Amazon is reportedly in talks to sell its in-house AI accelerator, AWS Trainium, directly to other companies for use in their data centers. If realized, the move would add a serious alternative to Nvidia, which dominates the AI chip market.
April 9, 2026 · Amazon / AWS
Amazon May Sell Its Trainium AI Chips Beyond the Cloud
CEO Andy Jassy says demand for Amazon's in-house silicon is very high — and the company is weighing direct, by-the-rack sales to outside firms, offering the market a fresh alternative to Nvidia. As of June 2026, it remains at the deliberation stage.
$20B+
Current annual run rate of the chip business
$50B
Potential scale as a standalone external business
~2 yrs
Likely window to begin external sales (no deal signed yet)
Business run rate vs. external potential
Same unit (USD, annual). Column height is proportional to value.
$20B+
Today
The Trainium roadmap
Trainium2
~30% price-performance edge over comparable GPUs. Sold out.
Trainium3
A further 30-40% gain over Trainium2. Rolling out from early 2026.
Trainium4
Full availability in ~18 months. Heavily pre-booked.
Reported cost savings vs. Nvidia GPUs
Customer-reported reductions on AWS instances.
Overall, price-performance gains of ~40-50% vs. Nvidia GPUs have been reported. Adopters via AWS include Anthropic, OpenAI, Databricks and others.
Why it appeals
40-50% cost savings vs. Nvidia GPUs
Keep existing PyTorch / TensorFlow workflows via Neuron SDK
Avoids the so-called "Nvidia tax"
The constraints
Trainium2/3 supply is tight — immediate use is hard
Software ecosystem less mature than Nvidia's
GPUs still chosen for cutting-edge, high-performance work
External hardware sales have not yet materialized — there is no signed deal and no direct buyer assessment. If Amazon proceeds, it would add a new option to an AI-chip market still led by Nvidia.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…