Latest Search
Quote
| Back Zoom + Zoom - | ||||
|
<Research> CICC: Amazon.com, Inc. (AMZN.US) Trainium Chip Series Continues to Win Customers, Growth Outlook Positive
Recommend 5 Positive 3 Negative 1 |
||||
|
||||
|
CICC published a report stating that on May 19, The Information reported that Amazon.com, Inc. (AMZN.US)'s Trainium has begun attracting smaller-scale AI developers, following earlier announcements by Anthropic and OpenAI on deploying Trainium. The broker believes that Amazon's Trainium chip series continues to gain customers, with promising growth prospects. In February, OpenAI announced that it had secured 2GW of Trainium computing capacity through AWS, covering generations including Trainium3 and the next-generation Trainium4. In April, Anthropic committed to spending more than USD100 billion on AWS over the next decade and announced that it had locked in 5GW of Trainium computing capacity to train and deploy Claude. The products include Graviton and Trainium 2 to 4, including Trainium2 launched in the first half of the year, as well as 1GW of Trainium2 and Trainium3 computing capacity to be launched before the end of 2026. The Information reported that software improvements have enabled smaller AI developers to migrate from GPUs to Trainium. Amazon stated last month that Trainium2 is largely sold out, Trainium3 is nearly fully booked, and most of the Trainium4 capacity, which is scheduled for mass production in about 18 months, has already been reserved. The broker believes that demand visibility and growth certainty for Trainium are high, potentially driving earnings growth for related suppliers. The report noted that the market previously held relatively low expectations for Trainium and considered it undervalued. From earlier concerns over performance issues to its recent customer wins, key factors include tight supply of NVIDIA GPUs, with Trainium deployment offering greater flexibility in computing resources; improved software support, with years of deep collaboration between Anthropic and Amazon enhancing the chips supporting software capabilities, facilitating deployment by other customers, who indicated that documentation and support have improved in recent months alongside better compatibility with open-source tools; lower costs compared with GPUs, enhancing cost performance, with Trainium customers indicating that for inference workloads it can save about 35% in costs compared with H100, while Amazon's CEO stated that Trainium3 may deliver 30% to 40% better cost performance than Trainium2; significant performance gains from generational upgrades, with Trainium3 UltraServer achieving 4.4x performance improvement and 3.9x memory bandwidth increase over the previous generation, and Trainium4 potentially delivering a threefold increase in FP8 performance; and Trainiums proprietary switching architecture supporting All-to-All communication, offering superior performance and total cost of ownership in MoE models. (ha/da) Auto-translated by AI This article was automatically translated by AI, the original language version should be considered the authoritative version. AASTOCKS.com Limited does not guarantee its accuracy or completeness and accepts no liability for any damages or losses arising from the use of this translation. More Details
AASTOCKS Financial News |
||||
Copied to Clipboard
 |
||||
