Amazon Says It Can Embiggen AWS Past “Multi-$100 Billion” With AI

Here is how you know the cloud revolution is not done: Throughout the 77 quarter financial history of Amazon Web Services, which was formally launched in March 2006, there have been so few quarters of sequential revenue decline that you can literally count them on one hand. And the last quarter it happened was in Q1 2023, and the sequential decline in revenue was a mere one-tenth of a percent an it doesn’t quite count. The other four times were real, and they happened many years ago, in a different era many years before the coronavirus pandemic and before the AI revolution truly got underway six years earlier than that.

This is an amazing thing, and a testament to the ingenuity, tenacity, and drive of AWS. Sequential growth like this is a lot harder to achieve than year on year growth each quarter, which incidentally AWS has been also able to do for every single quarter it has existed. The former implies the latter, but it is important to say it out loud.

So who the hell cares if AWS cloud growth did not make the expectations of the models Wall Street puts together for three quarters in a row? Maybe Wall Street has a problem, and AWS doesn’t. Microsoft did well with Azure in its most recent quarter because OpenAI had contractual arrangements to spend big bucks and its non-AI cloud business saw a bump.

Anyway, that impressive growth streak by AWS we outlined above does not mean that this level of growth, or even any growth at all, goes on forever. And here at the beginning of 2025 and at the leading edge of the commercialization of GenAI, it is reasonable to wonder if AWS will be able to get a second wave of high growth similar to what it has enjoyed a decade ago or if AI will just allow a more tepid level of growth. It is also reasonable to wonder if the use of more general cloud capacity will accelerate as companies rent rather than buy AI models and processing capacity.

On a conference call going over the first quarter of 2025 financial results for AWS and its parent company, Amazon chief executive officer Andy Jassy had this to say:

“It is useful to remember that more than 85 percent of the global IT spending is still on premises – so not in the cloud yet. It seems pretty straight-forward to me that this equation will flip in the next ten to twenty years. Before this generation of AI, we thought AWS had the chance to ultimately be a multi-$100 billion revenue run rate business. We now think it could be even larger. If you believe your mission is to make customers’ lives easier and better every day, and you believe that every customer experience will be reinvented with AI, you’re going to invest very aggressively in AI and that’s what we are doing.”

What Jassy does not say, and that we all know, is that for every $1 spent on IT hardware in the world, there is more than $3 spent on enterprise software and another $4 spent on IT services to help implement, manage, and maintain all of this stuff. While AI will more than double the amount of spending on datacenter systems in the past four years, which can help AWS grow, for AWS to grow more it will have to become more of a supplier of software and services. In the age of agentic AI applications replacing hand-coded applications written in Java, PHP, Python, or Node.js/JavaScript against databases and datastores, AWS can become such a software player.

We already think that a big chunk of revenues at AWS can be attributed to the software the comprises its massively distributed, global-scale data processing system, as the chart above shows. Software, we think, kissed 50 percent of AWS revenues this time last year, but the rise of AI hardware has compute taking up a larger and larger share of revenues once again. As AWS moves to its own Trainium and Inferentia AI compute engines, this revenue could shrink even as usage goes up because AWS is committed to provided a 30 percent to 40 percent lower cost per token generated compared to using Nvidia GPUs. This is the same spread that AWS Graviton Arm server chips offer to customers compared to X86 instances on the EC2 compute cloud.

The question is will AWS use AI to help customers eliminate a slew of software and services costs it currently has, and therefore cause a slowing in overall IT spending due to increased efficiencies but a much larger share of wallet spend to come to AWS? We think this is a highly likely strategy. And that would imply that AWS will not just dance around being a major software and services player, thereby competing against its thousands of major partners worldwide.

What other choice does it have, really?

In the meantime, let’s focus on the quarter that just ended. In the March quarter, Amazon posted sales of $155.67 billion, up 8.6 percent, with operating income of $18.41 billion, up 20.2 percent. That operating income for the whole company, which is being driven by Amazon’s media and advertising business and not just by AWS as in years gone by, takes some of the heat off AWS and thus allows it to think about its own investments rather than trying to carry the rest of the company. (The retail business is probably not all that profitable is out guess.) Amazon had a net income of $17.13 billion in the quarter, up 64.2 percent, and it grew its cash hoard by 28 percent to $92.57 billion.

The Amazon conglomerate has lots of maneuvering room, unlike a decade ago and unlike its situation before the pandemic even. So you can understand why founder Jeff Bezos feels like he can leave the company in the hands of Jassy and focus on other things like Blue Origin and the Washington Post.

During the March quarter, the AWS cloud business brought in $29.27 billion, up 16.9 percent and up 1.7 percent sequentially. This was fairly weak sequential growth by AWS standards and a little better than average for the year-on-year quarterly growth rates the company has experienced in 2023 and 2024. AWS had an operating income of $11.55 billion, up a very respectable 22.6 percent and representing 39.5 percent of revenues. This is the highest level of operating profitability that AWS has ever had in its history, and this is the third quarter in a row that this sentence has been true.

Clearly, Jassy is driving efficiencies at AWS.

According to Brian Olsavsky, AWS had a backlog of $189 billion, up 20 percent year on year, as the March quarter came to a close. This is an IBM-class services backlog, and it is for utility computing not consulting engagements that are people intensive as the old Global Services at Big Blue (a lot of which has been spun off as Kyndryl) was.

In the quarter, Amazon spent $25.1 billion in capital expenses, which was 67.1 percent higher than the year ago period.

Our best guess is that IT infrastructure represented about 85 percent of this spending, or $21.3 billion. Of this, we think about 90 percent of that IT infrastructure spending – or 19.2 billion – was for AI systems being installed for AWS to rent out.

Back in early 2020, when the coronavirus pandemic was just starting to hit, we estimate that around 30 percent of capital expenses spending was spent for AI systems and another 30 percent was spent on non-AI IT systems for AWS. The other 40 percent was for fulfillment centers and transportation fleets relating to the Amazon retail business. In the current quarter ended in March 2025, we think that AI systems for AWS comprised 76.5 percent of all CapEx for Amazon, and another 8.5 percent was for more general purpose IT systems for AWS. AI spending is now 7.4X higher than general purpose IT spending at AWS, and we wonder if this is a leading indicator for the industry at large or if this is just a temporary condition because AI is so new.

As 2025 comes to a close, the expectation is that AI system revenues will represent about half of all datacenter systems spending. The spending levels at AWS are 7.4 times higher than a 1:1 ratio, we think. It’s anybody’s guess if this represents the new normal. We just point out that this lopsided ratio seems likely at AWS right now.