AI Demand Is Overstated — Only Anthropic Is Being Realistic

TL;DR

AI token demand is artificially inflated by gaming, budget overruns, and unsustainable flat-rate plans, and Anthropic is the only major lab adjusting for it.

Key Points

• 1.AI token consumption is exploding beyond any budget. Uber's CTO revealed AI coding tools burned through the company's full-year AI budget by April; Goldman Sachs says inference costs are overrunning initial budgets by orders of magnitude and may rival engineering headcount.
• 2.Tokenmaxxing is corrupting demand signals. Meta and Shopify rank employees by AI token usage, not productivity output — Jensen Huang even said he'd be 'alarmed' if a $500K engineer didn't consume $250K in tokens — causing people to game the metric by burning spend with no real output.
• 3.Flat-rate plans proved financially ruinous for AI labs. A single $200 Anthropic Max plan was estimated to consume $2,000–$5,000 in compute; Anthropic killed unlimited subscriptions, cut off tools like OpenClaw, and moved enterprise customers to per-token billing to price for verifiable demand.
• 4.Anthropic's 'cone of uncertainty' framing exposes systemic overinvestment. Data centers take 1–2 years to build and companies are making billion-dollar bets on unverified demand; Anthropic's per-token model gives it clean, real revenue data ahead of its expected IPO, unlike rivals still guessing.