According to Jassy, Amazon's chip division could be valued at $50 billion, and he suggests that there is a possibility of selling them to external clients.

      In summary: Andy Jassy's annual letter to shareholders, released on April 9, 2026, indicates that Amazon's custom chip division, which includes Graviton, Trainium, and Nitro, generates over $20 billion in annual revenue, with year-on-year growth rates in the triple digits. If this business were sold like Nvidia's, Jassy estimates it would be valued at around $50 billion annually. He hints that Amazon might start selling these chips directly to third parties and justifies the company's $200 billion capital expenditure plan for 2026 as being based on committed customer demand rather than mere speculation.

      “Not based on a guess”: the $200 billion investment

      Jassy began the financial discussion by directly addressing the doubts surrounding Amazon's capital investments. “We’re not investing approximately $200 billion in capex in 2026 based on a guess,” he stated. “We’re not going to take a cautious approach. We’re investing to become a significant leader, and as a result, our future business, operating income, and free cash flow will be considerably larger.” This assertion comes from a company that experienced a drop in free cash flow from $38 billion to $11 billion last year, attributed to a $50.7 billion rise in capital expenditures, mainly directed towards AI infrastructure.

      The defense relies on existing customer commitments. Jassy pointed out that a significant portion of the expected CapEx deployment in 2026 has customer backing, highlighting OpenAI’s commitment of over $100 billion to AWS as an example. This commitment, which extends an existing $38 billion seven-year partnership established in November 2025, also involves OpenAI utilizing around two gigawatts of Trainium capacity through AWS infrastructure. SoftBank, which owns a majority stake in OpenAI and has been funding its infrastructure development through various means, including a $40 billion bridge loan, is essentially supporting the demand that Jassy references as justification for his CapEx position.

      A $50 billion chip enterprise in plain sight

      Amazon’s custom silicon initiative encompasses three product lines. Graviton is a tailored CPU that Jassy claims offers over 40% better price-performance than similar x86 processors, a market dominated by Intel and AMD. It is currently utilized by 98% of the top 1,000 EC2 clients, indicating a significant shift in the cloud computing economics over recent years. The demand is so robust that two major AWS clients inquired about acquiring all available Graviton capacity for 2026, which Amazon declined.

      Trainium is the AI training and inference accelerator that represents Amazon’s most direct competition to Nvidia. Jassy states that Trainium2 delivers approximately 30% better price-performance compared to rival GPU options and is largely sold out. Trainium3, which started shipping in early 2026, provides an additional 30 to 40% improvement over Trainium2 and is almost fully booked, with Uber among the companies migrating workloads to it. Trainium4, which is about 18 months away from widespread availability and integrates with Nvidia’s NVLink Fusion interconnect technology, has already seen significant reservations. Nitro, the custom network and security chip supporting AWS's virtualization layer, rounds out the trio. Jassy notes that collectively, these three product lines generate over $20 billion in annual revenue, experiencing triple-digit percentage growth annually. “If we were a standalone chip firm,” he remarks, “our chips would be producing over $50 billion in annual revenue.” Currently, this business operates exclusively within AWS; clients access Trainium and Graviton through EC2 instances rather than purchasing the chips outright.

      At scale, Jassy contends, Trainium will “save us tens of billions in capex annually and provide several hundred basis points of operating margin advantage compared to relying on third-party chips for inference.” This assertion is central to the rationale supporting the $200 billion CapEx strategy: custom silicon not only differentiates Amazon but also creates a structural cost advantage that accumulates over time as the ratio of inference to training in AI workloads continues to increase.

      The Nvidia relationship and the “new paradigm”

      Jassy is cautious in how he depicts the competitive landscape with Nvidia. “We have a strong partnership with NVIDIA and will always have customers who prefer NVIDIA,” he states, while also claiming that “nearly all AI up to this point has relied on NVIDIA chips, but a new shift is beginning.” Customers are expressing a desire for better price-performance, he notes. Nvidia, which reported earnings of $68.1 billion in the fourth quarter of 2025—an increase of 73% year-on-year—entered 2026 with significant market dominance that Amazon's custom silicon is gradually eroding within its AWS customer base rather than in the broader merchant market. The inclusion of NVLink Fusion in Trainium4 indicates that Amazon is fostering integration rather than isolation: customers can mix Trainium accelerators with Nvidia GPUs in the same system, maintaining options for enterprises