$530B in guarantees: Nvidia funded its own demand

Nvidia's total commitments and guarantees reached about $530 billion last quarter — up from $184 billion the quarter before, a near-tripling in three months. According to a Wall Street Journal investigation published Thursday, much of that money exists to do one thing: pay Nvidia's own customers to keep buying Nvidia's chips.

What the Journal found

The report, based on documents and internal accounts, describes a meeting in August 2026 between former CEO Jensen Huang and longtime CFO Colette Kress, now interim CEO after Huang's resignation this week. Kress wanted to rein in the circular-financing commitments — Nvidia funding AI startups and smaller cloud providers so they could afford to buy its GPUs. Huang refused.

The scale of what was on the table: early discussions floated underwriting roughly 8 gigawatts of OpenAI computing capacity, valued at up to $250 billion. Kress got the signed guarantee cap cut to about 4.25 gigawatts, or up to $125 billion. The Journal reports Huang privately described the problem as an "avalanche" of prepaid silicon and memory Nvidia had committed to buy from TSMC — and that the largest cloud companies were quietly moving away from Nvidia toward their own custom chips. Funding the buyers was the alternative to letting the surplus hit the open market and push prices down.

The balance-sheet symptoms line up with that story. Nvidia's receivables had ballooned to $63 billion, with average payment terms stretching from 45 to 60 days — customers taking longer to pay for chips they may not need. Industry analyst Ed Zitron estimates at least $200 billion worth of Nvidia GPUs are sitting idle in warehouses, bought by cloud giants that lack the grid capacity or the end-user demand to switch them on. Nvidia has since contacted TSMC to wind down upcoming chip orders, according to the report.

The market has already voted once: Nvidia shares fell 13% in the week the report emerged, its worst week since April 2025, wiping about $573 billion from the company's market value. Nvidia reports third-quarter results on Tuesday, November 17 — its first earnings since Huang's departure, with Kress presenting as interim chief executive.

The arithmetic that makes the "shortage" story collapse

For three years, the defining fact of the AI boom was that you could not get Nvidia GPUs. The waitlists, the allocations, the premiums — the whole industry organized itself around scarcity. The Journal's numbers say the scarcity was, at least in part, manufactured on the demand side:

• $530 billion in commitments and guarantees, per SemiAnalysis's reading of Nvidia's latest quarterly filing — up from $184 billion a quarter earlier.
• $279 billion of that is supply commitments for memory and chip wafers, up from $119 billion, per coverage of the same filing.
• $36 billion is locked in deals to buy back unused cloud capacity from the "neoclouds" — the small providers Nvidia financed, per the same coverage.
• $63 billion in receivables, with payment terms stretched to 60 days.

Nvidia has long rejected the circular-financing label, and Huang previously dismissed the idea. But the structure is hard to describe any other way: Nvidia's money went out as venture funding and guarantees, came back as GPU orders, and got booked as revenue. The company's own former CFO tried to cut the commitments in half. The question the report raises is not whether the revenue was collected — it was — but whether the demand behind it was real.

Why this matters beyond Nvidia's stock

This is the third leg of a story I have been tracking all week. On Wednesday, SoftBank wrote down $3.7 billion on turbine deposits for AI data centers that may never get built. On Thursday, the Blackwell fingerprint report showed export-banned GPUs in China's open-weight models. Now the demand side of the same boom turns out to be partly self-funded. Three different balance sheets, one pattern: the AI buildout's numbers were bigger than its end demand, and the gap was papered over with financing.

The honest test is November 17. If Kress uses her first earnings call to unwind the guarantees, the stock takes the hit now and the business gets smaller but real. If she defends them, the $530 billion keeps growing and the eventual correction gets bigger. Either way, the "shortage" is over as a story — not because supply caught up, but because the buyers were Nvidia all along.

Sources: The Wall Street Journal investigation as reported by Wccftech ("Instead Of Lowering GPU Prices, Former NVIDIA CEO Jensen Huang Reportedly Financed AI Startups To Absorb Excess Silicon," November 13, 2026) and PakTechStan ("Huang Financed AI Startups to Buy Nvidia's Own Chips as a TSMC Glut Looms, WSJ Reports," November 14, 2026); SemiAnalysis figures via the same coverage; Ed Zitron idle-GPU estimate via Wccftech. Nvidia has not confirmed the Journal's account, and no formal investigation has been publicly confirmed. Figures are as reported; the WSJ original is paywalled.