AI Markets: Nvidia pulls back on Ohio, targets $1 trillion, SpaceX reveals its shareholders

Nvidia is shrinking its financial guarantee for OpenAI’s mega data center in Ohio, even as it doubles its chip sales target to more than $1 trillion through 2027. Meanwhile, quarterly filings reveal who quietly bet on SpaceX, and Nebius is pushing up prices on its AI data center contracts.

Nvidia cuts its guarantee for OpenAI’s giant Ohio data center in half

Nvidia is scaling back its financial commitment to the biggest data center project ever announced, after investors raised concerns about the risk it was taking on.

The key points:

  • Nvidia was set to guarantee roughly $250 billion of the debt and lease tied to the 10-gigawatt campus OpenAI is leasing from SB Energy, SoftBank’s energy subsidiary, on the site of a former uranium enrichment plant in Piketon, Ohio.
  • That amount is being cut to less than $120 billion, according to the Wall Street Journal, and would now cover only the first phase of construction.
  • A deal could be signed as early as this weekend; a separate package of up to $350 billion is still under discussion to directly finance OpenAI’s chip purchases.
  • Nvidia shares fell about 5% when the news broke.

Why Nvidia is easing off the gas a little

The Ohio project would have made Nvidia the financial backer of a customer that doesn’t yet have an “investment grade” credit rating: OpenAI isn’t profitable and can’t borrow cheaply on its own for a project costing more than $500 billion (chips included). So Nvidia stepped in as guarantor, an arrangement that several observers, including investor Michael Burry, describe as “circular financing”: the chip seller indirectly guarantees the purchase of its own chips.

According to the Wall Street Journal, it was Nvidia’s own investors who pushed for a smaller commitment. The new structure limits Nvidia’s exposure to the first tranche of the data center rather than the entire 10 GW campus, a project estimated at more than $500 billion in total. The deal itself isn’t in question: OpenAI still needs this data center to deliver on its compute promises, and Nvidia still has every reason to want the project built.

Caution that doesn’t stop big bets elsewhere

This one-off pullback doesn’t change Nvidia’s underlying strategy. In early August, the company also signed agreements with six financial giants (including Goldman Sachs, BlackRock and KKR) to raise more than $500 billion in third-party capital for AI compute infrastructure. The difference: that money comes from outside investors, not from Nvidia’s own balance sheet, which reduces its direct risk if a project defaults.

For a European investor, the episode illustrates a simple point: financing structures around AI data centers are watched very closely by the market, and a mere adjustment to a guarantee is enough to move the stock of a company worth more than $5 trillion. Nvidia (Nasdaq) is not eligible for the PEA; exposure goes through a standard brokerage account or a UCITS-format US equity/technology ETF.

Sources:

SpaceX on the market: quarterly filings reveal who bet big

The 13F filings submitted to the SEC for the second quarter show, for the first time, the size of several major institutional investors’ positions in SpaceX.

The key points:

  • Harvard Management Company held $2.2 billion in SpaceX shares as of June 30, its largest US equity holding, ahead of Amazon and Nvidia.
  • The University of California reported a position of about $1 billion; the University of North Carolina and Washington University in St. Louis also benefited from the rally.
  • AMD holds about 3.3 million SpaceX shares, worth more than $565 million, a bet described as financial rather than strategic.
  • SpaceX, which went public at $135 per share in June and was valued at about $1.8 trillion in mid-August, remains well below its June peak.

Why everyone is showing their cards at the same time

It’s no coincidence that several funds disclosed their positions on the same day: the SEC requires institutional investors managing more than $100 million to file a 13F form every quarter, within 45 days of quarter-end. Since SpaceX has only been listed since June, this June 30 filing is the very first to reveal which major US funds already held shares before the IPO. Harvard, with $4.3 billion in total US equities in this filing, had therefore concentrated more than half of its listed portfolio in this single bet.

The AMD paradox: a shareholder, but shut out of AI compute

AMD’s case is even more telling. Elon Musk confirmed in early August that SpaceX’s artificial intelligence would run exclusively on Nvidia chips, cutting AMD out of orders tied to “space compute.” Even so, AMD chose to keep, or even increase, its financial stake in SpaceX: the bet is on the growth of Starlink (more than 12 million subscribers, up from 2.3 million at the end of 2023), satellite manufacturing and the expansion of launches, not on chip orders. In other words, getting left out of a contract doesn’t stop you from wanting a piece of the jackpot.

For a European investor, these filings mainly confirm one thing: SpaceX (SPCX, Nasdaq) is not eligible for the PEA, and neither are the big US stocks that hold it. Indirect exposure goes through a standard brokerage account or a UCITS-format US equity ETF, where SpaceX’s weighting is generally marginal.

Sources:

Nvidia doubles down: more than $1 trillion in AI chip sales targeted through 2027

Same week, two opposite signals from Nvidia: caution on one specific project in Ohio, but a doubled sales target across its entire AI chip lineup.

The key points:

  • Jensen Huang is raising the cumulative sales target for Blackwell and Vera Rubin chips over 2025-2027 from $500 billion to more than $1 trillion.
  • Data center revenue for the first quarter of fiscal 2027 (ended late January) jumped 92% year over year to $75 billion.
  • Nvidia attributes the doubling to “demand visibility,” driven in particular by the ramp-up of inference (using already-trained models, as opposed to training them).
  • Nvidia’s market cap exceeds $5 trillion.

Where the doubling comes from

Jensen Huang’s previous target, set in April, called for $500 billion in cumulative sales of Blackwell and Vera Rubin chips between 2025 and 2027. Four months later, he’s doubling it: more than $1 trillion over the same period, a figure that covers only the chips themselves, not complete systems or related services — so the real total could be even higher, according to Nvidia. The reasoning builds on the trajectory already seen: data center revenue is up 92% year over year, and orders from major customers (OpenAI, Microsoft, Meta, Anthropic) remain strong despite the debate over a possible bubble.

Execution risk, and the link to the AI bubble debate

A target like this carries obvious execution risk: Nvidia has to manufacture and deliver chips at unprecedented scale, manage an increasingly complex supply chain spanning several product generations, and keep a big enough technological lead that customers don’t turn to alternatives (AMD, hyperscalers’ in-house chips, Chinese competitors). It’s precisely because demand looks insatiable that Nvidia is simultaneously piling up financing deals for its customers — including the scaled-back Ohio guarantee mentioned above. Doubling the sales target while reducing direct financial exposure on a specific project isn’t contradictory: it’s a way to keep driving demand without carrying all the credit risk that comes with it alone.

For a European investor, Nvidia (Nasdaq) is not eligible for the PEA; access goes through a standard brokerage account or a UCITS-format US equity/technology ETF. This raised target is a statement from the company itself, not a guaranteed independent forecast.

Sources:

Nebius pushes up prices for its AI data centers: how far does the $100 billion math go?

AI compute provider Nebius is raising its contracted power capacity target for the third time this year, and negotiating its new leases at significantly higher prices.

The key points:

  • Nebius is raising its contracted power target from 4 to 5 gigawatts by the end of 2026, after already lifting it from 3 to 4 GW in May.
  • Its most recent contracts are going for $20 million to $25 million per megawatt for medium-term leases, and up to $40 million to $50 million for short-term ones.
  • Nebius has signed about $46 billion in cumulative five-year contracts with Meta and Microsoft.
  • Second-quarter 2026 revenue: $582 million, up 454% year over year; the stock rose 8.9% after the report.

The mechanics of the price increase

By simply multiplying the targeted 5 gigawatts by the $20 million per megawatt at the low end of the range, the trade press arrives at a potential of more than $100 billion in revenue. It’s important to be clear about what that figure represents: it’s not an official target published by Nebius, but an arithmetic extrapolation built from two real company data points (the targeted capacity and the per-megawatt price of the latest contracts). Nebius itself is more cautious in its official numerical guidance, targeting 2026 annual revenue of $3 billion to $3.4 billion and an annualized run rate of $7 billion to $9 billion by year-end.

What this price increase really tells us

What can be verified, however, is that Nebius is managing to lease its capacity at higher prices than a few months ago: contracts at $20 million to $25 million per megawatt are above previous levels, and the company says it could already sell all of its 2027 capacity on those terms if it wanted to, but prefers to keep some of it for spot customers willing to pay more. It’s a sign of a persistent scarcity of available AI compute capacity, more than confirmation of $100 billion in revenue.

For a European investor, Nebius Group (Nasdaq: NBIS) is a company incorporated under Dutch law but listed only in New York: it is not eligible for the PEA, and access goes through a standard brokerage account or a UCITS-format US equity/technology ETF.

Sources:

Edition prepared by our news-monitoring agent (data and sources as of Monday, August 17, 2026), reviewed in line with our editorial policy. General information, not investment advice: investing carries a risk of capital loss.