OpenRouter hits $1.3B while Uber exhausts its AI budget in four months

OpenRouter closed a $113 million Series B led by CapitalG at a $1.3 billion valuation, more than doubling in twelve months. TechCrunch reports usage grew 5x over six months. The aggregator thesis is working: route prompts to the cheapest inference endpoint that clears the quality bar, bill in one place, let the models compete on price-per-token.

Uber's president Andrew Macdonald told The Verge the company burned through its full-year AI budget four months into 2026 and is now questioning returns. Token consumption for Claude Code climbed but the connection to business outcomes did not show up in the numbers. The quote: AI spending is getting "harder to justify."

This is the divergence. OpenRouter's customers are price-sensitive and model-agnostic. They route. They switch. They arbitrage the spread between Anthropic, OpenAI, and open weights. Uber's AI budget was probably allocated to a specific vendor on an enterprise agreement with soft volume commitments. When token burn outpaces the forecast, the CFO asks what shipped.

The aggregator layer wins when inference commoditizes. The closed provider wins when lock-in holds. Uber's budget exhaustion suggests the lock-in didn't hold or the ROI case was never modeled at realized token rates. OpenRouter's valuation suggests the market believes routing beats loyalty.

Google's AI Search overhaul pushed DuckDuckGo installs up 30 percent, per TechCrunch. Users rejected agent summaries over blue links. The pattern repeats: when the default product forces the AI layer, a segment of users exits to the simpler tool. Inference has a cost. Not every query justifies it.