Hyperscaler Custom Silicon: TPU, Trainium, Maia, MTIA — 2026-09-11
Amazon has expanded its AI infrastructure strategy by signing a major multi-generational deal with Qualcomm for custom inference silicon, marking a significant shift beyond Broadcom and Marvell. Simultaneously, OpenAI has diversified its manufacturing risk by adding Samsung Foundry to its chip production program, while Anthropic continues its massive compute expansion with new agreements totaling hundreds of billions. These moves highlight the hyperscalers' aggressive push to secure supply chains and reduce reliance on Nvidia's dominance.
Hyperscaler Custom Silicon: TPU, Trainium, Maia, MTIA — 2026-09-11
Top developments
Amazon and Qualcomm Partner on Custom Inference Chips
On September 8, 2026, Amazon Web Services (AWS) announced a multi-generational partnership with Qualcomm to develop custom AI inference chips and 1.6T optical connectivity for AWS infrastructure. The deal includes a warrant allowing AWS to acquire up to 25 million Qualcomm shares, contingent on purchases reaching up to $60 billion. This move diversifies AWS’s silicon supply chain beyond Broadcom, which has traditionally designed its Trainium chips, and addresses the growing demand for efficient inference workloads as the company doubles its power capacity.
OpenAI Adds Samsung Foundry to Silicon Program
OpenAI is expanding its custom silicon strategy by partnering with Samsung Foundry for the production of next-generation AI accelerators, according to confirmation from OpenAI Korea GM Harrison Kim on September 10, 2026. This relationship operates through Broadcom’s existing $200 billion manufacturing MOU, providing OpenAI with a critical second foundry option alongside TSMC. This diversification mitigates single-source dependency risks for the company’s "Titan" program and other custom ASIC efforts, ensuring greater supply chain resilience in a constrained market.
Anthropic’s Compute Deals Reach $517 Billion
Anthropic has reportedly signed compute agreements worth up to $517 billion, securing at least 14.8 gigawatts of capacity within eleven months. These deals span cloud services, custom chips, and long-term data center leases, reflecting a significant reversal in the company’s previously cautious stance on compute spending. The agreements include commitments to use Google TPUs and AWS Trainium, reinforcing the strategic importance of hyperscaler custom silicon for leading AI labs.
Local view
In Japan, media outlets are closely following the Amazon-Qualcomm partnership, with Zaikei Shimbun reporting on the joint development of AI inference semiconductors for AWS and the specific financial structures, such as the purchase-linked warrants for up to 25 million shares. The coverage highlights how this deal positions Qualcomm as a new key player in the AI silicon race alongside established partners like Broadcom. Meanwhile, Forbes Japan discusses the broader implication that Amazon may not be able to win the AI competition relying solely on AWS infrastructure, suggesting a need for more diverse hardware strategies.
Context & numbers
- Market Share: Broadcom holds approximately 70%+ of the custom AI accelerator design services market, while Marvell commands 20–25%.
- Revenue Projections: Marvell projects up to $11 billion in AI ASIC revenue for 2026, driven by partnerships with Amazon (Trainium) and Microsoft (Maia).
- AWS Backlog: Amazon is sitting on a $496 billion backlog, which is driving its need to expand and diversify its AI chip suppliers.
On the radar
- Qualcomm Delivery Timeline: Industry analysts are watching whether Qualcomm can deliver the custom inference silicon before the current window for capacity expansion closes, given the tight timelines for 2027 deployments.
- Samsung Yield Rates: The success of OpenAI’s move to Samsung Foundry will depend on Samsung’s ability to meet yield requirements for advanced nodes compared to TSMC, a critical factor for future volume production.
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
