Key Takeaways:
Amazon's cloud unit is accelerating its custom chip push, raising third-quarter ASIC server shipments by as much as 30% as Anthropic's insatiable compute demand outpaces even 10-times growth planning.
Amazon.com Inc.'s AWS raised its third-quarter ASIC server shipment target by 20% to 30%, signaling accelerating adoption of its custom Trainium chips as Anthropic's compute needs outpace even aggressive growth forecasts.
"The buildout of AI factories, the largest infrastructure expansion in human history, is accelerating at extraordinary speed," Jensen Huang, chief executive officer of Nvidia Corp., said on the company's most recent earnings call.
Trainium 3 server components began shipping in May and are ramping monthly, with rack-level orders and rail components entering mass production in July, according to Taiwan-based supply chain sources cited by Digitimes. Trainium 2 has sold out, Trainium 3 is nearly fully booked, and some customers have already begun pre-ordering Trainium 4, Amazon CEO Andy Jassy said on the company's earnings call. AWS also started development of Trainium 5.
The ramp threatens Nvidia's dominance in AI inference chips, where AWS's custom silicon offers lower cost for large-scale workloads. DIGITIMES Research projects ASIC server shipments will grow 64.2% in 2026, outpacing GPU server growth of 43.8%. Amazon trades at roughly 22 times forward earnings, and in-house chips could meaningfully reduce its reliance on external GPU suppliers.
Anthropic's Insatiable Demand
Anthropic, in which Amazon is a major investor, is the primary force behind the accelerated timeline. The AI company has said its compute capacity planned on a 10-times growth basis still falls short of what it needs. The two companies deepened their partnership in April 2026 with a 10-year agreement, with Anthropic committing to expand its AWS compute purchases. OpenAI and Uber are also Trainium customers, and AWS's Bedrock managed AI platform now serves 125,000 enterprise clients, with inference workloads running primarily on Trainium.
Competitive Pressure Intensifies
The acceleration comes as competition in cloud AI intensifies. Alphabet Inc.'s Google Cloud, which leads in ASIC chip shipments with its Tensor Processing Unit, grew 63% to $20 billion in the most recent quarter. Meta Platforms Inc. recently announced plans to open its AI compute to external customers, directly competing with AWS, Microsoft Azure and Google Cloud for enterprise workloads. For AWS, custom chips offer a path to better margins and reduced dependency on Nvidia, whose data center revenue reached $75 billion last quarter with 75% gross margins. The Taiwan supply chain — including thermal solution providers Auras Technology and Mark, rail makers King Slide and Nan Juen, chassis suppliers Chenbro, and board assembler Accton Technology — is positioned to benefit from the higher shipment volumes.
This article is for informational purposes only and does not constitute investment advice.
