After a year of prodding staff to maximise AI use, a slew of global corporations are abruptly reversing course. Uber exhausted its 2026 AI budget by April, imposed monthly limits on coding tools and acknowledged it had become “harder to justify” spending on AI tokens. Walmart capped internal token consumption, Amazon warned engineers to stop using AI “just for the sake of using AI”, and Meta, after recording a steep jump in costs, is readying restrictions. The trigger is not technological failure but a shift in how AI companies charge. OpenAI and Anthropic have moved from flat subscriptions to token-based billing, a model that makes every prompt, automated workflow and autonomous agent task a direct—and growing—expense.

That billing change has exposed how the leap from simple chatbots to autonomous AI agents multiplies compute demands. CEO Sam Altman acknowledged that cost had become a “huge issue” for customers this year, while Cisco’s product chief noted that a single agent can consume as much infrastructure as dozens of humans working non-stop. The appetite for silicon is now rippling well beyond data centres. Apple CEO Tim Cook warned that memory and storage chip prices are being driven to unprecedented levels by AI demand, making iPhone price hikes “unavoidable”. Dell, Ford and a coalition of retailers and medical suppliers have sounded similar alarms, pointing to a squeeze on DRAM and NAND flash that threatens to feed into consumer inflation.

Viewed from Washington, the chip crunch complicates a geopolitical tangle. Microsoft has quietly turned China into its fastest-growing AI revenue market, channelling OpenAI’s GPT models through Azure to ByteDance, Ant Group and Tencent, even as OpenAI itself refuses to sell there and US lawmakers warn of intellectual property theft. Meanwhile, the rally in AI-related stocks faces a near-term stress test: Micron Technology reports earnings as a bellwether for chip demand, with investors watching for any softening that could challenge elevated valuations. Away from markets, the Reuters Institute’s 2026 Digital News Report finds that 10% of news consumers now access information via AI chatbots, with the share rising to 16% among the under-35s, accelerating a shift away from publisher websites and traditional broadcasters.

The rapid appropriation of AI is prompting more than a financial audit. Researchers are documenting a pattern of “cognitive offloading”, where users delegate everyday decisions—from what to eat to navigating relationships—to chatbots, risking a weakening of critical-thinking skills over time. Separately, studies in the journal Digital Journalism indicate that news audiences trust AI-generated content only when human oversight is clearly signalled, particularly where nuance is required. As businesses recalibrate their AI spend, they are shifting to layered strategies: reserving costly frontier models for complex tasks and using cheaper, smaller models for routine work, a practice that executives say can cut bills by up to 90%. The next milestone for the broader AI trade will be the Federal Reserve’s preferred inflation gauge and a final first-quarter GDP reading due next week, figures that will help determine whether the macro environment can support current levels of tech investment.