Amazon CEO Andy Jassy’s yearly letter to shareholders can be interpreted as a subtle challenge to competitors, akin to a musical “diss track” but delivered in corporate language rather than artistic verse.
To fully grasp the rivals Jassy addresses, one needs to be familiar with the industry landscape, though his letter also includes charming anecdotes, such as his youthful ambition to be a sportscaster and enjoying hockey with his father.
Jassy avoids direct confrontation, opting for a subtle strategy. For example, when addressing Nvidia, he states, “We have a strong partnership with NVIDIA, will always have customers who choose to run NVIDIA” and commits to supporting their chips within Amazon’s cloud infrastructure.
However, he also highlights: “While NVIDIA chips have historically dominated AI development, a new trend is emerging.” Jassy notes that AWS clients are seeking “improved price-performance,” which points to Amazon’s proprietary Trainium AI chips.
According to Jassy, the demand for this particular chip is exceptionally high, with the latest version, Trainium3, almost completely booked. Even more surprisingly, Trainium4, which won’t be released for another 18 months, is also nearing full capacity.
Consequently, Amazon’s chip division has achieved an annual revenue run rate of $20 billion. Jassy further speculates that if Amazon functioned as a standalone chip manufacturer selling to external clients, its annual recurring revenue could reach $50 billion.
Nevertheless, Nvidia reported $215.9 billion in actual revenue last year, suggesting Nvidia isn’t likely feeling threatened just yet. Despite this, Jassy positions Trainium as a promising and significant emerging contender.
Intel was also a focus of Jassy’s address. He emphasized that AWS’s internally developed Graviton CPU, which competes with Intel’s x86 architecture, “is now extensively adopted by 98% of the top 1,000 EC2 customers,” representing some of the globe’s largest corporations. He noted that two firms even sought to “acquire all of our Graviton instance capacity in 2026” (his emphasis). He clarified, “We cannot fulfill these requests due to the needs of other clients, but it illustrates the level of demand.”
Jassy further asserted that Amazon Leo, the company’s rival to Starlink, set to debut in mid-2026, is already demonstrating success. It has secured agreements with major entities including Delta Airlines, AT&T, Vodafone, Australia’s National Broadband Network, and NASA.
Intriguingly, he mentioned the possibility of Amazon venturing into robotics sales in the future. He suggested that data gathered from its 1 million warehouse robots could be leveraged to develop “robotics solutions” for both industrial and consumer markets. The question remains whether an Amazon humanoid robot is on the horizon. He also highlighted other Amazon ventures, such as same-day delivery, groceries, and drone services.
Primarily, Jassy aimed to justify the massive capital expenditures he has approved. In February, he revealed plans for $200 billion in capital spending in 2026, largely dedicated to expanding AWS data centers. This outlay surpasses that of other leading tech firms, who are also investing heavily in capital projects. Jassy’s appeal to shareholders is understandable, given Amazon’s stock drop below $200 per share and its failure to rebound.
“We are not making an approximate $200 billion capital investment in 2026 based on speculation,” he asserted, citing the agreement with OpenAI which includes a commitment for the AI model developer to spend $100 billion on AWS. However, some observers question whether OpenAI will fulfill all its spending pledges.
Addressing such skepticism, Jassy emphasized that in addition to OpenAI, “numerous other customer agreements have been finalized (and are currently confidential) or are far along in negotiations,” all geared towards utilizing AWS capacity.
Only time will tell. Individuals contributing to a market bubble rarely acknowledge its presence. Jassy acknowledged, “I’ve monitored the public discussion regarding whether this technology is exaggerated or if we are experiencing ‘a bubble.’” Yet, in his letter, he asserts that this does not apply to Amazon’s situation.
Correction: This article has been revised to clarify that the projected annual recurring revenue for Amazon’s proprietary chip sector encompasses various chips, not solely Trainium.