Jassy indicates that Amazon's chip division could be valued at $50 billion and suggests that it might sell the chips to external customers.

      In summary: Andy Jassy's annual letter to shareholders, released on April 9, 2026, indicates that Amazon's custom chip business, which includes Graviton, Trainium, and Nitro, generates over $20 billion in annual revenue, experiencing year-on-year growth at triple-digit rates. According to Jassy, if this business were sold on the open market like Nvidia, it would be valued at approximately $50 billion annually. He also indicates that Amazon may start selling these chips directly to third parties and defends the company's $200 billion capital expenditure plan for 2026, emphasizing that it is based on confirmed customer demand rather than assumptions.

      “Not on a hunch”: the $200 billion wager

      Jassy commenced the financial section of the letter by directly addressing the doubts surrounding Amazon’s capital commitments. “We’re not investing approximately $200 billion in capex in 2026 on a hunch,” he stated. “We’re committed to being a significant leader, and as a result, our future business, operating income, and free cash flow will be substantially greater.” This assertion comes from a backdrop where the company saw its free cash flow drop from $38 billion to $11 billion last year due to a $50.7 billion surge in capital expenditures, primarily allocated for AI infrastructure.

      The justification is rooted in established customer commitments. Jassy mentioned that a significant portion of the expected CapEx for 2026 has backing from customers, highlighting OpenAI’s more than $100 billion commitment to AWS as an example. This commitment builds on an existing $38 billion seven-year partnership initiated in November 2025, which includes OpenAI utilizing approximately two gigawatts of Trainium capacity through AWS infrastructure. SoftBank, which controls a majority share in OpenAI and is funding its infrastructure expansion via options like a $40 billion bridge loan, effectively supports the demand that Jassy cites as evidence for his CapEx position.

      A $50 billion chip venture in plain sight

      Amazon’s custom silicon initiative encompasses three product lines. Graviton is a custom CPU that Jassy states offers over 40% better price-performance compared to similar x86 processors, which are predominantly dominated by Intel and AMD. Currently, it is employed by 98% of the top 1,000 EC2 customers, indicating a significant shift in cloud computing economics over recent years. The demand is so high that two large AWS clients inquired if they could purchase all available Graviton capacity for 2026, which Amazon denied.

      Trainium serves as the AI training and inference accelerator that provides Amazon’s most direct competition to Nvidia. Trainium2, as described by Jassy, delivers about 30% better price-performance than equivalent GPU options and is largely sold out. Trainium3, which began shipping in early 2026 and offers an additional 30 to 40% improvement in price-performance over Trainium2, is nearly fully reserved, with Uber being one of the companies that have migrated workloads to it. Trainium4, which will be widely available in approximately 18 months and features interoperability with Nvidia’s NVLink Fusion interconnect technology, has also seen significant reservations. Nitro, the custom network and security chip essential for AWS's virtualisation layer, completes the three-product lineup. Together, Jassy notes, these three lines generate over $20 billion in annual revenue, with year-on-year growth at triple-digit percentage rates. “If we were a standalone chip company,” he states, “our chips would be producing over $50 billion in annual revenue.” Currently, this business operates entirely within AWS, with customers accessing Trainium and Graviton via EC2 instances rather than purchasing the chips outright.

      At scale, Jassy contends that Trainium will "save us tens of billions of capex dollars per year and offer several hundred basis points of operating margin advantage compared to relying on third-party chips for inference.” This claim is essential to the investment rationale supporting the $200 billion CapEx initiative: custom silicon serves as both a competitive differentiator and a structural cost advantage that will accumulate over time as the balance of inference to training in AI workloads continues to shift.

      The relationship with Nvidia and the “new shift”

      Jassy is deliberate in articulating the competitive landscape with Nvidia. “We have a strong partnership with NVIDIA and will always have customers who prefer to utilize NVIDIA,” he states, while also asserting that “virtually all AI to this point has been conducted on NVIDIA chips, but a new shift has begun.” Customers, he notes, “seek improved price-performance.” Nvidia, which reported $68.1 billion in revenue in the fourth quarter of 2025, marking a 73% year-on-year increase, entered 2026 from a position of market strength that Amazon's custom silicon is incrementally eroding within the AWS customer base instead of the wider merchant market. Trainium4's compatibility with NVLink Fusion indicates that Amazon is also establishing a bridge