Nvidia's Pause Button: Inside the Chip Giant's Retreat From Banker to Builder

Nvidia halts its AI financing initiative, citing control and antitrust worries, despite record $96.2B quarterly revenue and strong chip demand.

The most telling detail in Nvidia's latest earnings story isn't the $96.2 billion in quarterly revenue — it's what the company chose not to do. Reports emerged Friday that Nvidia has shelved its "AI Compute Partnership" financing initiative, a move driven by internal concerns about losing control of its own business and the prospect of antitrust scrutiny. The company that just posted a record quarter is, by its own hand, applying the brakes.

Shares fell 4.1 percent on the day to €187.78, a decline rooted not in disappointing fundamentals but in the optics of a chipmaker stepping back from a role it had increasingly come to play: banker to the AI infrastructure boom. The stock still sits roughly 7 percent above its level from 30 days ago, though it remains 7.3 percent below the 52-week high of €202.50 reached in May.

The Circular Economy of AI Capital

The financing pause marks a notable shift for a company that has blurred the lines between semiconductor supplier and capital provider. Nvidia has poured roughly $50 billion into AI labs that subsequently buy its own chips, and CFO Colette Kress has indicated that about 25 percent of the business expected in 2028 could come from such financed customers. The company has been building funding platforms exceeding $500 billion alongside partners including Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, and has committed up to $105 billion in leasing guarantees for OpenAI's Ohio campus, alongside a $1.5 billion investment in SB Energy.

Critics have called this circular financing — Nvidia effectively underwriting the very demand for its own products. Supporters counter that without these capital bridges, many AI clouds simply couldn't place the orders they need for long-term capacity. Both arguments carry weight, but the company's decision to pause its compute partnership initiative suggests the internal calculus has shifted. Financing an entire supply chain from chip to cloud capacity eventually makes you a bank — and banks attract regulators. Nvidia appears to have concluded it can maintain influence without absorbing that risk.

A Customer Base in Transition

The composition of Nvidia's customer base is changing faster than its stock price. While established hyperscalers grew their revenue contribution by 102 percent to $48.7 billion, a newer class of buyers is expanding even more rapidly. AI-native clouds, industrial firms, sovereign state projects and specialized providers like CoreWeave and Nebius collectively contributed $40.3 billion, up 138 percent year over year.

This diversification eases the concentration risk of relying on a handful of tech giants for future growth. Amazon Web Services has ordered an additional two million GPUs across the Blackwell Ultra and Rubin architectures for delivery in 2027 and 2028, building on an earlier commitment of more than one million units.

The Margin Squeeze Arrives

Growth at this scale carries a literal price. Kress has guided gross margin down to 71 to 72 percent by the fourth quarter of fiscal 2027, a steady decline from the 75.0 percent posted in the most recent quarter and the roughly 74.0 percent expected for the current period.

Memory costs are the primary culprit. TrendForce projects server DRAM prices will climb 13 to 18 percent in the third quarter of 2026, and Nvidia has responded by raising prices on its Vera Rubin and Grace Blackwell AI servers by 15 to 17 percent for deliveries starting in 2027. Memory now accounts for roughly a quarter of a rack's cost, while the GPU share has fallen from over 80 percent to about 50 percent. New memory fabrication capacity isn't expected online until 2029 or 2030, and TSMC, Nvidia's key manufacturing partner, plans price increases of 10 to 15 percent across all process nodes.

The company is meanwhile pushing forward on the product front. The Vera Rubin platform has entered full production, complemented by Vera, its first processor designed specifically for AI agents. Samsung has confirmed an 80 percent "golden yield" on HBM4, the next memory generation critical to Rubin production. Reports also surfaced of a potential $12.9 billion acquisition of Hugging Face, which would secure Nvidia a distribution channel for open-source models — though neither company has confirmed the transaction.

Confidence in the Numbers

The market's faith in the underlying business remains substantial. The board has authorized an additional $99.0 billion for share buybacks, following $26.0 billion returned to shareholders in the second quarter. Cathie Wood's ARK funds took advantage of the post-earnings dip to accumulate roughly $53 million in additional shares. The record date for the next quarterly dividend of $0.25 per share arrives September 10, 2026, with payment on October 1.

Analyst sentiment skews bullish despite the recent wobble. Truist raised its price target to $346, JPMorgan to $320, and Bernstein to $400, against a consensus of just above $320. Insider selling of more than $664 million this year — much of it from a single director — hasn't dented the overall outlook.

The stock's 13 percent gain over the past month and its position comfortably above the 200-day moving average suggest investors are digesting the tension between record growth and the quiet signals that even the industry's dominant chipmaker is starting to define its limits. The mature version of growth, it seems, means not financing everything just because you can.