Compute demand is straining supply
AI customers are trying to lock up massive cloud capacity as demand outstrips supply, pushing vendor durability into procurement conversations. Reports show enterprises seeking large blocks of capacity from hyperscalers while analysts keep strong buy views on GPU vendors as hardware ramps drive throughput gains. (networkworld.com) (ibtimes.com.au) (developer.nvidia.com)
Enterprises are trying to lock up massive cloud GPU and custom-chip capacity as AI compute demand far outstrips supply. (networkworld.com) Amazon Chief Executive Andy Jassy told shareholders that two large customers asked to buy all of AWS’s 2026 Graviton instance capacity, and that AWS added 3.9 gigawatts of new power capacity in 2025. (networkworld.com) AWS has responded to tight supply with product and price moves: Elastic Compute Cloud capacity-block pricing rose roughly 15 percent and some P5 instance hourly rates jumped from about $34.61 to $39.80 in the US East (Ohio) listing. (networkworld.com) Wall Street remains bullish on GPU vendors: NVIDIA shares closed near $188.63 on April 10, 2026, with average 12‑month analyst targets around $264–$275 and many firms keeping "Buy" or "Strong Buy" ratings. (ibtimes.com.au) Software and model changes are also shifting throughput: NVIDIA says MiniMax M2.7 is a 230‑billion‑parameter mixture‑of‑experts model that activates about 10 billion parameters per token and supports a 200,000‑token context to cut inference cost. (developer.nvidia.com) Analysts warn this is not just a supply chain problem: Moor Insights’ Matt Kimball called two customers seeking full buyouts a sign of strategic dependency, and Info‑Tech’s Scott Bickley said “everything is sold out across the board.” (networkworld.com) Jassy wrote AWS “can’t agree to those kinds of requests,” even as the company plans to double power capacity by the end of 2027, leaving procurement teams and hyperscalers to jockey for guaranteed compute through 2026–2027. (networkworld.com)
