Can Amazon Take Market Share From Nvidia in the AI Chip Race?

Amazon released software enabling its Trainium chips to run code written in a popular AI coding language, potentially making them more competitive with Nvidia's chips. Amazon said its chip business reached a $25 billion annual revenue run rate, slightly more than one-quarter of Nvidia's most recent quarterly revenue.

Analysis

The economic lever is lower software friction, not chip performance by itself. If the tool lets developers reuse existing code with limited rework, AWS can make Trainium a credible price/performance option for workloads already hosted on AWS. That could improve AWS’s bargaining power against GPU suppliers and shift some incremental AI infrastructure spend toward Amazon’s own silicon. The second-order benefit may be broader AWS retention: customers choosing its chips may become more embedded in its data, networking, and cloud services.

This is not yet evidence of material displacement of NVIDIA. Compatibility does not establish parity in performance, reliability, developer experience, or total cost once porting and operations are included; NVIDIA’s ecosystem and installed base remain the key counterweight. Nor does a chip revenue run rate establish contribution to earnings: verify gross profit, utilization, customer mix, and whether sales are internal or external before extrapolating value to AMZN. The run-rate/quarterly-revenue comparison is not like-for-like.

Near term, sentiment may favor AMZN and weigh on NVDA’s multiple at the margin, but a durable relative move needs customer adoption and workload-level economics. Over 1–3 months, watch independent benchmarks, production deployments, and AWS disclosures. Over 6–18 months, successful substitution could reduce AWS’s reliance on merchant GPUs and strengthen cloud differentiation. The thesis weakens if porting remains costly, Trainium capacity or performance disappoints, or AWS customers continue selecting NVIDIA despite the new tooling.

Key Decisions for Investors

• Do not chase a headline-driven AMZN/NVDA pair immediately. Put a relative-value position on the watchlist; consider a modest long-AMZN/short-NVDA expression only after evidence of production workloads moving to Trainium, with position sizing tied to the risk that this is tooling progress without meaningful substitution.
• Over the next 1–3 months, verify developer adoption, independent performance and total-cost benchmarks, production availability, and AWS customer references. Treat company-reported revenue run rate as insufficient without profitability, utilization, and customer-mix disclosure.
• Falsify the substitution thesis if customers still require substantial code changes, Trainium falls short on workload-level performance or reliability, or AWS disclosures show no meaningful uptake; reassess any relative position if those indicators emerge.
• For NVDA, distinguish a potential share-loss narrative from near-term earnings risk: absent evidence of workload migration, avoid treating compatibility alone as a reason to short. Monitor NVIDIA guidance and cloud-provider demand for confirmation.