Supply commitments jump $250 billion in a quarter
According to an analysis by SemiAnalysis, Nvidia is reinforcing its dominance in the AI accelerator market not only through its chips but also by pre-booking supply chain capacity. The company's supply and capacity commitments rose by roughly $250 billion in a single quarter, pushing total off-balance-sheet guarantees to $530 billion. Most of the increase comes from memory procurement.
Cheaper HBM4 than rivals
The analysis estimates that SK hynix could charge Nvidia between $3.00 and $3.30 per gigabit of HBM4 memory in 2027, while other first-tier customers such as AMD, Amazon, Broadcom and Google would pay between $3.70 and $4.10. Nvidia is said to give SK Group preferential access to its AI GPUs as part of an informal arrangement.
Breakdown of the guarantees
Based on an SEC filing:
- $366 billion in future commitments ($279 billion of which is supply and capacity, mainly memory and manufacturing)
- $56 billion in additional commitments ($36 billion in AI cloud take-or-pay arrangements, $20 billion in data-center leases)
- $108.5 billion in maximum gross exposure ($105 billion lease/credit support for an OpenAI affiliate at SB Energy's PORTS campus in Ohio)
Internal doubts
According to The Wall Street Journal, then-CEO Jensen Huang raised concerns at an August meeting with senior executives about the massive volumes of wafers and memory Nvidia had committed to buy. Then-CFO Colette Kress also questioned the criteria for selecting financial support recipients and the potential returns, and reportedly halved Nvidia's initial commitment to OpenAI's Ohio data center project before the agreement was signed.



