Hyperscaler Custom Silicon: TPU, Trainium, Maia, MTIA — 2026-09-08
Qualcomm has secured a significant role in AWS's AI infrastructure buildout, joining Broadcom as a key design partner for Amazon's Trainium chips. Simultaneously, Anthropic has solidified its massive compute commitments, signing a reported $35 billion deal with Lambda Labs and reaching total compute agreements worth up to $517 billion. Broadcom also raised its fiscal 2027 AI chip revenue forecast to $115 billion, signaling sustained hyperscaler demand for custom silicon over standard GPUs.
Hyperscaler Custom Silicon: TPU, Trainium, Maia, MTIA — 2026-09-08
Top developments
Qualcomm Enters AWS Trainium Design Partnership
On September 8, 2026, it was reported that Amazon Web Services (AWS) has engaged Qualcomm to assist in its AI silicon development efforts, adding a major new name to the custom chip race alongside incumbent Broadcom. This move comes as AWS sits on a $496 billion backlog and seeks to double its power capacity, driving the need for additional design partners to accelerate Trainium deployment. The partnership signals AWS's intent to diversify its supply chain and potentially reduce reliance on a single ASIC designer for its next-generation accelerators.
Anthropic’s Compute Commitments Surge to $517 Billion
Anthropic has reportedly signed compute agreements totaling up to $517 billion within eleven months, securing at least 14.8 gigawatts of capacity through deals with AWS, Fluidstack, Nscale, SpaceX, and Lambda Labs. This includes a newly reported $35 billion deal with Lambda Labs specifically for Nvidia GPU cloud capacity, which will influence Claude API rate limits and model routing. These massive commitments highlight the critical importance of securing both custom silicon (via AWS Trainium) and standard GPU capacity for large-scale AI training and inference.
Broadcom Raises FY2027 AI Revenue Forecast to $115 Billion
Broadcom Inc. announced on September 7, 2026, that it now expects AI chip revenue of approximately $115 billion for the fiscal year ending October 2027, an increase from its prior forecast. This upward revision underscores the continued strong demand from hyperscalers like Google, Meta, and OpenAI for custom ASICs designed by Broadcom. The forecast suggests that despite the rise of alternative designers like Marvell and now Qualcomm, Broadcom remains the dominant force in the custom AI accelerator market, powering key chips such as Google's TPU and Meta's MTIA.
Software Giants Face Scrutiny Over Hardware Experience
A report from Seoul Economic Daily on September 7, 2026, highlighted growing skepticism regarding the ability of software-centric giants like Microsoft, Meta, OpenAI, and Anthropic to successfully manage their own custom chip projects. While these companies are racing to build proprietary silicon to reduce costs and dependency on Nvidia, analysts note their lack of deep hardware engineering experience poses significant risks. This commentary adds context to the recent partnerships with specialized designers like Broadcom, Marvell, and now Qualcomm, who provide the necessary hardware expertise.
Local view
South Korean media outlet Seoul Economic Daily (en.sedaily.com) has published analysis questioning the strategic viability of software-focused companies attempting to master complex hardware design. The article notes that while the move toward custom silicon is driven by cost efficiency and supply chain control, the technical barriers remain high for firms without a legacy in semiconductor engineering. This perspective contrasts with the bullish outlooks from US-based financial outlets focusing on Broadcom's revenue growth.
Context & numbers
- AWS Backlog: Amazon Web Services currently holds a $496 billion backlog, driving aggressive expansion in AI infrastructure capacity.
- Broadcom Revenue Forecast: Projected AI chip revenue of ~$115 billion for FY2027, up from previous estimates.
- Anthropic Compute Spend: Total committed compute agreements reported at up to $517 billion, including $35 billion with Lambda Labs.
- Market Share: Broadcom maintains a dominant position in the custom AI accelerator design services market, estimated at over 70% share as of earlier 2026 reports, though competition is intensifying with Marvell and now Qualcomm.
On the radar
- Qualcomm's Delivery Timeline: Analysts are watching closely to see if Qualcomm can deliver on AWS's requirements quickly enough to impact the Trainium roadmap, given the tight window for next-generation deployments.
- Anthropic's IPO Path: The scale of Anthropic's $517 billion in compute commitments may influence its future public offering strategy or capital structure, as noted by financial observers tracking its "cautious compute stance" reversal.
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
