The "Usage Only Goes Up" Growth Story Just Hit the Finance Department's Budget — From today (Jul 6) Tesla meters employee AI spend: $200/week per person, manager sign-off to exceed. The trigger: engineers burning "thousands of dollars of tokens each week"; the company spent six months consolidating scattered usage into central procurement, then slapped on quotas. It's not alone — Uber blew through its entire 2026 AI budget by April and switched to $1,500/month per head; Meta, Amazon and Walmart are capping or steering staff to cheaper tiers. Sources: The Information / Electrek / IBTimes. Judgment: (1) My March line "models go from moat to utility" described the supply side commoditizing; these caps are the second half — the demand side is now managed like a utility too. When electricity became a utility, factories' first move wasn't celebrating cheap power, it was sub-metering each floor and setting budgets. Tesla just installed that sub-meter on tokens — enterprises are seeing agentic AI's full, undiscounted cost for the first time. (2) Don't call it bearish. Burning enough to need a cap, engineers preferring to file approvals rather than stop — that's the hardest penetration evidence there is. It's the flip side of MIT's "95% of pilots have zero impact" (ENTRY 93): the small slice that IS in real use runs hot enough to lose control. Value didn't fail to land; it landed somewhere expensive enough that the CFO has to step in. (3) It also caps a myth — "usage-based billing, only goes up, no ceiling." Quotas say: there's a ceiling, and it's the customer's budget discipline. Buyers now manage AI spend like a cloud bill; CFO and procurement enter. Primary-market read: discount the "consumption-driven" slice of AI companies' ARR — its ceiling is no longer capability but the finance department's patience. The moat migrates from "who's smarter" to "who gets more out of each token" — token efficiency is the new gross margin. GPT-5.6 Terra at half price, Sonnet 5 at $2/M — the price war is fighting over these newly-metered, unit-price-watching buyers. (4) Two threads: the cloud-bill explosion birthed FinOps (CloudHealth, Datadog); the token-bill explosion is birthing "AI FinOps / usage observability / cost attribution" — Tesla's DIY token-ranking dashboard is the gen-1 product, and independent companies will take that demand. And the ugly tell: Grok's beta is exempted from the $200 cap — even though engineers prefer Claude on merit, budget gets weaponized as a channel to funnel staff toward Musk's own model. Which model is "default" in the enterprise will next be decided by cost policy, not capability (after ENTRY 90 "licensed release," ENTRY 85 "model-swappable becomes table stakes" — now "cost-swappable" becomes table stakes). For founders: stop pitching "unlimited usage"; show how much business each token buys. Whoever makes the token bill controllable, attributable and optimizable owns the surest "meter" business in this utility era.