Amazon's custom chip business, encompassing its Graviton processors and Trainium AI accelerators, has achieved an annual revenue run rate exceeding $20 billion, demonstrating triple-digit percentage year-over-year growth. This significant milestone was revealed by CEO Andy Jassy in his recent letter to shareholders, highlighting an extraordinary demand for these proprietary chips that is currently outstripping supply. The robust performance of Amazon Web Services (AWS) is closely tied to this internal chip development, as the company navigates a period of intense innovation in artificial intelligence.
Jassy's letter underscored the overwhelming market appetite for Amazon's custom silicon. He stated, "two large AWS customers have already asked if they could buy all of our Graviton instance capacity in 2026," a request Amazon could not fulfill due to existing customer commitments. This anecdote illustrates the severe capacity constraints AWS faces despite its rapid expansion in power infrastructure, with plans to double total power capacity by the end of 2027. The CEO emphasized that the company's chips business is "on fire" and is poised to become "much larger than most think."
The strategic importance of these custom chips extends beyond merely meeting internal AWS needs. Jassy projected that if Amazon's chip business operated as a standalone entity, selling to third parties, its annual run rate could reach approximately $50 billion. This indicates a potential future where Amazon could become a significant external supplier in the semiconductor market. The company's Trainium2 and Trainium3 AI chips are already largely sold out or nearly fully subscribed, with a substantial portion of Trainium4 (due in about 18 months) already reserved.
The development of proprietary silicon like Graviton and Trainium is central to Amazon's strategy to enhance price-performance for customers and improve AWS's economic model. Graviton processors offer up to 40% better price-performance than traditional x86 processors, while Trainium2 boasts a 30% advantage over comparable GPUs. This focus on in-house chip design aims to reduce reliance on external providers and secure a competitive edge in the rapidly evolving cloud and AI landscape.
