsoft, and Amazon.
The 30 billion dollar February check is what landed instead.
The Bull Case Is Just as Loud
Altimeter Capital CEO Brad Gerstner has predicted Nvidia could become the world's first 10 trillion dollar company. He made that call this month, with Nvidia's market capitalization sitting at 5.23 trillion dollars and the stock up 83% over the prior twelve months.
The thesis is not subtle. Nvidia controls the chip every frontier lab needs. It now invests in the companies that buy those chips. It now sits in the cap tables of the public companies that build the infrastructure those chips depend on. Each leg of the table reinforces the others. The company has, in effect, vertically integrated the AI economy without acquiring any of the participants.
The bull case has internal evidence to support it. Nvidia told investors at GTC in March that it expects to generate at least one trillion dollars in cumulative revenue from its Blackwell and Rubin chips through the end of 2027. The equity commitments help ensure the customers exist to absorb that revenue.
How The Numbers Stack Up
What The Counter-Argument Sounds Like
The circular-deals critique is not the only objection in circulation. Two more carry weight on Wall Street and in policy circles.
The first is concentration risk. With 30 billion dollars of Nvidia's 40 billion sitting in a single private company, any pause in OpenAI's growth, regulatory action, or eventual IPO mark-down would directly hit Nvidia's balance sheet at a scale that no historical chipmaker has ever absorbed in a single equity position. CNBC's reporting did not include the structure of the stake, the lock-up terms, or the accounting treatment. Wall Street analysts are still asking those questions.
The second is accounting optics. When a chip vendor invests in a customer that then buys chips from that vendor, the GAAP treatment is straightforward, but the appearance to outside observers is murkier. The SEC has not, to date, raised concerns about Nvidia's specific structures. It has, however, signaled increased attention to round-tripping disclosures across the AI sector. The SEC's stance could change quickly if the IPO market opens for one of Nvidia's portfolio companies and the disclosures arrive in front of public-market investors.
There is also a quieter dissenting view inside Nvidia's own investor base. Some institutional shareholders have started asking, on earnings calls, whether the company's investment portfolio should be carved out and reported separately, the way Berkshire Hathaway reports its insurance float versus its operating businesses. Nvidia has not provided that breakdown.
What It Means For Practitioners
For ML engineers and data scientists, the $40 billion number is less abstract than it looks.
It means the companies they buy compute from, deploy models on, and build infrastructure with are increasingly the same entity, three layers up the cap table. The startup using Nvidia inference chips, hosted on a datacenter Nvidia partially owns, training a model whose lab Nvidia partially owns, is a less hypothetical configuration than it sounds.
The downstream effect is on supplier independence. Smaller frontier labs that want to ship without taking Nvidia capital, like Mistral or DeepSeek, are now visibly on a different track. Mistral borrowed 830 million dollars to buy chips outright rather than take equity from a chip vendor. DeepSeek trained its trillion-parameter model on smuggled Nvidia hardware precisely because it could not get into the official Nvidia-investment ecosystem. The bifurcation is no longer just about geopolitics. It is about who is inside the Nvidia cap-table flywheel and who is outside it.
For practitioners building on top of these stacks, the question worth asking on any vendor RFP this quarter is whether the vendor is taking Nvidia equity. The answer increasingly correlates with the vendor's ability to ship at scale, and with the strategic optionality it will have if Nvidia's pace of investment ever slows.
Nvidia is no longer just selling shovels in a gold rush. In four and a half months, it has put 40 billion dollars of equity to work in the people doing the digging, with 30 billion of that in one customer alone. The structure works as long as AI demand keeps growing and OpenAI does not stumble. If either condition fails, the circle gets very visible very fast.
The cleaner read on the year so far is this. Nvidia is using its market position and its 5.23 trillion dollar balance sheet to write the equity checks that lock the rest of the AI economy into its silicon. As one Wedbush analyst put it, the circle is real. It is also, for now, holding up.
Sources
- Nvidia has already committed $40B to equity AI deals this year (TechCrunch, May 9, 2026)
- Nvidia embraces AI investor, topping $40 billion in equity bets 2026 (CNBC, May 9, 2026)
- Nvidia's AI Investment Bets Top $40 Billion In 2026, Led By OpenAI Stake (Benzinga, May 9, 2026)
- Nvidia Commits Over $40B to AI Equity Deals in 2026 Led by $30B OpenAI Investment (The AI Insider, May 11, 2026)
- NVIDIA tops $40bn in AI equity bets in 2026, anchored by $30bn OpenAI investment (The Next Web, May 9, 2026)
- Nvidia's AI empire: A look at its top startup investments (TechCrunch, Jan 2, 2026)
- Jensen Huang says Nvidia is pulling back from OpenAI and Anthropic, but his explanation raises more questions than it answers (TechCrunch, Mar 4, 2026)