
Nvidia Hoards $530 Billion in Guarantees to Lock Up AI Supply
Nvidia's off-balance-sheet guarantees hit $530 billion, up from $184 billion, per SemiAnalysis; internal records show Huang and Kress questioned the strategy.
- By
- Sophie Lindqvist
- Filed
- Channel
- AI & Compute
- Read
- 3 min read
Nvidia held $530 billion in gross off-balance-sheet financial guarantees at the end of its fiscal second quarter, up from $184 billion three months earlier, according to SemiAnalysis. The near-quarter-trillion-dollar increase came mostly from memory procurement. The figure is confirmed in a filing with the U.S. Securities and Exchange Commission. That is the hard center of this story.
The breakdown matters. The SEC filing, as parsed by SemiAnalysis, shows:
- $366 billion in future commitments, of which $279 billion covers supply and capacity, mainly memory and manufacturing;
- $56 billion in additional commitments, including $36 billion in AI cloud take-or-pay agreements and $20 billion in data-center leases expected to be reassigned;
- $108.5 billion in maximum gross exposure under guarantees, of which $105 billion is a lease and credit-support guarantee tied to SB Energy's PORTS Technology Campus in Ohio for an OpenAI affiliate, plus $3.5 billion for other AI-cloud partner land, power and shell guarantees.
Nvidia also carries $99 billion in equity investments, per SemiAnalysis.
Why does this size matter?
Nvidia's dominance rests on more than chip performance and its software stack. The company has locked up a substantial share of global capacity for advanced logic at TSMC and high-bandwidth memory elsewhere. That grip limits how fast rivals can ramp their own accelerators. Jensen Huang negotiated the supplier agreements personally with major contractors in Taiwan and South Korea, before his departure.
The scale also buys price. SemiAnalysis estimates SK hynix could charge Nvidia between $3.00 and $3.30 per gigabit of HBM4 in 2027. Other first-tier customers — AMD, Amazon, Broadcom and Google — would pay between $3.70 and $4.10. The analysts describe an informal arrangement: Nvidia gives SK Group preferential access to its AI GPUs in return.
These are estimates, not contract terms. The pricing gap rests on one analyst house's model. Nvidia and SK hynix have not confirmed it.
Did Nvidia's own leadership doubt the strategy?
Yes, and this is the part the $530 billion headline does not say. At an August meeting with senior executives, Huang raised concerns about the volumes of wafers and memory Nvidia had committed to buy, according to The Wall Street Journal. Cloud providers were seeking to cut their reliance on Nvidia hardware in favor of alternative accelerator platforms.
Then-CFO Colette Kress questioned the criteria for selecting recipients of financial support and the potential returns. She reportedly cut Nvidia's initial financial commitment to OpenAI's Ohio data-center project in half before the two sides signed an agreement on August 17. That facility is expected to house roughly $300 billion worth of Nvidia processors.
Huang went further. He reportedly described Kress, now interim CEO, as a mere "reporting CFO" rather than an executive capable of strategic decisions. He also questioned the financing strategy's merits in the event of his own departure. A founder doubting the machine he built is a signal worth more than any roadmap slide.
What did the guarantees buy?
Demand, among other things. Nvidia extended financial guarantees to emerging cloud providers and data-center developers who otherwise could not afford its hardware at scale. The guarantees create customers for the very capacity Nvidia has reserved. They also create exposure if those customers fail.
What changes after Huang?
Possibly the supplier relationships. SemiAnalysis argues that without Huang's personal ties to semiconductor executives, Nvidia may struggle to keep preferential treatment. The analysts project that customers other than Nvidia could account for over 70% of TSMC production capacity in 2028 — a forecast, not a confirmed allocation.
That shift cuts two ways. It could let competitors ship far more AI processors. It could also make Taiwanese and South Korean suppliers more hesitant to invest in added capacity without the assurances Huang once provided personally, leaving the industry's next expansion underwritten by nobody in particular.
Source: Tom's Hardware
More from Sophie Lindqvist
Show full bio
News editor covering money, funding and market shifts around Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
1,055 articles
Related articles
nvidia-carries-164-5-billion-in-commitments-into-ceo-vacuum
Nvidia Carries $164.5 Billion in Commitments Into CEO Vacuum
nvidia-s-500-billion-ai-financing-plan-meets-wall-street-skepticism-4222ffca
Nvidia's $500 Billion AI Financing Plan Meets Wall Street Skepticism
nvidia-s-125-billion-guarantee-shifts-ai-debt-risk-to-insurers-3e980ffc
Nvidia's $125 Billion Guarantee Shifts AI Debt Risk to Insurers
morgan-stanley-restores-nvidia-as-top-semiconductor-pick-300-target-932a6602
Morgan Stanley Restores Nvidia as Top Semiconductor Pick, $300 Target



