Amazon Moves to Sell Its Custom AI Chips to Challenge Nvidia's Market Dominance

The move stems from Amazon CEO Andy Jassy's annual shareholder letter in April, in which he noted that if AWS's homegrown chip business were standalone and sold chips produced that year to both AWS and third parties, the annual run rate would be approximately $50 billion. Jassy further stated that there was so much demand for the chips that AWS would likely sell racks of them to third parties in the future. AWS spokesperson Doron Aronson confirmed to TechCrunch that while the company has historically declined requests to sell chips directly, it's quite possible AWS will sell racks of them to third parties.

The talks are in early stages, and Amazon has not disclosed specific potential buyers or a launch timeline. The company's fabrication partner is likely TSMC, though TechCrunch notes that AWS would need to secure additional wafer allocation to supply external customers without reducing capacity for its own cloud infrastructure.

Current Market Context and Scale

Nvidia currently controls an estimated 70 to 80 percent of the AI accelerator market, with pricing power directly tied to the absence of credible competition at scale. A $50 billion competitor would not be insignificant—it would be comparable to Intel's annual revenue—though Nvidia is currently on a $326 billion revenue run rate. However, the strategic importance extends beyond raw market share. Amazon's move introduces a second large-scale supplier with deep manufacturing and logistics capabilities, something that could reshape the competitive dynamics of the AI infrastructure market.

Amazon's historical resistance to selling chips externally has been rooted in the economics of its cloud business model. AWS makes money through a waterfall effect: while it charges customers directly for AI tokens processed on its cloud, it also charges for storage, security, networking, and monitoring services. Additionally, Amazon has touted that its chip capacity has been selling out faster than it can produce, with Jassy noting in the shareholder letter that Trainium capacity sold out almost instantly, as did capacity for the next generation, Trainium4, which won't be available for more than a year.

Manufacturing Constraints and Competitive Implications

The fundamental challenge facing Amazon's external chip ambitions lies in semiconductor manufacturing bottlenecks. Advanced semiconductor fabrication requires facilities that cost over $20 billion to build and years to qualify for production. Both Amazon and Nvidia depend on TSMC for manufacturing, meaning that while Amazon has developed differentiated chip designs, it ultimately faces the same foundry constraints as its competitors. TSMC has recently supplanted Apple to become the foundry's largest customer, and competing for wafer allocation against Nvidia—TSMC's primary customer—represents the central tension in Amazon's strategy.

This manufacturing dependency creates a paradox: Amazon's ability to challenge Nvidia at the chip sales level depends on securing sufficient TSMC capacity, but TSMC's allocation decisions reflect market demand and customer relationships. The foundry concentration means that supply-chain fragility, which has defined AI infrastructure buildouts since 2022, depends on a single manufacturing partner. The risks tied to this concentration include both inflated pricing and extended build timelines, as customers lack alternatives when a single supplier controls production capacity.

Why This Matters for the AI Infrastructure Market

Amazon's potential entry into external chip sales addresses one of the most critical supply-side constraints in AI infrastructure. Since 2023, the concentration of chip supply has been a central bottleneck for enterprises and hyperscalers attempting to deploy large-scale AI systems. A credible second supplier with manufacturing scale, customer relationships, and logistics infrastructure could disrupt both Nvidia's pricing power and the supply shortage that has characterized the market. For customers already using AWS services, purchasing Trainium chips externally would extend Amazon's ecosystem advantage—they could potentially negotiate volume discounts or integrated service packages that leverage AWS's broader cloud offerings.

However, the competitive impact depends on execution. Amazon must successfully secure wafer allocation, validate chip performance against Nvidia's latest architectures, and convince data center operators and enterprises that Trainium chips deliver sufficient performance and reliability advantages to justify migration costs. The company has strong incentives to succeed: if AWS can establish itself as a credible chip supplier, it diversifies its revenue stream and strengthens its competitive moat by reducing customer dependence on Nvidia.

What to Watch

The critical near-term milestone is any formal product announcement detailing external chip availability, pricing, and performance specifications. Watch for announcements around Trainium 3, which is expected to use TSMC's 3nm-class process node—a step that would narrow the performance gap with Nvidia's latest architectures. Any disclosure about TSMC wafer allocation or manufacturing partnerships would clarify whether Amazon can produce chips at the volume required to serve both internal and external demand. Finally, monitor which customers—whether hyperscalers, enterprises, or specialized AI infrastructure companies—become the first external buyers. Early adoption by major customers would signal genuine market traction and competitive viability against Nvidia.