Jassy states that Amazon's chip division could be valued at $50 billion and suggests that there may be plans to sell them to external customers.

      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 growth rates in the triple digits year-over-year. Jassy estimates that if the business were traded publicly like Nvidia, it could be valued at approximately $50 billion annually. He also hints that Amazon might start selling these chips directly to other companies, defending the company’s $200 billion capital expenditure plan for 2026 as based on solid customer demand rather than speculation.

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

      Jassy began the letter by addressing the skepticism regarding Amazon’s capital investments. “We’re not investing approximately $200 billion in capex in 2026 on a hunch,” he stated. “We’re not adopting a conservative approach. We’re positioning ourselves as a leading force, which will result in significantly larger future business, operating income, and free cash flow.” This assertion is made in the context of a substantial decline in free cash flow from $38 billion to $11 billion last year, primarily due to a $50.7 billion rise in capital spending, mostly for AI infrastructure.

      The defense of this spending hinges on existing customer commitments. Jassy highlighted that a significant portion of the 2026 CapEx is backed by customers, citing OpenAI's commitment of over $100 billion to AWS as an example. This commitment, which extends a previous seven-year partnership worth $38 billion established in November 2025, involves OpenAI utilizing around two gigawatts of Trainium capacity through AWS. SoftBank, which has a majority stake in OpenAI and is financing its infrastructure development through a $40 billion bridge loan, is essentially supporting part of the demand that Jassy now references as justification for his CapEx strategy.

      A $50 billion chip business in plain view

      Amazon’s custom silicon initiative encompasses three product lines. Graviton is a tailored CPU that Jassy claims provides over 40% better price-performance compared to similar x86 processors, the market dominated by Intel and AMD. Currently, it is in use by 98% of the top 1,000 EC2 customers, indicating a notable shift in cloud computing economics observed over several years. Demand is so strong that two significant AWS clients inquired about purchasing all available Graviton capacity for 2026, but Amazon declined.

      Trainium serves as the AI training and inference accelerator, representing Amazon's most direct challenge to Nvidia. Trainium2, which offers around a 30% improvement in price-performance over comparable GPUs, is largely sold out. Trainium3, which commenced shipping in early 2026 and boasts an additional 30 to 40% enhancement over Trainium2, is nearly fully booked, with companies like Uber allocating workloads to it. Trainium4, expected to launch in about 18 months and compatible with Nvidia’s NVLink Fusion technology, has already seen significant reservations. Nitro, the custom chip that supports AWS's virtualization framework, completes the triad. Jassy states the three chip lines collectively generate over $20 billion in annualized revenue, growing at triple-digit rates yearly. “If we were a standalone chip company,” he mentions, “our chips would be generating over $50 billion annually.” Currently, this business operates solely within AWS; customers can access Trainium and Graviton through EC2 instances rather than directly purchasing the chips.

      Jassy argues that at scale, Trainium will “save us tens of billions of capex dollars per year and provide us with several hundred basis points of operating margin advantage compared to relying on external chips for inference.” This assertion is central to the rationale for the $200 billion CapEx plan: custom silicon is not just a distinguishing factor but a long-term cost advantage that will grow as the ratio of inference to training in AI workloads rises.

      The Nvidia relationship and the “new shift”

      Jassy thoughtfully articulates the competitive landscape with Nvidia. “We maintain a robust partnership with NVIDIA and will always have customers who opt for NVIDIA,” he states, while also emphasizing that “virtually all AI to date has operated on NVIDIA chips, but a new shift is underway.” Customers are looking for improved price-performance, he points out. Nvidia, which reported $68.1 billion in revenue in the fourth quarter of 2025, a 73% year-over-year increase, began 2026 in a strong market position that Amazon’s custom silicon is gradually eroding from within the AWS client base, rather than through broader market competition. The incorporation of NVLink Fusion in Trainium4 signifies that Amazon is creating a bridge, not a barrier: customers can integrate Trainium accelerators with Nvidia GPUs within the same systems, maintaining flexibility for enterprises that heavily rely on Nvidia’s software ecosystem.

      The letter’s most significant implication regarding chips may lie in a simple statement