Sunday Jun 28
STRATEGYOther

Firms Crack Down On AI Spending

26JUN
BUDGET CUTBLANK CHECKCAPPED

The blank check for AI is gone. Companies that once told staff to use as much AI as possible are now capping it. Uber burned a year's budget in four months.

The shift has a name: tokenmaxxing, where teams maxed AI use to look productive. The bills came in. The returns did not.

Lindy moved all its traffic off Claude to cheaper DeepSeek. A D.A. Davidson analyst warned big customers may cap runaway token spend.

This bites right as OpenAI and Anthropic prep IPOs. Growth built on unlimited spending now has to show real returns.

full brief & sources

⚡ Why this matters

  • The 'spend at all costs' era that fueled OpenAI and Anthropic is ending.
  • Enterprises now demand clear ROI before approving AI budgets.
  • Cheaper models like DeepSeek get a real opening on price.

🔍 What happened

  • CNBC reported the efficiency shift June 26.
  • Uber blew its entire annual AI budget in four months, then added spend tiers from $1,500 a month.
  • Lindy's CEO moved 100% of traffic off Claude to DeepSeek on cost.
  • Companies are policing employees who max AI on small tasks.
  • Both OpenAI and Anthropic filed confidential IPO papers in early June.

💬 Smart takes

  • D.A. Davidson: big enterprise customers may start limiting out-of-control token spend.
  • Lindy CEO: 'you could see that cost curve crash to the ground.'
  • Skeptic: efficiency talk is cyclical; usage often rebounds once tools prove out.

🧭 Where this goes

  1. Likelyfrontier labs push cheaper, smaller models to defend revenue.
  2. Likely'cost per outcome' replaces token volume as the buying metric.
  3. PossibleDeepSeek and other low-cost rivals win share from price-sensitive teams.
  4. PossibleAI line items get their own FinOps tracking and approval gates.
  5. Wild Carda big public customer drops an AI vendor and dents its growth story pre-IPO.

🥄 The Spoon Take

Tokenmaxxing was always a vanity metric. Usage is not value. As the IPOs near, the question flips from 'how much AI did you use' to 'what did it earn you.' The labs that win next sell outcomes, not tokens. The cheap-model threat is now real.

🤔 Pushback

Efficiency drives come and go; the same buyers cutting spend today often expand it again once a workflow clearly pays off.

·CNBC - Tokenmaxxing to efficiency·TechCrunch - Capping AI budgets·Fortune - Tokenmaxxing is dead