NextFin News - Amazon CEO Andy Jassy issued a direct challenge to the semiconductor establishment on Thursday, revealing that the company’s internal chip business has reached a $20 billion annual revenue run rate. In his annual shareholder letter, Jassy framed Amazon’s custom silicon—specifically the Trainium AI chips and Graviton CPUs—not merely as internal cost-saving tools, but as a formidable alternative to industry leaders Nvidia and Intel. The disclosure comes as Amazon prepares to deploy a staggering $200 billion in capital expenditures for 2026, a figure that underscores the company’s aggressive pivot toward infrastructure independence.
The scale of demand for Amazon’s proprietary hardware appears to be outstripping supply. Jassy noted that capacity for the upcoming Trainium3 chip is nearly sold out, while the Trainium4, which is not expected to be available for another 18 months, is already seeing similar levels of pre-commitment. This forward-looking demand suggests that cloud customers are increasingly seeking "price-performance" alternatives to Nvidia’s dominant H-series and Blackwell architectures. While Nvidia reported a massive $215.9 billion in revenue last year, Jassy’s narrative suggests a structural shift is underway, postulating that if Amazon sold its chips to third parties rather than utilizing them within AWS, its chip revenue run rate would already sit near $50 billion.
Intel faces an equally pointed threat from Amazon’s Graviton CPUs. According to Jassy, 98% of the top 1,000 EC2 customers now use Graviton, which competes directly with Intel’s x86 architecture. The CEO claimed that two major customers recently attempted to reserve the entirety of Amazon’s Graviton instance capacity for 2026, a request the company had to decline to maintain service for other clients. This high utilization rate among enterprise giants indicates that the transition away from traditional silicon providers is no longer a niche experiment but a core component of cloud strategy for the world’s largest corporations.
The aggressive expansion into hardware is the primary driver behind Amazon’s record-breaking $200 billion capex guidance for 2026. This spending plan, which significantly exceeds the projected investments of peers like Microsoft and Google, has been met with some skepticism by investors. Amazon’s stock has struggled to recover since dipping below the $200 mark earlier this year, as the market weighs the potential for long-term dominance against the immediate pressure on margins. Jassy defended the outlay, asserting that the investment is based on clear demand signals rather than a "hunch," citing a massive $244 billion AWS backlog as evidence of the underlying growth trajectory.
However, the path to unseating established chipmakers is fraught with technical and competitive hurdles. While Amazon’s internal ecosystem provides a guaranteed market for its silicon, Nvidia’s software moat—specifically its CUDA platform—remains a significant barrier for developers who have built their entire AI workflows around Nvidia’s tools. Furthermore, the $200 billion spending spree raises questions about capital efficiency. If the anticipated AI "supercycle" fails to deliver sustained revenue growth, Amazon could find itself with a massive, underutilized infrastructure that weighs on its balance sheet for years.
Beyond the data center, Jassy’s letter signaled broader ambitions that could disrupt other sectors. He confirmed that Amazon’s satellite internet project, Project Kuiper (referred to as Amazon Leo), is on track for a mid-2026 launch and has already secured contracts from Delta Airlines and NASA. He also hinted at a future in commercial robotics, suggesting that the data gathered from the one million robots currently operating in Amazon warehouses could be monetized as "robotics solutions" for industrial and consumer use. For now, the focus remains on the silicon war, where Amazon is betting that its scale can finally break the hardware monopolies that have defined the first phase of the AI era.
Explore more exclusive insights at nextfin.ai.
