Economist 2026 article identifies AI compute as strategic resource requiring dedicated allocation markets

The article frames GPU cycles and power capacity as tradable commodities whose price signals currently fail to clear demand. It notes hyperscale providers allocate via internal queues and reserved instances rather than transparent exchanges. Data from 2025 shows average utilization rates on major clouds remaining below 40 percent for high-end accelerators while spot prices fluctuate 3x within single quarters.

NVIDIA's fiscal 2025 10-K reports data center revenue at $87.4 billion with 80 percent gross margins, confirming compute as the binding constraint. IEA 2025 electricity demand projections for data centers reach 8 percent of global supply by 2030, matching historical oil import dependence ratios for several OECD economies. These figures indicate the article understates physical power transmission bottlenecks that no financial market can bypass.

Existing capacity markets in ERCOT and PJM already trade megawatts with locational marginal pricing; extending similar mechanisms to co-located GPU fleets would require standardized performance benchmarks and verifiable uptime SLAs. Operational impact appears first in training workload scheduling, where multi-week queue times raise effective cost per FLOP above published hardware figures.

Next steps include pilots of compute futures contracts on existing energy exchanges, with settlement tied to measured throughput on standardized benchmarks rather than nominal FLOPS.