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

      In summary: In his annual letter to shareholders released on April 9, 2026, Andy Jassy disclosed that Amazon's proprietary chip business, encompassing Graviton, Trainium, and Nitro, is generating over $20 billion in annual revenue with year-on-year growth rates in the triple digits. Jassy indicated that if this business were publicly traded like Nvidia, it would be valued at around $50 billion annually. He also hinted that Amazon might start selling these chips directly to outside parties and justified the company's $200 billion capital expenditure plan for 2026, asserting it is based on confirmed customer demand rather than speculation.

      "Not based on a guess": the $200 billion wager

      Jassy began the financial portion of the letter with a strong rebuttal to the skepticism surrounding Amazon’s capital investments. “We’re not making a $200 billion capex investment in 2026 based on a guess,” he stated. “We won’t adopt a conservative approach. We’re investing to be a significant leader, and as a result, our future business, operating income, and free cash flow will be considerably larger.” This statement comes from a company that faced a drop in free cash flow from $38 billion to $11 billion the previous year due to a $50.7 billion rise in capital spending, primarily allocated to AI infrastructure.

      The argument is supported by existing customer commitments. Jassy mentioned that a significant portion of the expected 2026 CapEx has already been backed by customers, citing OpenAI’s more than $100 billion commitment to AWS as an example. This commitment expands on a pre-existing $38 billion partnership established in November 2025 and includes OpenAI utilizing about two gigawatts of Trainium capacity via AWS infrastructure. SoftBank, which owns a majority stake in OpenAI and has been funding its infrastructure through various means, including a $40 billion bridge loan, is effectively backing some of the demand Jassy references to justify his CapEx position.

      A $50 billion chip business in plain sight

      Amazon's custom silicon initiative consists of three product lines. Graviton is a tailored CPU that Jassy claims outperforms comparable x86 processors by over 40% in price-performance, dominating 98% of the top 1,000 EC2 customers. This shift reflects a long-term change in cloud computing economics. The demand is so high that two major AWS customers inquired about purchasing all available Graviton capacity for 2026, which Amazon declined.

      Trainium serves as the AI training and inference accelerator directly competing with Nvidia. According to Jassy, Trainium2 delivers about 30% better price-performance than competing GPUs and is mostly sold out. Trainium3, which began shipments in early 2026, offers an additional 30 to 40% increase in price-performance over Trainium2 and is nearly fully booked, with companies like Uber already migrating workloads to it. Trainium4, which is approximately 18 months away from wider availability and will feature compatibility with Nvidia’s NVLink Fusion technology, has already seen significant reservations. Nitro, the custom security and networking chip that supports AWS's virtualization layer, rounds out the trio. Jassy stated that these three lines generate over $20 billion in annual revenue with year-on-year growth rates in the triple digits. “If we were an independent chip company,” he remarked, “our chips would yield over $50 billion in annual revenue.” This business exists solely within AWS at the moment; customers utilize Trainium and Graviton via EC2 instances instead of purchasing chips directly.

      Jassy argues that at scale, Trainium will “save us tens of billions of capex dollars yearly and provide several hundred basis points of operating margin advantage compared to relying on other chips for inference.” This assertion lies at the core of the rationale behind the $200 billion CapEx plan: custom silicon is not merely a competitive edge but a structural cost benefit that accrues over time as the ratio of inference to training in AI workloads continues to rise.

      The relationship with Nvidia and the “new shift”

      Jassy is deliberate in presenting the competitive landscape with Nvidia. “We maintain a strong partnership with NVIDIA, and there will always be customers who opt to use NVIDIA,” he stated, while also claiming that “virtually all AI thus far has operated on NVIDIA chips, but a new shift has begun.” Customers, he notes, “are seeking better price-performance.” Nvidia, which reported a revenue of $68.1 billion in the fourth quarter of 2025, increasing 73% year-on-year, began 2026 from a strong market position that Amazon's custom silicon is gradually eroding within its AWS user base rather than in the wider merchant market. The inclusion of NVLink Fusion in Trainium4 indicates that Amazon is creating a bridge instead of a wall: customers can combine Trainium accelerators with Nvidia GPUs in the same system, retaining flexibility for enterprises heavily invested in Nvidia's software.

      However,