Amazon Prepares to Compete with NVIDIA in the AI Chip Market
Fifty billion dollars. That is the projected annual revenue Amazon could generate by selling its custom Trainium AI chips to external customers — a figure roughly equivalent to Intel's entire yearly intake, and a direct shot across NVIDIA's $326 billion bow.
The cloud giant is no longer content to keep its silicon in-house.
The Shift from Walled Garden to Open Market
Until recently, AWS deliberately withheld Trainium from the open market. The logic was straightforward: customers using Amazon's custom chips were also tethered to its data storage, networking, and security stack. Lock-in drove recurring cloud revenue. That calculus is now changing.
AWS AI chief Peter DeSantis has confirmed, according to Bloomberg, that negotiations are underway to sell Trainium chips to third-party companies for deployment in their own data centers. Amazon CEO Andy Jassy flagged the pivot in his April shareholder letter, citing demand so intense that current Trainium capacity — including next-generation Trainium4 units expected to ship next year — is already fully booked. The company is weighing the sale of complete server rack systems, not just individual chips, as confirmed by AWS representative Doron Aronson.
Production Bottleneck, Not Demand Problem
The constraint is not market appetite. It is fabrication. Amazon depends on TSMC for chip manufacturing, which means it is now competing head-to-head with NVIDIA and Apple for foundry capacity — a contest that will test supply chain leverage and procurement strategy.
Scaling production sharply enough to serve both internal AWS workloads and a new external customer base is the core engineering and logistics challenge. If Amazon can solve it, the competitive dynamics of the AI hardware market shift measurably: a credible second source for high-performance AI accelerators would introduce pricing pressure where NVIDIA currently operates with near-monopoly margins.
What This Means for the Ecosystem
For the broader technology landscape, the signal is structural. AI infrastructure costs remain the primary barrier to entry for startups, researchers, and enterprises outside the Fortune 500. An alternative supplier with Amazon's manufacturing partnerships and capital reserves could yield meaningful cost reductions — not through charity, but through basic competition.
The $50 billion revenue projection is not yet realized. It hinges on Amazon's ability to ramp TSMC orders, build out distribution channels for enterprise hardware, and prove Trainium's performance parity in diverse, non-AWS environments. What is clear: the AI chip market is moving from single-supplier dominance to a two-player contest. The data points to Amazon taking that transition seriously — with billions already committed and customer demand outpacing supply.