Nvidia AI equity bets top $40 billion as circular-deal scrutiny grows

Nvidia’s stakes in artificial-intelligence customers crossed $40 billion this week, sharpening a long-running argument about whether the chipmaker is bankrolling a strategic moat or a closed loop of self-dealing as OpenAI moves toward an initial public offering.

The running tally combines $30 billion committed to OpenAI, up to $3.2 billion to glass and fibre supplier Corning, and $2.1 billion in cloud operator IREN. Earlier positions in xAI, CoreWeave and Anthropic sit alongside. Critics call the pattern an “AI money-go-round” because the proceeds typically come back to Nvidia in the form of chip orders. At a Morgan Stanley conference on 4 March, chief executive Jensen Huang told investors the OpenAI cheque “might be the last” before the start-up’s IPO. He also flagged that a separate $10 billion stake in Anthropic would not be topped up.

Add the funding-round contributions and the figure climbs higher. The OpenAI commitment was part of a $110 billion round that drew a further $50 billion from Amazon and $30 billion from SoftBank. As part of that pact, Nvidia secured 3 gigawatts of dedicated inference capacity and 2 gigawatts of training capacity on its forthcoming Vera Rubin systems, locking in multi-year off-take from the most prominent buyer of advanced GPUs.

Corning’s stake, set at up to $3.2 billion, is tied to fibre-optic and packaging materials used in Nvidia’s networking switches. The supplier reported $4.35 billion in first-quarter sales and $0.70 in earnings per share. IREN, building data-centre capacity in Texas and British Columbia, posted a third-quarter net loss of $247.8 million on AI Cloud services revenue of $33.6 million. Both deals show the breadth of Nvidia’s footprint, which now reaches suppliers, model labs, neoclouds and merchant cloud operators.

What the bears see

The structural objection is straightforward. Nvidia takes equity in a customer. The customer uses some of that money to buy Nvidia chips. Nvidia recognises revenue. Sceptics say the loop inflates demand metrics that justify a market value of roughly $5.2 trillion and a price-earnings ratio of 43.92, on the trailing year’s $120.07 billion of net income.

Mizuho Securities chip analyst Jordan Klein put it bluntly in a quarterly note circulated to clients. The so-called neocloud investments “feel more questionable to me and likely investors,” Klein wrote, adding that “it smells like you are pre-funding the purchase of your own GPUs and products.” Bloomberg has since mapped the web of cross-investments that runs between Nvidia, Microsoft, Oracle, AMD and the largest AI labs. Money, chips and cloud credits keep cycling through the same handful of names.

Sarah Friar, OpenAI’s chief financial officer, has sharpened the unease. Friar warned investors this quarter that without faster revenue growth the start-up could struggle to honour its computing contracts. The remarks landed alongside Wall Street Journal reports that OpenAI has been missing internal growth targets. Even diluted across a $110 billion syndicate, Nvidia’s exposure to a single counter-party of that scale is now a sustainability question for the wider chip cycle. Comparisons to the late-1990s telecoms-and-equipment boom, when vendor financing inflated demand for switching gear, are turning up in fund letters.

What the bulls see

The counter-argument is that Nvidia is buying optionality, not flattering its books. Wedbush Securities analyst Matthew Bryson conceded the “circular investment” critique in a note this month but argued that “successful execution could create a powerful competitive edge” by pinning down chip uptake at customers that might otherwise default to AMD or to in-house silicon. Bryson holds an outperform rating with a $300 price target. That implies an 11 per cent uplift from current levels.

Inside Nvidia, executives frame the equity programme as connective tissue. The stakes are meant to lock in long-term off-take agreements and align customers on architecture choices before rivals close the gap. Forthcoming software stacks tied to Vera Rubin and to the Spectrum-X networking platform are central to that thesis. The same logic underpins Nvidia’s appearances on the supply side, including the Intel manufacturing partnership reportedly under discussion with Apple and the broader memory-supplier rally that has lifted Micron 38 per cent in a week.

The OpenAI walkback

Huang’s Morgan Stanley remarks doubled as a pivot. The September 2025 Nvidia-OpenAI infrastructure announcement had carried a $100 billion headline. By March, Huang said that figure was “probably not in the cards,” partly because OpenAI’s IPO timetable will limit further private-market investment by chipmakers. The $30 billion stake stands, but the distance between the original ambition and the executed cheque has become a touchstone for sceptics who saw the larger number as evidence of strategic over-reach.

Concerns about the durability of these arrangements resurfaced this week as Broadcom shares fell 4 per cent on reports that OpenAI’s custom-silicon financing had stalled at $18 billion. The episode underlined how dependent the AI capital stack remains on a small group of underwriters, and how quickly sentiment can turn when one funding leg wobbles.

AMD pulls ahead

The market has priced circularity risk into Nvidia’s relative performance. The chipmaker is up about 11 per cent year-to-date. AMD has rallied roughly 97 per cent on the back of hyperscaler design wins. AMD’s first-quarter revenue rose 38 per cent to $10.25 billion, with data-centre sales up 57 per cent to $5.78 billion and free cash flow climbing 253 per cent to $2.566 billion. Meta has committed to a 6-gigawatt deployment of AMD’s Instinct accelerators, and OpenAI signed for 6 gigawatts of the MI450, a procurement scale that until last year had been the preserve of Nvidia’s Hopper and Blackwell lines.

The sector trade has broadened beyond the duopoly leader. Qualcomm gained 8 per cent on its earnings beat and a fresh OpenAI partnership for inference silicon. The S&P 500 and Nasdaq closed at records, with AI optimism doing most of the heavy lifting. Nvidia’s 85 per cent share of the AI accelerator market is intact. Its premium over peers, however, has compressed as buyers signal a willingness to dual-source.

What investors will watch

The most immediate test is OpenAI’s IPO filing. Bankers have indicated the prospectus could land within twelve months. Once it does, Nvidia’s $30 billion stake gets a public mark and OpenAI must disclose whether its revenue is ahead of or behind the curve set by its computing commitments. Friar’s contracts warning is the elephant in that room.

Nvidia’s August earnings call is the next checkpoint. Management has so far avoided breaking out how much of customer-side demand is downstream of its own equity contributions. After Klein’s note made the rounds, analysts are likely to push harder for that disclosure.

There is also a regulatory overhang. The White House has begun scoping a review of frontier-AI models that could limit the volumes AI majors are committing to spend, and dictate which jurisdictions host the resulting capacity. A tightening of export rules on advanced GPUs would interact directly with Nvidia’s offtake assumptions, and with the value of the equity book it is now sitting on.

For now the trade is binary. Bryson sees the equity programme as Nvidia’s most durable advantage. Klein sees it as a flag. Whether the $40 billion turns into third-party chip orders, or stays inside a loop investors eventually discount, decides which of them was right.