Where Will Amazon Be in 5 Years? - The Motley Fool

Semiconductors

Amazon's Custom Chip Run Rate Tops $25 Billion as Trainium Goes Standalone

Amazon's custom chip run rate topped $25 billion, with Graviton and Trainium commitments up twentyfold in Q3. Trainium goes standalone next year.

By
Nathan Brooks
Filed
Channel
Semiconductors
Read
3 min read

Amazon's custom silicon business reached an annual revenue run rate above $25 billion in the second quarter, and revenue commitments for its Graviton and Trainium chips grew twentyfold in the third quarter compared with the year-earlier period. That is not a rounding error in a company of Amazon's size. That is a new business being born in public view.

How Big Is the Chips Operation Now?

The numbers come with unusual precision for a company that, as the reporting notes, does not always break out figures for every part of its operations. The custom chip run rate — more than $25 billion annualized as of Q2 — sits inside AWS, which itself reached a $184 billion annualized revenue run rate in the latest quarter.

AWS remains the No. 1 cloud provider worldwide. It posted that figure within a quarter in which Amazon's total sales hit $201 billion, with retail in the U.S. and internationally accounting for 77% of the total. Retail still pays the lights. Cloud and silicon pay for the future.

The distinction matters for anyone reading Amazon's income statement. E-commerce drives revenue; AWS drives profit. Custom chips now sit at the intersection, growing at triple-digit rates while earlier generations of the same silicon continue to see rising demand.

What Are Graviton and Trainium, and Why Do Customers Care?

Graviton is a central processing unit — the commodity chip found in every computer — designed in-house by Amazon. Trainium is purpose-built for artificial intelligence work. Both compete for attention inside AWS against chips from Nvidia, which customers can also rent.

The pitch is cost. Amazon says customers saved more than $2 billion in the recent quarter simply by using its custom chips instead of alternatives. For a cloud provider whose customers watch their compute bills the way farmers watch rainfall, that figure does more selling than any brochure.

Demand has a structural driver as well. Amazon points to agentic AI — systems that act rather than merely answer — as the next big growth area, and one that will keep chips scarce and orders full.

What Changes With a Stand-Alone Chips Business?

Until now, Graviton and Trainium have lived exclusively inside AWS. Customers have begun asking Amazon to sell them outside that walled garden. Amazon said it expects to do exactly that, with Trainium, later next year.

The move would convert an internal cost advantage into an external revenue line. It also places Amazon in more direct competition with merchant silicon vendors. A stand-alone Trainium business means selling chips to companies that may run them in rival clouds or their own data centers. That is a different commercial posture, and arguably a more interesting one.

Where Does the Five-Year Math Land?

Amazon chief Andy Jassy predicts AWS will become a trillion-dollar revenue business within five to 10 years. On the strength of that forecast and current demand, the reporting projects that Amazon's annual revenue — $716 billion last year — will very likely surpass $1 trillion five years from now.

The path is not free of tolls. The AI build-out is expensive and may continue to weigh on free cash flow, a pattern some investors punish. Amazon's stock has fallen 15% since the start of August, which the account attributes partly to this dynamic.

The counter-argument is monetization. New data centers coming online over the next few years should start generating returns, and the stand-alone chips business adds a fresh revenue driver on top. Amazon cannot pass up the AI opportunity, the analysis concludes, and shareholders stand to reap the rewards of that spending if the current demand holds.

The verdict, for now: a $25 billion run rate, a twentyfold jump in commitments, and a trillion-dollar forecast. Few companies get to bet the balance sheet with those cards.

Source: Google News: AI chips

Share this article:

More from Nathan Brooks

Nathan Brooks

Show full bio

Senior reporter covering people, launches and day-to-day practice in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

713 articles

Related articles

« Previous article