Uber's Q3 Reveals AI Agent Adoption Still Nascent, but GPU Demand Looms

Uber's third-quarter earnings call offered a reality check on the much-hyped threat of AI-agent disintermediation. Executives revealed that third-party bots accounted for a mere four rides and approximately 500 Uber Eats orders during the Halloween week, a negligible slice of the 320 million total transactions on the platform.

The numbers underscore how far current agent ecosystems — such as Meta's Muse or OpenAI's Dots — remain from replacing direct app usage. Friction in authentication, payment, and real-time decision-making still steers users toward Uber's native experience, a dynamic the company expects to persist in the near term.

Rather than ceding ground, Uber is doubling down on first-party AI. The recently launched Cart Assistant for Uber Eats, which builds grocery lists from recipe images or text prompts, exemplifies a strategy to keep engagement inside its own interface while still leaving the door open for future third-party integrations.

Why it matters for GPU / AI infrastructure

Today's low agent volumes mask a looming compute surge. Training and serving the large language models that power consumer-facing agents demand massive GPU clusters. As Uber, DoorDash, and other platforms roll out proprietary AI features and eventually open APIs for external agents, inference workloads will scale exponentially. For GPU cloud providers like AiGpu, this translates into sustained, high-density demand for low-latency inference capacity — especially in regions where delivery and mobility super-apps operate.

The trajectory is clear: agent-mediated transactions will grow, and the underlying infrastructure must be ready. Early movers who secure scalable, sovereign GPU supply will capture the next wave of AI-driven commerce.