• Uber adopts Amazon’s custom chips to cut AI computing costs and reduce Nvidia reliance
  • Move signals industry shift towards cloud providers’ proprietary silicon for large-scale AI

What happened

Uber, a ride-hailing and food delivery platform, deepens AWS partnership to run AI workloads on custom chips amid rising compute costs

Uber is expanding its use of Amazon Web Services’ (AWS) custom-designed chips to power its artificial intelligence workloads. The company is leveraging AWS’s specialised silicon—widely understood to include Trainium and Inferentia processors—to improve efficiency in training and deploying AI models.

The shift comes as Uber seeks to optimise the cost-performance balance of its AI operations, which underpin core services such as ride matching, pricing algorithms and delivery logistics. By using AWS’s chips, Uber aims to reduce its dependence on more expensive, general-purpose GPUs typically supplied by Nvidia.

Amazon, a US e-commerce and cloud computing giant, has positioned its custom chips as a lower-cost alternative for large-scale AI workloads, particularly for inference tasks and certain training scenarios. Uber’s adoption reflects a broader trend among tech firms looking to diversify their compute stack as demand for AI infrastructure surges.

Why it’s important

As AI demand drives chip shortages and rising cloud costs, companies adopting hyperscaler silicon gain pricing leverage and architectural flexibility, reshaping competitive dynamics in AI infrastructure.

Uber’s move highlights a structural shift in the AI ecosystem: cloud providers are no longer just infrastructure vendors but increasingly compete at the silicon layer. By adopting AWS’s custom chips, Uber not only reduces cost exposure but also aligns more closely with Amazon’s vertically integrated AI stack. This could accelerate innovation cycles while reinforcing vendor lock-in risks.

More broadly, the decision underscores intensifying competition between proprietary cloud chips and Nvidia’s dominant GPU ecosystem. As hyperscalers refine their in-house processors, enterprises may increasingly adopt hybrid compute strategies—balancing performance, cost and availability—to sustain AI growth at scale.

Also read: Broadcom and Google strike long-term deal on custom AI chips

Also read: Meta unveils four custom chips to power AI and recommendations