Anthropic's Trainium deal challenges Nvidia's dominance
J.P. Morgan's Michael Cembalest identifies Anthropic's use of Amazon's Trainium chips as a major threat to Nvidia, with hyperscalers seeing 30-40% cost savings. Nvidia's accelerator revenue share may drop from 85% to 75%. Despite recent software vulnerabilities, the shift to custom silicon continues.
*this image is generated using AI for illustrative purposes only.
A structural shift in artificial intelligence (AI) infrastructure is challenging Nvidia Corp.’s market dominance, according to J.P. Morgan Asset Management’s Michael Cembalest. He highlights Anthropic’s decade-long pledge to deploy its models on Amazon.com Inc.’s custom Trainium processors as the "strongest third-party" endorsement of specialized application-specific integrated circuits (ASICs) to date. This development follows Anthropic’s commitment to spending more than $100 billion on Amazon Web Services (AWS) technology to secure massive data center capacities.
The Silicon Shift
The migration toward custom silicon underscores a growing trend among hyperscalers developing in-house chips. J.P. Morgan Equity Research indicates these companies are reporting total cost of ownership reductions of 30% to 40% compared to traditional merchant GPU fleets. Amazon maintains that its custom AI silicon offers high performance at a significantly lower cost, providing a direct competitive alternative to high-priced hardware infrastructure.
Market Share Projections
Cembalest notes that this transition contributes to a gradually "declining share" of global accelerator revenue for Nvidia. The company's market share is projected to slip from 85% in 2023 down to an estimated 75%. This shift reflects the increasing viability of ASICs as a substitute for general-purpose GPUs in AI workloads.
Volatility and Long-Term Scale
The hardware expansion coincides with operational challenges. Access to Anthropic’s advanced "Mythos-class" reasoning models, commercially known as Fable 5, recently faced an abrupt "global shutdown" after Amazon’s software testing exposed critical vulnerabilities that could act as a catalyst for "cyberattacks." Despite these export blocks, the underlying custom silicon framework remains intact. Cembalest concludes that while tech hardware dependencies are permanently shifting, the broader financial projections of these premium AI labs remain "speculative, uncertain and subject to revision."
Amazon Stock Performance
AMZN shares have risen by 1.43% year-to-date, declined by 12.09% over the last month, and advanced 12.30% over the year. The stock closed 0.57% higher at $234.11 apiece on Tuesday and was lower by 0.37% in premarket trading on Wednesday. Benzinga’s Edge Stock Rankings indicate that AMZN maintains a weak price trend in the short, long, and medium terms, with a moderate quality score.
| Metric | Performance |
|---|---|
| Year-to-date | +1.43% |
| Last month | -12.09% |
| Over the year | +12.30% |
| Tuesday close | $234.11 (+0.57%) |
| Wednesday premarket | -0.37% |
Will other hyperscalers follow Amazon's lead in aggressively developing custom ASICs to reduce their dependency on Nvidia?
How will Nvidia respond strategically to the projected decline in accelerator market share from 85% to 75%?
Can the cost efficiency of custom silicon sustain the projected growth of AI labs despite the current speculative financial outlook?
19
Create Custom Screeners 
Access 200+ Pre-built Filters 
Use 50+ Readymade Screener 
