AI Firms Ration Access as Global Compute Shortage Deepens

The All-You-Can-Eat AI Era Is Over

The artificial intelligence industry is facing its most severe infrastructure crisis in years, as explosive demand for computing power collides with a cascade of supply chain disruptions. Major AI providers including OpenAI, Anthropic, and GitHub are rationing access to their models, overhauling pricing structures, and making painful decisions about which products survive.

Demand Outpaces Supply

OpenAI's API token usage surged from six billion per minute in October to fifteen billion per minute by the end of March, driven largely by autonomous coding agents and enterprise tools. The company is shutting down its Sora video generation app on April twenty six to redirect computing resources toward higher-priority products. OpenAI's CFO Sarah Friar has described spending much of her time hunting for near-term compute capacity.

Anthropic's Claude API logged just ninety eight point nine five percent uptime over ninety days, well below the ninety nine point nine nine percent industry standard. The company has tightened session limits, introduced off-peak usage incentives, and blocked flat-rate subscriptions from powering third-party agent tools.

Supply Chain Under Siege

The crisis runs deeper than simple demand growth. Spot DRAM prices have surged nearly seven hundred percent in some cases as manufacturers divert production to AI infrastructure. The disruption of Qatar's helium exports, triggered by military strikes on the Ras Laffan Industrial City in February, removed roughly thirty percent of global semiconductor-grade helium supply overnight. Helium is essential for cooling the extreme ultraviolet lithography machines used to manufacture advanced chips.

Spot GPU rental prices for NVIDIA's Blackwell series have climbed forty eight percent in two months, reaching four dollars and eight cents per hour. Lead times for data centre chips now stretch thirty six to fifty two weeks.

Alibaba Bets Big While Others Cut Back

While Western labs tighten access, Alibaba is moving in the opposite direction, committing four hundred and thirty two million dollars to subsidise its Qwen chatbot and pricing API access at roughly one eighteenth of comparable Western offerings. The strategy aims to funnel users into Alibaba's broader commerce and cloud ecosystem, though it has come at a steep cost, with operating income falling seventy four percent in the most recent quarter.

OpenAI projects fourteen billion dollars in losses for twenty twenty six, with no break-even expected until after twenty thirty.