Amazon plots direct challenge to Nvidia by opening Trainium chip sales to third parties

AWS Signals Shift on Chip Sales Strategy

Amazon Web Services is exploring the sale of its custom-designed Trainium AI chips to data center operators outside the AWS ecosystem, according to reporting by TechCrunch. Amazon AI Chief Peter DeSantis disclosed the discussions to Bloomberg but declined to name potential external buyers. According to TechCrunch, these negotiations remain preliminary and stem from a strategic reorientation outlined by Amazon CEO Andy Jassy in his April 2026 shareholder letter.

Jassy’s April filing contained a striking valuation claim: if AWS’s chip business operated as an independent unit and sold inventory produced in 2026 to both AWS and third parties “as other leading chips companies do,” it would generate approximately $50 billion in annual run rate. This projection underscore AWS’s shift from treating chips as a proprietary competitive moat to treating them as a potential standalone revenue stream.

Manufacturing Constraints Versus Market Demand

The paradox underpinning AWS’s deliberation is severe capacity oversubscription paired with constrained supply. According to TechCrunch, Jassy stated in April that current Trainium capacity sold out almost instantly, and pre-orders for the forthcoming Trainium4 (unavailable for more than a year) had already exhausted future allocation. This scarcity exists despite adding OpenAI to the models AWS serves, compounding the pressure on production.

Selling external inventory would require AWS to dramatically increase foundry capacity beyond current manufacturing partnerships, particularly with TSMC. A critical complication: TSMC, according to TechCrunch, recently surpassed Apple to become the foundry’s largest customer—a position largely driven by Nvidia’s GPU orders. Any attempt by Amazon to negotiate surplus Trainium production would compete directly for TSMC’s finite wafer capacity, a negotiation likely favoring the larger, more established chip buyer.

Competitive Positioning Relative to Nvidia

The $50B run-rate estimate positions Amazon as a substantial semiconductor player—comparable to Intel’s annual revenues—but does not immediately threaten Nvidia’s $326B current run rate. However, according to TechCrunch, Nvidia CEO Jensen Huang has already signaled expansion beyond GPUs into AI CPUs, a territory that overlaps with traditional x86 markets held by Intel and AMD. This suggests the competitive landscape is fragmenting: Nvidia expanding upmarket into CPUs; Amazon attempting to establish external chip sales; and traditional server chip makers defending margins.

AWS spokesperson Doron Aronson confirmed to TechCrunch that the company “may sell racks of [chips] to third parties in the future,” a reversal from AWS’s historical practice of reserving chips exclusively for internal cloud services. This shift signals Amazon’s willingness to optimize for revenue and capacity utilization over vertical integration—a strategic gamble that hinges on manufacturing availability and customer adoption rates.

Why This Matters

If Amazon successfully scales third-party chip sales, it fragments Nvidia’s near-monopoly on accelerator supply and forces cloud-native AI infrastructure buyers to evaluate a credible alternative. Teams already embedded in AWS (and facing queue delays for Trainium capacity) gain relief; prospective customers evaluating long-term GPU economics gain leverage over Nvidia’s pricing. However, the outcome depends entirely on TSMC capacity allocation: without sufficient wafer supply, Amazon cannot materially dent Nvidia’s dominance. Watch for AWS’s next earnings call disclosures on chip availability and manufacturing partnerships—those metrics will signal whether this is a credible competitive move or a revenue aspiration unmatched to supply-chain reality.