Amazon to Sell Trainium Outside AWS in Late 2027, Jassy Says

Amazon CEO Andy Jassy said on the company’s 29 October earnings call that customers are encouraging Amazon to make its Trainium AI chip available outside of AWS, and that it intends to do so late in 2027, according to a FirstSquawk newswire headline from the call.

Amazon’s own results releases put its chips business at an annual revenue run rate of over $10 billion at the end of 2025 and over $25 billion two quarters later. By our arithmetic, that is about 2.5 times on the floors Amazon states; the actual multiple may be higher or lower, and the later figures also count Nitro.

Short explainer · Amazon Trainium outside AWS

A vertical news Short narrated from this piece’s headline and opening line.

Business Pill · Vertical Integration

An evergreen Business Engineer explainer on vertical integration. It explains the general idea only and does not discuss Amazon or Trainium.

The key insight: Amazon’s releases put its chips business at an annual run rate of over $10 billion at the end of 2025 and over $25 billion by the second quarter of 2026, and on the 29 October call Jassy said Amazon intends to make Trainium available outside of AWS late in 2027, according to reports of the call.

What Jassy Said on the Call

FirstSquawk’s headline from the call reads: “AMAZON CEO SAYS THAT WITH TRAINIUM CHIP DEMAND SO STRONG, CUSTOMERS ARE ENCOURAGING AMAZON TO MAKE IT AVAILABLE OUTSIDE OF AWS, AND IT INTENDS TO DO SO LATE IN 2027”. An X account that posts call excerpts, @pequityresearch, quoted Jassy as saying: “With Trainium demand so strong, customers continue encouraging us to make it available outside of AWS. We intend to do so late in 2027”.

Amazon had not published a transcript of the call when we wrote this, so these lines come from reports of the call, not from an Amazon document.

Another account, @firesidealpha, quoted Jassy as saying “revenue commitments for Graviton and Trainium have each increased more than 20 times year over year”. A machine transcript of the call published by Gloom carries the same line (see the update below).

Amazon chips business annual revenue run rate as stated in its results releases: over $10 billion (Q4 2025, Tr
Amazon chips business annual revenue run rate as stated in its results releases: over $10 billion (Q4 2025, Trainium and Graviton), over $20 billion (Q1 2026, incl. Nitro), over $25 billion (Q2 2026). The Q3 2026 release gives no figure. Sources: Amazon results releases.

What Was Posted on X

Wall St Engine on X, 29 Oct: Amazon plans to begin selling its Trainium AI chips outside AWS in late 2027, per CEO Andy Jassy.

Shay Boloor on X, 29 Oct: Amazon plans to start selling Trainium chips outside AWS in late 2027, with CEO Andy Jassy expecting “a very significant business selling chips externally”; Boloor notes Jassy says the same chips generate more cash flow inside AWS over 5 to 6 years.

The account @firesidealpha posted a longer excerpt it attributes to Andy Jassy on the chip business: the chips business growing triple-digit percentages year over year, Graviton and Trainium revenue commitments each up more than 20 times year over year, and AI as a tailwind for Graviton because “post-training, reinforcement learning, and agent tool use is mostly done on CPUs versus AI accelerators”.

What Amazon’s Releases Say About the Chips Business

In its fourth-quarter 2025 release, Amazon said “Trainium and Graviton now have a combined annual revenue run rate of over $10 billion and growing at a triple digit percentage year-over-year”.

Its first-quarter 2026 release said it “Exceeded $20 billion annual revenue run rate for Amazon’s chips business—inclusive of Graviton, Trainium, and Nitro—which is growing triple-digit percentages year-over-year”. The second-quarter release said it “Exceeded a $25 billion annual revenue run rate for its chips business, growing triple-digit percentages year-over-year”. The third-quarter release gives no chips run rate.

The definitions differ: the first figure covers Trainium and Graviton, and the later ones add Nitro, so the series is not like for like, and because each figure is a floor, the 2.5 times is approximate.

Who Has Committed to Trainium

Amazon’s first-quarter 2026 release said it “Secured a commitment from OpenAI to consume approximately two gigawatts (GW) of Trainium capacity through AWS infrastructure to power its frontier models and advanced workloads, which begins ramping in 2027”, and that “Anthropic will secure up to five gigawatts (GW) of current and future generations of Amazon’s Trainium chips”. By our arithmetic, the two labs account for up to about 7 gigawatts.

The same release said Amazon “Landed 2.1 million+ AI chips over the past 12 months, more than half of which were Trainium, and announced one million+ NVIDIA GPUs to be deployed starting in 2026”. The third-quarter release adds that TorchNeuron “lets developers run existing PyTorch code (which most already use) on Trainium without rewriting a thing”.

The Google Comparison

Google already sells its TPU systems beyond its own cloud: Google Cloud chief Thomas Kurian said on Alphabet’s 28 October call that TPU systems are offered in customers’ own data centers and through other providers, including Oracle and Crux AI. Our piece on Google selling TPUs to rivals and our piece on Google’s TPU inventory set out how that shows up in Alphabet’s accounts.

Jassy’s late-2027 date is a stated intention reported from the call; Amazon has not given terms, customers or prices for sales outside AWS. Our results piece has Amazon’s quarter, our piece on Jassy’s projection has the capex guide, and our Bedrock piece covers the model layer.

The Structural Read

As we read the releases, Trainium’s two largest named buyers are AI labs: OpenAI’s roughly two gigawatts begins ramping in 2027, and Anthropic will secure up to five gigawatts.

As we read the call reports alongside Alphabet’s 28 October call, Google Cloud chief Thomas Kurian said Google’s TPU systems are offered in customers’ own data centers and through other providers, which is the kind of sale Jassy described for Trainium.

As we read the releases, Amazon reports buying from Nvidia as well as building Trainium: its first-quarter release counted 2.1 million+ AI chips landed in 12 months, more than half of them Trainium, and announced one million+ NVIDIA GPUs to be deployed starting in 2026.

FirstSquawk headline from the Amazon Q3 call, 29 October 2026

“AMAZON CEO SAYS THAT WITH TRAINIUM CHIP DEMAND SO STRONG, CUSTOMERS ARE ENCOURAGING AMAZON TO MAKE IT AVAILABLE OUTSIDE OF AWS, AND IT INTENDS TO DO SO LATE IN 2027”

Three Implications

FOR AI LABS Amazon’s releases name OpenAI and Anthropic as multi-gigawatt Trainium customers; neither release gives prices or contract values.

FOR NVIDIA AND GOOGLE Jassy’s stated plan, as reported, would put Amazon’s in-house AI chip on sale beyond its own cloud, as Google does with its TPU systems.

FOR INVESTORS Amazon has given no terms, customers or prices for selling Trainium outside AWS, and the “more than 20 times” commitments line comes from reports and a machine transcript of the call, not an Amazon document.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Nine Layers of AI.

The analysis puts it this way: “This entire ecosystem has been shaped by constraints, bottlenecks, chokepoints, and scarcity at every layer.”

As we read it, an AI chip sold beyond its owner’s cloud moves a company from renting compute to supplying the chip layer itself. The analysis was published in May 2026 and does not discuss Amazon, Trainium or this quarter.

Update (29 Oct, transcript): A machine transcript of the call published by Gloom carries the “more than 20 times” line: “revenue commitments for Graviton and training have each increased more than 20 times year over year”, where “training” is the transcript’s rendering of Trainium. Asked about selling chips externally versus using them inside AWS, Jassy said, per the same transcript, that the number of companies with a significant cloud business and a chip strong enough to sell externally is effectively two (the transcript renders the word as “too”), and that using chips inside the cloud means “you actually will end up driving more absolute revenue and free cash flow from those chips because you get to use them over you know five to six years”. He added: “we believe we’re going to have a very significant business selling chips externally and we’re going to start in late 2027”. CNBC quoted him separately as saying: “With Trainium demand so strong, customers continue encouraging us to make it available outside of AWS. We intend to do so late in 2027.” Amazon had not published its own transcript when we added this.

What Is Not Established

The call lines come from newswire and X reports, not an Amazon transcript, and the “more than 20 times” line comes from an X post and a machine transcript of the call. Amazon has not said how it would sell Trainium outside AWS, to whom or at what price. The approximate 2.5 times and the 7 gigawatt figure are our arithmetic from Amazon’s releases.

This Week In The AI Supercycle

Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls

This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.

Read the full issue →

Business Engineer Framework

The Nine Layers of AI

A Business Engineer map of the AI stack, layer by layer, and where the constraints sit.

Read the Map of AI →

The Bottom Line

Jassy said on the call that Amazon intends to make Trainium available outside of AWS late in 2027, according to reports of the call, after Amazon’s releases put its chips business at over $25 billion of annual run rate in the second quarter.

95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. Jassy’s call remarks come from a FirstSquawk newswire headline and X posts reporting the 29 October call; Amazon had not published a transcript. Chips figures and Trainium commitments come from Amazon’s quarterly results releases (8-K Exhibit 99.1). The 2.5 times and 7 gigawatt figures are our arithmetic.

Sources: FirstSquawk headline from the Amazon Q3 call (29 Oct 2026) · @pequityresearch call excerpt (29 Oct 2026) · Amazon 8-K Ex.99.1, Q3 2026 results · Amazon 8-K Ex.99.1, Q2 2026 results · Amazon 8-K Ex.99.1, Q1 2026 results

Scroll to Top
FourWeekMBA