AI compute and power strain
Demand for advanced AI compute is outpacing the physical capacity of data centres, forcing rationing and price increases as buyers scramble for GPUs and electricity. Nvidia says it has more than $1 trillion in GPU orders through 2027 as hyperscalers keep buying, while reports say outages and energy limits are already constraining deployments and raising costs. (cnbc.com) (enterpriseai.economictimes.indiatimes.com)
Artificial intelligence companies are running into a simpler limit than software: not enough chips, data-center space, and electricity. (cnbc.com) Nvidia Chief Executive Jensen Huang said in March that the company sees more than $1 trillion in orders for its Blackwell and Vera Rubin systems through 2027. CNBC reported on April 14 that Nvidia shares had risen 18% over a 10-day streak, their longest run since 2023. (cnbc.com 1) (cnbc.com 2) Those systems are the specialized processors that train and run large artificial intelligence models, and buyers now need whole clusters of them plus networking gear, cooling equipment, and steady power. The Economic Times, citing The Wall Street Journal, reported on April 14 that companies are already rationing access and raising prices as outages and energy limits slow deployments. (sec.gov) (enterpriseai.economictimes.indiatimes.com) The bottleneck is no longer just chip fabrication. A usable artificial intelligence data center also needs land, transformers, transmission, backup power, liquid cooling, and permits, and each piece can delay the rest. (energy.gov) (enterpriseai.economictimes.indiatimes.com) The U.S. Department of Energy said in December 2024 that data-center load growth had tripled over the past decade and could double or triple again by 2028. Lawrence Berkeley National Laboratory estimated data centers used about 4.4% of U.S. electricity in 2023 and could reach 6.7% to 12% by 2028. (energy.gov) (eta-publications.lbl.gov) That helps explain why cloud groups and start-ups are scrambling for long-term supply instead of buying chips one quarter at a time. Nvidia told investors in its fiscal 2025 annual report that data-center demand was driven by accelerated computing and artificial intelligence, and that it began shipping production Blackwell systems in the fourth quarter of that fiscal year. (sec.gov) The rush has also shifted the economics of artificial intelligence services. When compute is scarce, providers can cap usage, delay customer rollouts, or charge more for premium access rather than promise unlimited capacity they cannot deliver. (enterpriseai.economictimes.indiatimes.com) Not everyone sees the strain lasting forever. Nvidia has argued that newer systems deliver more output per watt and per dollar, while utilities, developers, and governments are racing to add generation and build more data-center capacity. (cnbc.com) (energy.gov) For now, the artificial intelligence boom is being governed by physical infrastructure as much as by software code. The next fight is not only over who has the best model, but who can secure the chips and megawatts to run it. (cnbc.com) (enterpriseai.economictimes.indiatimes.com)
