Curzio Research framed the latest earnings season as the payoff Wall Street had been demanding after two years of extraordinary capital spending. The underlying results support the core observation. Alphabet reported Google Cloud revenue of $24.8 billion for the June quarter, up 82% year over year; Amazon reported $42.2 billion in AWS sales, up 37%; and Microsoft said Azure revenue topped $100 billion in its fiscal year while Azure growth reached 43% in the June quarter.
The important qualification is that these numbers prove strong demand for cloud services during the AI buildout. They do not yet isolate a clean, audited AI return on investment. The companies report AI as a driver of cloud growth, but they do not disclose a common measure showing what portion of each cloud dollar comes from model training, inference, AI agents, conventional virtual machines, databases, storage, security services, or migration work.
That distinction changes how the “AI payoff” should be read. Wall Street now has proof that the infrastructure owners are monetizing demand. It does not yet have a fully comparable accounting of whether every new dollar of AI-oriented data-center investment earns an attractive return over the life of those assets.
The $370 Billion Figure Is Real, but It Is Not an AI Revenue Number
The combined figure works as a rough scale marker. Annualizing Google Cloud’s $24.8 billion quarter gives about $99 billion; Amazon’s stated AWS run rate is about $169 billion; and Microsoft has now disclosed annual Azure revenue above $100 billion. Together, that is approximately $368 billion to $370 billion.
But treating the total as “AI cloud revenue” would be an overreach. AWS includes its broad infrastructure, platform, database, analytics, security, and application-service businesses. Google Cloud includes Google Cloud Platform and Google Workspace. Azure’s disclosed annual revenue similarly spans a large mix of compute, data, networking, platform services, enterprise software consumption, and AI workloads.
Microsoft’s disclosure is especially significant because it finally gives Azure a public revenue milestone rather than only a percentage growth rate. But Microsoft still does not publish Azure’s operating income as a separate segment, nor does it disclose how much Azure revenue comes from generative AI services or workloads linked to OpenAI. That leaves investors and customers with a high-confidence growth number but a less precise picture of the economics underneath it.
Amazon is more explicit about the scale of its AI business, saying its AI business has passed a $25 billion annual revenue run rate and is growing at a triple-digit percentage. Yet that figure sits within AWS rather than on top of it. It cannot be added to AWS revenue, and it does not establish that all of the recent acceleration in AWS came from AI.
The practical conclusion is straightforward: the $370 billion figure shows hyperscalers have enormous, fast-growing revenue engines available to absorb AI demand. It does not show that $370 billion is revenue generated by AI.
Google Cloud Shows the Strongest Near-Term Conversion of Demand Into Revenue
Alphabet’s second-quarter result is the sharpest example of the current shift. Google Cloud’s 82% revenue growth to $24.8 billion was accompanied by $8.8 billion in operating income, producing an operating margin near 36%. Google said the increase was driven by demand for AI infrastructure and AI solutions.
That margin matters. A cloud business growing quickly while expanding profit is a much different proposition from a vendor spending heavily to support a free or low-margin AI product. Google Cloud’s result suggests that, at least in the latest quarter, customers were purchasing high-value services quickly enough to outpace the immediate cost of serving them.
Alphabet also disclosed a $514 billion cloud backlog, a measure of contracted performance obligations not yet recognized as revenue. It is a striking number, but it needs careful interpretation. Backlog is not cash, and it is not a guarantee that all future revenue will arrive on the same timetable. It can contain multi-year commitments, services not yet delivered, and contracts whose consumption can vary as customers change deployment plans.
Still, the backlog is meaningful because it answers a basic capacity question: Google is not expanding data centers solely on the assumption that applications will appear later. It has a large amount of contracted future business to serve. The company’s capital-spending plans and the backlog are therefore linked by something more concrete than AI enthusiasm.
Alphabet’s cash flow shows the other side of the equation. Reuters reported that the company’s free cash flow turned negative in the second quarter as infrastructure outlays surged, despite the record cloud result. The company can be generating attractive cloud operating income while still consuming cash at the corporate level to build the data centers, power systems, networking, and accelerators needed to meet future commitments.
For enterprise customers, this can translate into a better near-term supply outlook but not necessarily lower prices. When capacity is constrained and contracted demand is high, cloud providers have less incentive to compete aggressively on GPU or premium AI-service pricing.
Azure’s Growth Is a Windows Story, but Its AI Mix Remains Opaque
Microsoft’s Azure growth is particularly consequential for Windows-focused organizations because Azure is tied tightly to the software estate many enterprises already operate: Active Directory and Entra, Windows Server, SQL Server, Microsoft 365, GitHub, Power Platform, Defender, Fabric, and Copilot services.
Microsoft told investors that Azure surpassed $100 billion in annual revenue in fiscal 2026 and grew 43% in the final quarter. Reuters reported that the result eased some investor concerns about the company’s huge AI infrastructure spending, while Microsoft said it had more than 30 million paid Microsoft 365 Copilot seats.
Those are real signs of commercial traction. But they still leave two gaps that matter to CIOs and investors alike.
First, Microsoft’s Azure growth rate combines AI demand with the rest of Microsoft’s cloud portfolio. A customer modernizing a Windows Server estate, increasing SQL consumption, moving VMware workloads, or purchasing cybersecurity services can all contribute to the broader cloud result. Azure is clearly benefiting from the AI cycle, but Microsoft does not provide enough detail to separate AI-created growth from accelerated spending across the company’s larger cloud stack.
Second, capacity has been a constraint, not merely a talking point. Microsoft said earlier in 2026 that it was prioritizing delivery of capacity and improving fleet efficiency, and it has continued to describe demand as strong enough to require massive capital commitments. For an IT department, that means availability of the right GPU configuration, region, storage throughput, networking profile, and service quota can matter as much as a published list price.
The operational risk is that enterprises confuse access to an AI-branded service with guaranteed production capacity. A Copilot pilot can be bought with a license. A production AI workload involving Azure AI Foundry, model endpoints, private data, retrieval pipelines, GPUs, governance, and high availability demands a capacity plan. The systems integrator pitch may be about model choice; the real bottleneck is often compute allocation and data movement.
AWS Demonstrates Why Infrastructure Providers Have an Advantage
AWS posted $42.2 billion in second-quarter sales and $16.6 billion in operating income, giving the segment an operating margin of roughly 39%. Amazon called the 37% growth rate its fastest in 18 quarters and said both AWS AI and its custom-chip business had passed $25 billion annual run rates.
The numbers reinforce a reality that receives less attention than the contest between model makers: Amazon, Microsoft, and Google can make money regardless of whether an enterprise standardizes on OpenAI, Anthropic, Gemini, open-weight models, or a proprietary model.
That does not make the cloud providers invulnerable. Customers can optimize workloads, repatriate predictable compute, use multiple clouds, buy dedicated capacity, or move inferencing to edge and on-premises systems. But training and serving frontier-scale models, or running high-volume enterprise inference, demands a mix of chips, data-center power, networking, storage, orchestration, and managed services that few organizations can assemble quickly.
AWS’s result also illustrates why investors are focusing on operating income rather than revenue growth alone. A 39% segment operating margin suggests AWS remains capable of financing a substantial part of Amazon’s broader AI buildout. Yet Amazon’s corporate free cash flow can still come under pressure when investment in data centers rises faster than cash generated from operations. A profitable cloud segment and a cash-intensive company can coexist.
The Missing Metric Is Return on Incremental Capital
S&P Global Ratings estimates that Alphabet, Amazon, Meta, Microsoft, and Oracle will spend more than $700 billion on capital expenditures in 2026. Microsoft has outlined roughly $190 billion in calendar-year capital expenditure, while Alphabet raised its own spending outlook as cloud demand accelerated. The scale is now large enough that growth rates alone cannot settle the investment case.
The useful test is not whether cloud revenue is growing; it plainly is. The question is whether incremental cloud operating profit, over several years, exceeds the cost of the incremental data-center and accelerator capacity required to produce it.
That calculation takes time because the costs arrive first. Construction, power agreements, network gear, chips, depreciation, leases, and financing hit before all of the capacity is occupied. Revenue is then recognized as customers consume services, sometimes under contracts that run for years. A bumper quarter can demonstrate demand, but it cannot prove the full life-cycle return of a data-center fleet commissioned in 2026.
For IT leaders, the lesson is less abstract. The hyperscalers’ current spending makes it safer to plan for AI workloads that need serious scale, but it also strengthens their negotiating position in the short term. Capacity-constrained platforms can dictate reservation terms, regional availability, service limits, and the premium attached to top-tier accelerators.
The latest results establish that AI has become a major growth catalyst for cloud computing. They do not establish that the AI buildout is cheap, complete, or evenly profitable. Customers lining up for capacity are real; the enduring payoff will depend on whether those customers keep consuming enough compute after the current race to deploy AI moves from experimentation into routine production workloads.
References
- Primary source: Curzio Research
Published: August 7, 2026 at 6:26 PM UTC
Loading…
www.curzioresearch.com - Related coverage: apple.com
Apple reports second quarter results - Apple
Apple today announced financial results for its fiscal 2026 second quarter ended March 28, 2026.www.apple.com