Nvidia Reveals Its Last Private Bet on OpenAI in 2026

Jensen Huang just told a room full of investors that Nvidia is finished buying its way into the two companies that helped make it the most valuable business on Earth. Speaking at the Morgan Stanley Technology, Media and Telecom conference, the Nvidia chief executive said the company's recent $30 billion commitment to OpenAI's funding round and its $10 billion stake in Anthropic are "probably the last" private checks it will write to either lab. The reason is not a loss of conviction. It is the opposite. Both labs are sprinting toward public listings, and once a company files to go public, the early-stage investment door swings shut for good.

What Actually Happened

Huang framed the comments as a matter of timing rather than strategy. Nvidia's $30 billion position in OpenAI and its $10 billion position in Anthropic were struck while both companies were still private, when a strategic backer could buy equity at negotiated terms. That window is now closing. Anthropic moved first, filing confidentially with the Securities and Exchange Commission on June 1 in a clear signal of intent to go public. OpenAI, meanwhile, has begun early preparations for what bankers are openly describing as a potential $1 trillion listing, a figure that would rank among the largest debuts in market history.

The financial backdrop explains the urgency. OpenAI has crossed $25 billion in annualized revenue, while Anthropic is approaching $19 billion on the same basis, with run-rate figures that some bankers expect to push higher by year-end. Those numbers turn both labs from speculative bets into businesses with the scale public markets demand. Huang's logic is simple: at this stage, the only way for Nvidia to add to its stakes is to buy shares on the open market like everyone else, which strips away the strategic premium that made the original investments attractive in the first place.

The setting matters. The Morgan Stanley TMT conference is where institutional investors take the temperature of the technology cycle, and Huang's remarks landed as a deliberate piece of guidance rather than an offhand aside. He paired the investment commentary with confirmation that Nvidia's upcoming Vera microprocessor will enter full production in the third quarter, with OpenAI, Anthropic, and SpaceX among its first major customers. The message was coordinated: the chips keep shipping, the commercial relationships hold, but the equity chapter with these two labs is effectively closed.

Nvidia has spent the past two years stitching itself into its largest customers through equity. It committed capital to OpenAI, Anthropic, and a roster of cloud and infrastructure players, building a web of relationships that tie chip demand to ownership stakes. Huang did not say Nvidia would sell any of those positions. He said it would stop adding to the OpenAI and Anthropic holdings specifically, because both are about to become public companies where the rules of engagement change entirely.

Why This Matters More Than People Think

Nvidia's investments were never only about returns. They functioned as demand insurance. By owning slices of the labs buying its most expensive chips, Nvidia aligned its balance sheet with the buyers driving its growth, and it gave those buyers a reason to keep ordering Blackwell and the coming Vera Rubin systems. Ending the private investment chapter does not unwind that alignment, but it does mark the moment when the relationship shifts from founder-and-backer to vendor-and-public-company, with quarterly scrutiny on both sides.

For OpenAI and Anthropic, an IPO is a forcing function. Public markets reward disclosure, margins, and predictable growth, not the open-ended capital incineration that has defined frontier model training. Both labs spend billions on compute, much of it flowing straight back to Nvidia. Once their income statements are public, every investor will be able to see exactly how much of their revenue converts to GPU purchase orders, which makes the Nvidia relationship far more visible and far more politically charged than it is today.

There is also a governance dimension that gets lost in the headline numbers. Nvidia sitting on the cap tables of its biggest buyers raised antitrust eyebrows in Washington and Brussels, where regulators have started asking whether a dominant chip supplier owning equity in its customers distorts a market it already controls. By declining to deepen those positions as the labs go public, Nvidia sidesteps a regulatory question that was only going to get louder, and it does so while keeping the commercial relationships that actually drive its revenue fully intact.

The bear case, however, is straightforward and worth stating plainly. Critics argue that Nvidia's circular financing, where it invests in customers who then spend that capital on Nvidia chips, has inflated demand signals that look stronger than underlying end-user adoption. If OpenAI and Anthropic enter public markets and their disclosures reveal thinner enterprise demand than the headline revenue suggests, the entire AI capital cycle could re-rate downward, and Nvidia would be exposed both as a supplier and as a shareholder. Huang stepping back from new private bets may read as confidence, but skeptics will read it as a chipmaker quietly reducing its concentration before the scrutiny of public markets arrives.

The Competitive Landscape

Nvidia is not the only strategic investor circling these labs. Microsoft remains OpenAI's largest backer and infrastructure partner, Amazon and Google have each poured billions into Anthropic alongside their cloud commitments, and SoftBank has emerged as a heavyweight financier of the broader OpenAI ecosystem. Each of these players faces the same closing window Huang described, which means the next year will likely bring a scramble to lock in final private positions before the IPO paperwork makes that impossible.

The competitive twist is that Nvidia's chips are no longer the only game in town. Google trains and serves models on its own TPUs, Amazon pushes Trainium and Inferentia, and Anthropic has reportedly explored running Claude inference on Microsoft's custom Maia silicon. As the labs diversify their compute, the strategic value of a Nvidia equity stake softens, because owning a piece of a customer matters less when that customer is actively building alternatives to your product. Huang's pullback may partly reflect that reality, not just IPO mechanics.

The pullback also reshapes how rivals position themselves. AMD has spent the past year courting the same labs with its MI-series accelerators and a more open software stack, pitching itself as the supplier that does not also want to own you. Custom silicon efforts at Amazon, Google, and now Microsoft give the labs leverage they did not have eighteen months ago. When the largest buyer of your product is also a shareholder, every negotiation carries a conflict of interest, and Huang reducing Nvidia's appetite for new stakes quietly removes a talking point that competitors had been using against it.

History offers a sharp parallel. In the late 1990s, networking and telecom suppliers financed the very carriers buying their equipment, a practice called vendor financing that propped up demand until the dot-com collapse exposed how much of it was hollow. Lucent and Nortel learned that lesson the hard way when their customer-financing books turned toxic. Nvidia's situation is healthier, because OpenAI and Anthropic generate real revenue that the carriers of 2000 never had, but the structural echo is unmistakable, and it is the reason every Nvidia disclosure about customer investments now draws extra attention.

Hidden Insight: The IPO Is the Real Audit

The most overlooked angle is that these IPOs will function as the first independent audit of the entire AI buildout. For three years, the scale of model training spend has been asserted rather than verified, communicated through funding announcements and capacity deals rather than audited financials. When OpenAI and Anthropic file full S-1 documents, the world will finally see gross margins, customer concentration, deferred revenue, and the precise share of spending that goes to compute. That transparency will either validate the trillion-dollar narrative or puncture it.

Consider what the audit actually exposes. Frontier labs book revenue from API usage, enterprise contracts, and consumer subscriptions, but their cost of goods sold is dominated by compute they rent or buy, and a large share of that flows to a single supplier. A public S-1 will force these companies to disclose supplier concentration, and analysts will immediately model how a change in Nvidia pricing or a shift to in-house chips would hit margins. That kind of sensitivity analysis has been impossible from the outside until now, and it will reframe how the market values both the labs and Nvidia itself.

Huang's decision to stop investing privately is a tell about timing, not doubt, but it also removes Nvidia's ability to shape these companies from the inside. As a private backer, a strategic investor gets board access, information rights, and a seat at the table during pivotal decisions. As a public-market holder, Nvidia becomes one shareholder among thousands, with the same delayed quarterly visibility as any index fund. Walking away from new private rounds means accepting a more arm's-length relationship with the two customers that matter most to its revenue.

There is a reflexive loop hiding here that few are pricing in. Nvidia's stock strength has been built partly on the perception of insatiable lab demand, and that demand has been partly funded by capital that Nvidia and its peers injected into the labs. If the IPOs replace strategic capital with public-market capital that is more sensitive to burn rates, the labs may face pressure to slow spending precisely when Nvidia needs them to accelerate. The very event that validates the labs financially could introduce the first real discipline on their compute budgets, and that discipline lands directly on Nvidia's order book.

There is a deeper signal in the sequencing. Anthropic filing before OpenAI suggests the smaller lab sees a strategic advantage in reaching public markets first, capturing investor capital and prestige before its larger rival. If Anthropic's debut goes well, it raises the bar and the valuation expectations for OpenAI's far larger listing. If it stumbles, it could chill sentiment for the whole category. The order of these IPOs may end up mattering as much as their size, and Nvidia, as a holder in both, is exposed to the outcome either way.

The uncomfortable truth this story challenges is the assumption that frontier AI is a private, founder-controlled domain insulated from the discipline of public markets. The moment these labs list, they inherit the obligations of any public company: forecast accuracy, margin defense, and answers when growth slows. The same capital intensity that let them outspend everyone in private becomes a liability when shareholders demand a path to profit. Nvidia closing its private chapter is the first visible acknowledgment that the rules are about to change for everyone in the AI economy.

What to Watch Next

In the next 30 days, watch for Anthropic's S-1 to move from confidential to public filing, which would expose its financials and the exact terms of Nvidia's $10 billion stake. Any disclosure about customer concentration or the share of revenue tied to a single cloud partner will move sentiment across the sector. Watch too for any signal from OpenAI on listing timing, because the gap between the two debuts will shape how investors price the category.

Over the next 90 days, track whether other strategic backers rush to finalize private positions before the IPO window closes. A flurry of late-stage rounds from Microsoft, Amazon, Google, or SoftBank would confirm Huang's read that the door is shutting. Watch also for Nvidia's own quarterly commentary on how it accounts for these stakes, since mark-to-market swings on a $40 billion combined position could begin to move its reported earnings in ways investors have not had to model before.

One underappreciated marker is lockup and insider-selling behavior after each debut. If Nvidia signals it intends to hold its stakes through the lockup and beyond, that is a vote of long-term confidence. If it trims at the first opportunity, the market will read that as the supplier hedging its own demand thesis. Either path will be visible in public filings within roughly six months of each listing, giving investors a rare, concrete read on whether the company that knows AI demand best is leaning in or quietly stepping back.

Over the next 180 days, the leading indicator to follow is the first full quarter of public disclosure from either lab. If gross margins and enterprise demand validate the spending, expect the AI capital cycle to extend and Nvidia's order book to stay full. If the numbers reveal heavier reliance on circular financing than bulls assume, the re-rating could be swift and broad. Either way, the era of taking AI economics on faith ends the day these companies ring the opening bell.

Nvidia stops investing in OpenAI and Anthropic not because it doubts them, but because the IPO is about to turn faith into a number everyone can check.

---