Amazon Web Services is intensifying its strategic focus on custom-designed chips. Supply chain sources indicate that AWS has informed its suppliers of plans to increase third-quarter ASIC server shipments by 20% to 30% above the original schedule, signaling strong confidence in the market outlook for its Trainium 3 processors.
According to a report citing sources within the Taiwanese supply chain, motherboard-level components for servers equipped with the Trainium 3 chip began shipping in May and are currently in a phase of monthly production ramp-up. This upward revision by AWS is interpreted by industry suppliers as a move to secure orders early and capture market share, rather than merely a routine demand adjustment.
Multiple supply chain sources have pointed out that Anthropic is a key driver behind this accelerated procurement. The AI company has reportedly stated that its computing power planning, based on a tenfold growth benchmark, remains insufficient and requires urgent scaling to fill the gap.
The news has positively impacted several Taiwanese companies within the AWS Trainium 3 server supply chain, with potential for upward revisions to their operational performance in the second half of the year. Concurrently, industry attention is turning to whether AWS's strong market momentum might lead to an earlier-than-expected launch of the next-generation Trainium 4, further strengthening its competitiveness in the ASIC market.
Supply Chain Enters Full-Scale Production Ramp-Up
The Taiwanese supply chain for AWS Trainium 3 servers has established a relatively comprehensive ecosystem. This includes companies like Auras, Macase, and CCI for thermal solutions; King Slide and AIC for slide rails; Auras and Chenbro for server chassis; and Accton as the L6 motherboard-level assembly provider.
Regarding the shipping schedule, L6 motherboard-level components have been shipping since May with incremental monthly increases. Orders for L11 rack-level units and components like slide rails are set to enter mass production in July. The overall supply chain is currently in a shipment ramp-up phase, with volume acceleration expected in the third quarter. Following AWS's increased procurement, shipment growth for the involved companies is projected to exceed prior estimates, potentially leading to further positive revisions in their second-half performance.
Strong Demand from Key Clients Like Anthropic and OpenAI
Supply chain sources believe Anthropic is a primary driver for AWS's advanced procurement. Amazon, the parent company of AWS, is a major investor in Anthropic. The two companies reinforced their partnership in April, signing a 10-year collaboration agreement under which Anthropic committed to expanding its purchase of AWS compute power. Anthropic has publicly indicated a significant shortfall in its current computing power planning, which is based on tenfold growth, necessitating rapid scaling.
Beyond Anthropic, OpenAI and Uber are also significant customers for AWS Trainium. Amazon's Bedrock, a generative AI hosting platform for enterprise clients, now boasts 125,000 customers. Its inference workloads are primarily serviced by Trainium, and the expanding demand from the enterprise side provides crucial support.
Amazon CEO Andy Jassy stated in a recent earnings call that Trainium 2 is sold out, Trainium 3 is almost fully booked, some customers have begun reserving the next-generation Trainium 4, and the company has concurrently initiated the development process for Trainium 5.
Intensifying Competition Spurs AWS to Scale ASIC Ambitions
Competitive pressures in the cloud computing market are prompting AWS to accelerate its efforts. While AWS remains the leader in the public cloud market, with Microsoft and Google in close pursuit, it faces direct pressure in the ASIC chip segment. Google's TPU currently holds the top spot in shipments, compelling AWS to rapidly enhance the capabilities and scale of its Trainium offerings.
Furthermore, Meta's recent announcement of plans to open its proprietary AI computing power for external use—allowing businesses and developers to pay for its AI models or directly rent bare-metal GPU compute—publicly pits it against AWS, Microsoft, and Google in the battle for enterprise clients. This further intensifies the competitive landscape and increases the urgency for AWS to expand.
Industry analysis forecasts that in 2026, shipments of high-end AI server chips will see GPU servers grow by approximately 43.8% year-over-year, while ASIC server shipments are expected to grow more significantly, at around 64.2%. With both AWS Trainium and Google TPU servers projected to enter a volume acceleration phase in the second half of 2026, the overall ASIC server market is poised for accelerated expansion.
Comments