The most striking figure in NVIDIA’s quarterly report isn’t a doubling of revenue—it’s $279B, the new supply-and-capacity commitments added in the quarter. Last quarter, it was $119B. In terms of structure, most of the amount goes to memory procurement. The gross margin guidance fell from 75% to 74%, and the CFO said plainly that memory prices are extremely volatile. This means the bottleneck for expanding AI compute has shifted up to storage cells—not CoWoS, not HBM. NVIDIA is locking in capacity for the next two years with its balance sheet, at the cost of near-term gross margin. AWS’s order for 2 million GPUs and the government data centers’ purchase of 100,000 GPUs are two sides of the same story: demand is real, but supply is exclusive. Beneficiaries: DRAM/NAND vendors get guaranteed orders, but price elasticity is capped. Losers: second-tier cloud providers and neocloud—after the top hyperscalers take their allocation, they’re left with less capacity and higher prices. Next, watch whether the Q3 gross margin can hold at 74%, and the spot price trend for memory. If memory prices pull back in Q4, the “bottleneck” narrative will soften; if they keep rising, $279B is NVIDIA’s full-scale lockout of competitors. Source: ca.investing.com (2026-08-26), ca.investing.com (2026-08-27).