Amazon to Sell Trainium AI Chips to External Buyers

The company plans to open its proprietary silicon to commercial sales by late 2027 to expand its footprint in the AI market.

Updated on Oct. 29, 2026 in Artificial Intelligence

Isometric editorial illustration of a stylized computer processor on a circuit board, representing Amazon's shift to selling AI hardware. — Amazon plans to commercialize its proprietary Trainium AI chips by late 2027, seeking to expand its influence in the competitive AI hardware market. AI Illustration. Upload story photo >

Live Poll

Would Amazon selling AI chips directly to competitors improve the market for AI development?

Amazon plans to open its proprietary Trainium AI chip line to external customers by late 2027. This move follows internal AWS demand that has already secured revenue commitments exceeding $225 billion from partners including OpenAI and Anthropic.

Why it matters

By transitioning from an exclusive cloud-only hardware provider to a merchant silicon vendor, Amazon aims to capture a larger share of the AI compute market. This shift creates a new revenue stream and challenges the current dominance of specialized GPU providers.

The upcoming Trainium3 architecture delivers a 30-40% improvement in price-performance compared to the second-generation Trainium2, which itself demonstrated a 30% advantage over existing standard GPUs.

The players

Amazon

A global cloud computing and e-commerce entity that designs custom AI silicon and manages the AWS infrastructure platform.

OpenAI

A research and deployment organization focused on generative AI models that has secured significant compute commitments on Amazon hardware.

Anthropic

An AI safety and research firm that develops large language models and currently utilizes Amazon's proprietary chip ecosystem.

The details

Amazon uses its custom silicon to accelerate machine learning workloads by optimizing tensor processing specifically for training large language models. Developers manage this hardware through TorchNeuron, an open-source tool that maps PyTorch code—a widely used machine learning library—to Trainium-specific instructions. This software stack removes the need for manual kernel optimization, allowing code developed on general-purpose hardware to execute on Amazon's specialized silicon.

Mid-2026: Amazon began direct negotiations with potential external hardware buyers.
October 2026: Amazon released the open-source TorchNeuron software suite.
Late 2027: The scheduled window for commercial availability of Trainium chips.
2028: The year through which current AWS hardware capacity remains primarily committed.

The Tech Race

Amazon currently competes directly with proprietary silicon programs from major cloud providers like Google and Microsoft. Selling Trainium externally marks a tactical shift to compete for data center hardware budgets beyond the AWS ecosystem.

Data center operators and enterprises building large-scale AI applications can expect new hardware procurement options by late 2027. Developers currently using PyTorch will be the primary users, as the platform remains dependent on the TorchNeuron tool for hardware compatibility.

The takeaway

The move signifies an escalation in the competition for AI-optimized silicon that could alleviate hardware supply constraints for major AI labs. Observers should track the 2027 rollout to see how Amazon bridges the performance gap between proprietary cloud stacks and general-market AI hardware.