Cloud Quarterly: Azure AI pop, AWS supply pinch and Google execution

Artificial intelligence is currently running on a one-lane CapEx highway that’s capacity-constrained. Hyperscaler cloud providers are the biggest contributor to the momentum and as such, investors keep rating with cloud growth rates.

But this past quarter you needed to read the footnotes and the wonderful print to transcend the huge investments fueling the wave. Microsoft Corp. put up eye-popping Azure growth again, but a giant a part of that acceleration is Azure serving AI inference — notably OpenAI’s ChatGPT — now neatly included within the revised definition of what’s included in Azure. That is great for headlines but not conducive to apples-to-apples comparisons over time.

Meanwhile, Amazon Web Services Inc. delivered the biggest revenue for cloud infrastructure and sent a transparent message that demand exceeds supply — meaning growth is capped by power and components, not pipeline. That creates a weird optics penalty — AWS showing growth within the high teens on a $120 billion-plus run rate and it’s deemed “concerning.” But it surely also telegraphs future upside as capability goes online and depreciation cycles progress.

The somewhat stealth story is Google LLC. Google Cloud posted a robust quarterly print with solid top-line growth and steadily improving operating margin — and GCP (the infrastructure-as-a-service and platform-as-a-service portion of the business) is growing materially faster than cloud overall – our estimate is sort of 40%. Google’s backlog is constructing at greater than $250 million and $1 billion-plus deals proceed to shut due to Google’s underlying tech, a robust data foundation and its AI prowess. As the combination shifts toward AI-heavy, infrastructure-centric workloads, it becomes a tailwind for Google, which continues to lag the size of AWS and Azure.

The common thread is a capital expenditure arms race that faces real constraints. All three players said that power, sites, servers and lead times will dictate who wins AI inference and who can monetize it. Microsoft is capacity-constrained, AWS says it is going to take “several quarters” to rebalance, and Google raised its CapEx guide. As such, we consider strongly that this will not be a one-quarter story; it’s a multiyear power grab that may determine margin structures and money flow for the following cycle. All told, the the massive three cloud players are on a pace to spend about $240 billion this calendar yr on CapEx with, AI revenue coming in at about 10% of that figure. We clearly have a giant hurdle before that massive investment pays back.

On this Breaking Evaluation, we update you on our latest take of the cloud infrastructure market with a fresh take a look at growth rates in context, market share projections for the yr, and the puts and takes for every of the massive three from their earnings prints. We’ll also share some spending data from Enterprise Technology Research and a take a look at the road ahead.

Q2 at a look

The massive three U.S. hyperscalers just put up roughly $60 bilion of IaaS/PaaS revenue, and all-in IaaS/PaaS grew about 27% year-over-year. Importantly, that’s a one-point acceleration sequentially — all while demand continues to be outrunning supply.

Margins proceed to be divergent. AWS got here in around 33% operating margin, down from 39% last quarter as depreciation and its seasonal stock based comp step-up hit — all with continued capability constraints. Google Cloud, while far less profitable continues to indicate material improvement. Operating margin increased to roughly 21%, up from ~18% last quarter and 11% a yr ago — showing, regular execution. Microsoft is a little more opaque on its cloud operating margin as Azure drags margins down, but the corporate gave us some solid hints and a first-time Azure revenue disclosure. Microsoft Cloud revenue was $46.7 billion, up 27%, with gross margin 68%, down 2 points aI infrastructure scales and the combination shifts toward lower margin Azure. But for the primary time, Microsoft put a number on Azure at $75 billion for the fiscal yr — but remember, recent definition changes and the AI inference boost muddy apples-to-apples comparisons and contain a clever communications strategy by Microsoft.

CapEx is the true story, nevertheless. The trio — plus Meta Platforms Inc. — shelled out about $87 billion within the June quarter. By our math, ~75% of Amazon’s CapEx is for AWS, underscoring the provision buildup. The underside line is Azure’s AI pop is real (but fattened by ChatGPT inference), AWS is constrained but investing through it, and Google is quietly widening margins while it scales.

Let’s take a take a look at the important thing performance indicators and do a fast scan of the metrics that matter:

Microsoft: Microsoft Cloud got here in at $46.7 billion for the quarter (+27% y/y) and Intelligent Cloud $29.9 billion (+26%) with Azure and other cloud +39% y/y (CC); Microsoft didn’t disclose its quarterly Azure performance but we’ve mapped our models to the $75 billion figure and estimate Azure IaaS/PaaS at ~$22.2 billion; with Azure AI services at ~$3 billion. Management guided that demand exceeds supply into 1H FY26 (2H CY 25), with its Q1 FY CapEx guided to $30 billion-plus.

AWS: AWS revenue was $30.9 billion (+17.5% y/y; $124B ARR) with 32.9% operating margin pressured by seasonal SBC and better depreciation; we peg AI services at around $2B for AWS. Backlog hit $195 bilion and CapEx stayed heavy ($24 biion of Amazon’s $31.4 billion in Q2 went to AWS by our estimates). Supply constraints proceed but are expected to enhance over several quarters.

Google Cloud: Google Cloud posted $13.6 billion (+32% y/y) and 20.7% operating margin; we estimate GCP (IaaS/PaaS) at ~$7.2B, growing “much faster” than Google Cloud overall, with GCP AI services around ~$0.6B. Capex was $22.4B in Q2 and the 2025 outlook was raised to ~$85 billion (higher again in 2026) with backlog at $106 billion and capability tight into 2026.

Positives and negatives for the massive three

Now lets take a look at the plusses and minuses for every player.

Microsoft quick read

Pluses: Azure continues to be ripping — ~39% growth with management guiding to ~37% cc next quarter even while capacity-constrained. We predict it is going to be even higher – perhaps as high as 40% with GPT-5. A $368 billion RPO (+37% y/y) says the the corporate has a protracted runway, and foundry/agent adoption combined with surging token throughput says Microsoft has found a real AI platform flywheel. But it surely appears heavily tied to OpenAI, which is broadening it ecosystem horizontally with latest open source models. Software efficiencies are kicking in with (“~90% more tokens per graphics processing unit” yr on yr) – without this, the gross-margin drag from scaling AI infrastructure could be much worse.

The opposite welcome positive is the $75 billion Azure disclosure.

Minuses: Microsoft still won’t disclose an Azure operating margin and is cagey around cloud operating margin. Overall Microsoft Cloud gross margin fell ~2 points year-over-year, pressured by Azure’s AI mix. Supply stays tight through 1H FY26, implying some demand is being pushed, and a big piece of Azure’s momentum appears tied to ChatGPT inference and to reporting definitional changes, which confuse the expansion rate relative to historical reporting.

AWS quick read

Pluses: Still the revenue leader and despite the drop in operating margin from the high 30s to ~32.9%, AWS stays profitable for what is basically an organization that sells access to hardware resources. Demand exceeds supply; backlog sits at ~$195 billion (+25% y/y) and the AI stack runs deep and wide — from Trainium2 to Bedrock plus AgentCore/Kiro announced at NYC Summit and an enormous ecosystem that offers AWS customers selection and the corporate  full-stack leverage.

Minuses: Headline growth of ~17.5% trails Azure/GCP and invites lazy comparisons. Let’s face it – AWS is getting outmarketed immediately by Microsoft in relation to cloud momentum. Heavy CapEx means elevated depreciation and, with capability expected to rebalance over “several quarters,” near-term upside could also be capped.

Google quick read

Pluses: Google Cloud is hitting a pleasant combo of scale and improving profitability — ~32% growth with ~20.7% operating margin — and GCP (IaaS/PaaS + AI) is growing much faster than overall cloud (tCR est. ~39%). Backlog momentum is real at ~$106 billion, with $250 million-plus deals doubling and $1 billion-plus wins in 1H25 matching all of 2024.

Minuses: Capability stays tight into 2026, which could pace revenue recognition, and Alphabet flagged an acceleration in depreciation as CapEx grows — each potential margin headwinds. Disclosure on pure IaaS/PaaS stays less granular, so we’re counting on qualitative signals and survey data for GCP’s exact mix shift.

The scoreboard on IaaS/PaaS revenue and market share

Let’s zoom out and try the annual revenue and market share forecasts:

Cloud infrastructure is now a quarter-trillion-dollar market. In 2025, the Big 3+ Alibaba will clear about $259 billion in IaaS/PaaS revenue, up roughly 26% yr over yr. AWS stays the revenue leader at ~$126.6 billion, but its growth cadence — ~18% — is well below Microsoft Azure and Google Cloud Platform, each running high-30s. Azure is the massive story: ~$87.7 billion in 2025E IaaS/PaaS, with growth juiced by AI inference volumes and Microsoft’s shifts in what it counts as ‘Azure.’ GCP is a solid No. 3 at ~$28.3 billion, growing about as fast as Azure from a smaller base.

Share shifts proceed. AWS drops about 4 points of IaaS/PaaS share since 2024 to ~48.6%. Microsoft picks up ~4 points to ~35.3%, while Google holds regular around ~10%. Put in another way: Azure + GCP are growing at a combined ~38% versus AWS at ~18% — that gap continues to draw negative attention to AWS.

Underneath the revenue lines is an unprecedented CapEx cycle. The massive three are on pace to pour roughly $240 billion a yr into data centers, power and servers to feed AI demand. Yet 2025 AI services revenue is just on the order of ~$25 billion from these three, which suggests heavy front-loading — that’s, a number of spend now for monetization that (hopefully) scales later. The underside line is AWS still throws off essentially the most dollars, but Microsoft’s definition-plus-AI strategy is gaining share fast, and Google is quietly compounding with evidence of improved profitability and a protracted runway.

Microsoft’s clever shell game

Let’s take a look at the nuances of interpreting the expansion rates and the games vendors play with numbers.

The Azure revenue trajectory has been quite remarkable as shown within the chart above. Entering COVID, we estimate that Microsoft Azure was generating roughly $4 billion per quarter. Today, our estimates indicate that Azure is running at greater than $21 billion per quarter growing within the high 30% range. Note the revenue flattening in 2024 roughly coincides with the removal of the Enterprise Mobility, Security and Power BI per-user pricing revenue inside Azure definitions. It replaced those low growth businesses with smaller but faster growing AI. As such you see the Azure curve steeply rises into 2025 as Azure AI services kick in. The timing was right and Microsoft disclosed that Azure has surpassed $75 billion.

[See our detailed analysis of the Azure financial disclosures and the definitional changes]

Microsoft’s FY 2025 ends in CY Q2 2025 and the $75 billion figure corresponds to the CY periods comprising Q3 2024 through Q2 2025. In our view, Microsoft cleverly timed its reclassification of Azure revenue and the $75 billion disclosure to underscore its substantial revenue base and its significantly higher growth rate relative to AWS.

But that is the “growth optics” slide – the games people play.

The underside half of the above chart tries to place everyone on the identical field. AWS did ~$30.9 billion IaaS/PaaS within the quarter, up 17.5%. Mapping to the $75 billion figure, now we have Microsoft at ~$22.2 billion, +39%. When you normalize for the ChatGPT uplift and Microsoft’s definitional tweaks, our range for ‘core Azure’ is ~$18.5 billion to $19.0 billion growing within the mid-30s; or, by its previous definitions, 20%. At that size, AWS historically also grew within the mid-30s, so the meme that “AWS is lagging” is usually rate-of-change noise. Google Cloud Platform is ~$7.2 billion, growing ~38% to 39%. Strong, but when AWS was at ~$7 billion, it was growing at ~45%, so this isn’t unprecedented hypergrowth.

Net-net: The market is ripping, but lots of Azure’s pop is AI inference optics. George Gilbert points out that Microsoft doesn’t want to spotlight the proven fact that it has a ChatGPT tiger by the tail. Probably 50% of OpenAI’s revenue goes to COGs – that’s, Azure fees. The purpose is, strip that out and the expansion curves of AWS and Azure converge when normalizing for size of business over time. Nonetheless – AWS stays the dollar leader, Azure is gaining share with these AI/OpenAI tailwinds, and GCP is executing cleanly off a smaller base.

ETR spending data confirms the speed macro

Let’s take a look at a number of the spending data from ETR on the massive three cloud players.

ETR’s Shared-Net Rating data shows spending momentum still favoring Azure. Remember that is account-based not revenue-based. In Jul 2025, Azure posts a 60.2% Shared Net Rating on 986 accounts, well ahead of AWS at 45.3% (N=784) and GCP at 36.2% (N=437) out of 1,777 total shared responses. The 40% dotted line indicates highly elevated spending velocity, so despite AWS’s lower rating it’s still elevated. During the last 18 months Azure has held a persistent 60%+ Net Rating, consistent with AI-driven expansions and Microsoft’s bundle leverage. AWS softened in late 2023, then recovered into the mid-40s — regular but still capacity-limited. GCP’s momentum cooled through 2024 into mid-2025, now mid-30s, possibly reflecting greater, longer deal cycles somewhat than broad-based expansions. The online is: In overlapping accounts, account-share momentum = Azure > AWS > GCP.

Market penetration shows a possible shift in priorities

Here’s an analogous slide but it surely shows the penetration into the information set.

Surprisingly, all three are showing deceleration within the pervasiveness metric. This doesn’t fly within the face of the rapid and accelerating cloud growth rates, nevertheless what it speaks to is spending is shifting to other categories, namely AI; and possibly some pause in AI experimentation on the cloud and a few considered moving toward on-premises – or balancing in that direction. Again, this data doesn’t represent spending levels, somewhat it represents account penetration. Microsoft is ubiquitous and speaks to why Azure’s penetration is so high. AWS continues to broaden its footprint with shared-account penetration GCP continues to be the third player and quite a distance from the leaders.

Trying to 2H 2025 and into 2026

Let’s close with a take a look at things to observe going forward.

Here’s the underside line heading into H2 and the 2026 window on the bull and bear cases: We’re tracking three swing aspects that may resolve the leaderboard — agent adoption, power and silicon. If pilots flip to production and enterprise agents get embedded across Microsoft M365, Google Workspace and custom apps, we may get a second-wave of cloud migration and real productivity gains.

For AWS, which lacks a ubiquitous collaboration software platform it must rely totally on its builders. Offerings around Q Developer are critical for developers to create software and apps on the AWS platform. But when customers don’t see broad-based return on investment and so they suffer from proof-of-concept fatigue, things could slow. After all, we’re at all times watching the macro spending picture as well, which immediately is cautious outside AI.

We’re also taking a look at the pace of CapEx buildouts and permitting – if regulations keep dragging and “short on power” becomes the norm, growth gets slowed by lack of kilowatts, not underlying demand.

For silicon designers, like these three – if price/performance keeps improving not only with Nvidia Blackwell GPUs, but in-house silicon similar to AWS’ Trainium and Google’s TPUs, and higher inference economics, hyperscalers can keep pricing power and this can pull adjoining services. If GPUs remain constrained or customers are concerned about CUDA lock-in and high training costs, that might be a headwind and the bear case is migrations slow and customers may pull back a bit.

Sovereign cloud is a monetization tailwind. Microsoft has the strongest hybrid story of the massive three with its legacy on-premises Azure stack footprint and there’s evidence that its non-cloud business stays strong. Google might be much more incentivized to partner for on-prem deployments, while AWS doubles down on VPC-native sovereignty and has Outposts in its back pocket. Policy is the wild card; a pro-innovation framework with licensing/data-use clarity supports spending, while broad antitrust/tariff shocks or an AI-ROI vacuum will cause buyers to tap the brakes.

Our base case is that pilots begin converting, power eases progressively, custom silicon improves inference unit economics, and sovereignty continues to be in demand. But the massive concern is enterprises proceed to struggle with good data. Bad data will constrict AI payback and agentic adoption. That every one said, CapEx continues to be the mainspring of AI enthusiasm and the Azure playbook is popping. AWS will not be standing still and Google’s quiet execution is showing improved mix and margin, which is encouraging.

Watch these signs: agent adoption, weekly energetic users and token volumes. As well as listen to CapEx versus AI revenue realization, the share of inference on non-GPU silicon, sovereign-cloud RFP velocity and any shift in liability rules.

If the players hit on these, the cloud up-cycle runs all the way through 2026. That is our expectation. That is our base case.

What’s yours? Please get in contact and tell us.

Disclaimer: All statements made regarding corporations or securities are strictly beliefs, points of view and opinions held by SiliconANGLE Media, Enterprise Technology Research, other guests on theCUBE and guest writers. Such statements usually are not recommendations by these individuals to purchase, sell or hold any security. The content presented doesn’t constitute investment advice and mustn’t be used as the premise for any investment decision. You and only you might be chargeable for your investment decisions.

Disclosure: Most of the corporations cited in Breaking Evaluation are sponsors of theCUBE and/or clients of theCUBE Research. None of those firms or other corporations have any editorial control over or advanced viewing of what’s published in Breaking Evaluation.

Image: geralt/Pixabay