Urgent Boost for Self-Developed AI Chips: Amazon AWS Raises Q3 ASIC Server Shipment Targets by Up to 30%.

TradingKey - Amazon ( AMZN) AWS is accelerating its self-developed AI chip strategy. According to supply chain sources, AWS has notified relevant suppliers to increase its Q3 2026 ASIC server shipment volume by 20% to 30% compared to the original plan. This move is interpreted as a sign of AWS's strong optimism regarding the market prospects of its latest Trainium 3 chip, as well as an important signal of its intent to preemptively capture ASIC market share.

As AWS's first AI chip to utilize a 3nm process, Trainium 3 represents a comprehensive leap in performance over the previous generation, Trainium 2. Its computing performance reaches 2.52 petaflops, which is twice that of Trainium 2. Its memory capacity has increased by 1.5 times, memory bandwidth has improved by 1.7 times, and energy efficiency has surged by over 4 times. Supporting trillion-parameter models and multimodal tasks, the chip can scale to clusters of up to 144 chips, representing AWS's latest technological strength in the AI computing power field.

According to DigiTimes, citing supply chain sources, server motherboard-level components equipped with the Trainium 3 chip began shipping in May of this year and are currently in a month-over-month volume ramp-up phase.

With this upward revision of the shipment plan, the shipment growth of relevant supply chain manufacturers will be higher than previously expected, leaving room for further upward revisions in their second-half operating performance.

The Taiwanese supply chain for AWS Trainium 3 servers has formed a complete ecosystem. This includes AVC, Micro-Star (Maik), and Zen Voce in the cooling sector; King Slide and Nan Juen International for slide rails; AVC and Chenbro for chassis; and Accton as the L6 motherboard-level assembly plant. Orders for L11 cabinet-level assembly and components like slide rails will enter mass production in July, with volume growth expected to surge in the third quarter.

Strong Demand from Major Customers, Anthropic Becomes Main Driver

Supply chain sources pointed out that AI startup Anthropic is one of the key drivers behind AWS's accelerated inventory pull this time. As a major investor in Anthropic, Amazon signed a 10-year cooperative agreement with the company in April this year, under which Anthropic committed to expanding its procurement of AWS computing power. Anthropic has publicly stated that there is still a significant gap in its computing power planning, which is currently based on a 10x growth benchmark, and this gap needs to be filled as soon as possible.

In addition to Anthropic, OpenAI and Uber are also major customers of AWS Trainium. Meanwhile, Bedrock, the generative AI managed platform launched by Amazon for enterprise customers, now has 125,000 clients. Its inference workloads are primarily serviced by Trainium, with continuously expanding enterprise-side demand serving as a crucial anchor of support.

Amazon CEO Andy Jassy revealed during a recent earnings call that Trainium 2 is sold out, Trainium 3 is almost fully booked, and some customers have already begun reserving the next-generation Trainium 4, while the company has simultaneously initiated the development process for Trainium 5. This series of moves indicates that market demand for AWS's self-developed AI chips far exceeds expectations and is currently in a phase of rapid expansion.

As the competitive landscape tightens, AWS urgently needs to strengthen its market position in ASICs.

Intense competition in the cloud computing market is prompting AWS to accelerate its self-developed chip layout. Although AWS remains the leader in the public cloud market, Microsoft and Google are closely following behind, with competition in the AI computing power field becoming increasingly white-hot.

In the ASIC chip field, Google's TPU shipments rank first in the market, exerting direct pressure on AWS and forcing it to accelerate the enhancement of Trainium's performance and scale.

In addition, Meta recently announced plans to open up its own AI computing power to the public, allowing enterprises and developers to pay to use its AI models or directly rent bare-metal GPU computing power, openly competing with AWS, Microsoft, and Google for enterprise customers. This has further intensified market competition and heightened AWS's sense of urgency to expand.

DIGITIMES Research analyzes and forecasts that in the high-end AI server chip market in 2026, the annual growth rate of GPU server shipments will be about 43.8%, while the annual growth rate of ASIC server shipments will be even more significant, expected to reach 64.2%. As both AWS Trainium and Google TPU servers enter the volume production phase in the second half of 2026, the overall ASIC server market is expected to experience accelerated expansion.

AWS's upward revision of Trainium 3 shipments this time is not only a response to market demand, but also an important measure to consolidate its competitive advantage in the AI computing power field.