Amazon Accelerates Custom AI Chip Production to Meet AWS Demand
Amazon Web Services reported a 39.3% revenue increase to $46 billion in the third quarter of 2026, driven by surging enterprise demand for artificial intelligence infrastructure. To sustain this growth and manage supply chain constraints, Amazon is advancing its proprietary silicon roadmap and preparing to sell its Trainium AI accelerators to external customers by late 2027.
How Amazon’s Graviton and Trainium Chips Lower AI Costs
Amazon’s custom-chip business has surpassed a $25 billion annual revenue run rate, fueled by enterprise adoption of its Graviton central processing units and Trainium AI accelerators.
According to MarketWise, these in-house chips offer meaningful cost savings compared to Nvidia hardware, making them highly attractive for businesses running large-scale AI workloads. Major customers, including OpenAI, Anthropic, Meta Platforms, and Uber, are actively securing AWS custom-chip capacity.
Because advanced chip designers rely on the same manufacturer and receive fixed allocations, newer Trainium capacity is nearly sold out, and future generations are being booked well ahead of launch. This high demand underscores why Amazon is prioritizing its custom silicon to maintain competitive cloud pricing.
Why Amazon Will Sell Trainium Chips Outside AWS
CEO Andy Jassy announced on the Q3 earnings call that Amazon will start selling its Trainium AI chips outside Amazon Web Services in late 2027.
Selling externally allows Amazon to book revenue immediately when the chip leaves inventory, rather than realizing the value gradually through AWS cloud consumption. Jassy highlighted the rarity of this capability in the current market, noting that Google already offers its TPU chips outside its own facilities.
Jassy stated that "the number of companies who have a significant cloud business and a chip strong enough to sell externally is effectively 2". This move transitions Amazon from a pure cloud services provider into a direct silicon supplier for enterprises that want to build their own AI infrastructure.
What Amazon’s $220 Billion Capital Expenditure Means for Cloud Buyers
To support this expanding silicon and cloud footprint, Amazon’s Q3 capital expenditures reached $75.9 billion, with full-year 2026 spending expected to total approximately $220 billion. Management expects to spend meaningfully more in 2027 to meet ongoing AI infrastructure demand.
Amazon’s capital investment outpaces its primary hyperscaler competitors, as shown in their 2026 forecasts:
| Company | 2026 Capital Expenditure Forecast |
|---|---|
| Amazon | $220 billion |
| Alphabet | $195 billion to $205 billion |
| Microsoft | $170 billion |
For IT decision-makers, this massive capital injection signals that AWS is aggressively expanding its AI compute capacity. The company forecasts AWS growth above 40% in both 2027 and 2028, supported by a $578 billion backlog of customer commitments.
How IT Leaders Should Adjust Their AI Compute Strategy
With Trainium capacity nearly sold out and external sales launching in late 2027, businesses relying on AWS for AI workloads must secure their compute allocations early to avoid capacity bottlenecks.
IT leaders should evaluate their current AI infrastructure costs. If they are currently using third-party GPUs via AWS, migrating applicable workloads to Graviton or Trainium instances could yield significant cost reductions.
Furthermore, the impending availability of Trainium chips outside AWS means enterprises might soon have the option to purchase or lease this silicon directly. This will allow organizations to build on-premises or hybrid AI deployments, bypassing traditional cloud consumption models entirely and giving IT teams greater control over their long-term AI hardware strategy.
