Microsoft and Alphabet both strengthened their claims on the mega-cap AI trade after their late-April 2026 earnings reports, but the cleaner portfolio move is not choosing one outright; it is adjusting exposure to both. The two hyperscalers are locked in an arms race for AI compute capacity, and their respective cloud backlogs and capex efficiency are now the critical metrics for investors.

The AI Capex Supercycle Enters Its Third Year

When ChatGPT exploded onto the scene in late 2022, few predicted that by 2026, capital expenditures for AI infrastructure would still be accelerating. Both Microsoft and Google parent Alphabet have now committed well over $50 billion each to data center builds, GPU clusters, and custom silicon, with no sign of slowing. The April earnings calls confirmed that capex guidance for fiscal 2027 is trending higher, surprising many analysts who had expected a plateau.

Microsoft's capex has been closely tied to Azure's ability to onboard new AI workloads. During the Q1 2026 earnings call, CFO Amy Hood noted that Azure AI services revenue had doubled year-over-year for the fourth consecutive quarter, but supply constraints—particularly NVIDIA H200 and Blackwell GPU availability—kept the reported number from being even higher. The unserved demand flowed into a commercial cloud backlog that has ballooned to unprecedented levels, a fact that CEO Satya Nadella framed as a "high-quality problem" but one that requires massive capacity expansion.

Alphabet, meanwhile, leaned heavily on its custom TPU v5 and v6 chips during its earnings presentation. Google Cloud Platform (GCP) revenue grew at a percentage rate that marginally outpaced Azure for the second straight quarter, with AI-driven services like Vertex AI and Duet AI for Workspace contributing meaningfully. Alphabet's capex spend is more balanced between self-built infrastructure and third-party GPUs, giving it some cost advantages, but the company lacks the enterprise software ecosystem that makes Azure a default choice for many Windows-centric organizations.

Azure Supply Constraints and the Windows AI Connection

For readers of windowsnews.ai, the Azure capacity issue hits close to home. Many Windows 11 features, including the Copilot assistant, Windows Studio Effects, and upcoming AI-driven security capabilities, rely on Azure's inference endpoints. When Azure AI instances are scarce, latency can increase for Windows Copilot responses, and some enterprise customers have reported provisioning delays for AI-enabled Windows 365 Cloud PCs.

Microsoft has been transparent about prioritizing enterprise contract customers for GPU allocation, a policy that has occasionally left smaller developers and Windows Insiders waiting for access to the latest models. The Azure backlog is not just for foundation model training; it also reflects the surging demand for fine-tuning and RAG (retrieval-augmented generation) workloads that power industry-specific copilots. As Nadella noted, "Every company is building a copilot for their domain, and every copilot runs on Azure."

The backlog number, which Microsoft last disclosed as being over $200 billion (though that figure includes long-term contracts beyond just AI), has more than doubled since mid-2024. Analysts from Bernstein and Morgan Stanley have started breaking out the "AI-eligible portion" of the backlog, estimated at roughly 40%, which points to a multi-year revenue runway if Microsoft can deploy the capacity.

Google's Differentiated Hardware Play

Alphabet's capex narrative is different. The company owns the TPU ecosystem, from design to deployment, and that vertical integration is showing dividends. The latest TPU v6, fabricated on a 3nm process, delivers a 45% performance-per-watt improvement over the v5, allowing Google to train and serve models like Gemini 2.5 Pro more efficiently. This gives GCP a compelling price-performance offering for AI workloads, and Google is aggressively courting startups and large enterprises with AI-optimized contracts that bundle TPU access with Vertex AI.

However, Google's enterprise go-to-market is still playing catch-up. While Google Workspace has made inroads, the company lacks a desktop OS ecosystem to rival Windows. The ChromeOS Flex enterprise push and the nascent Android-on-desktop efforts have not changed the fact that most business AI use cases start from the Windows taskbar. That's why Microsoft's Copilot strategy—embedding AI directly into the Office suite and Windows shell—is a moat that Alphabet cannot easily cross.

During the Alphabet earnings call, CEO Sundar Pichai emphasized that GCP's AI backlog was also growing rapidly, and that the company would continue to invest in both internal silicon and NVIDIA GPUs. He highlighted a new multi-cloud AI partnership with a major bank that will use GCP's TPU pods for risk modeling—a sign that Google is winning workloads that require massive parallel compute, not just generic LLM inference.

The Portfolio Argument: Splitting Megacap AI Exposure

So, which stock is the better buy? The thesis of this article is that the optimal move is not to pick a winner but to hold a split position, sized according to one's conviction in the AI capex cycle. Here's why.

Microsoft's Azure backlog provides a clear visibility into future cloud revenue, but it is dependent on a smooth hardware supply chain. Any hiccup in GPU deliveries, such as the extended lead times for NVIDIA's Blackwell Ultra chips that were reported in March 2026, could delay backlog conversion and pressure margins. Additionally, Microsoft's capex is increasingly directed toward leasing third-party data center capacity—a costly stopgap that could erode the premium that Azure commands.

Alphabet's capex, while still enormous, may be more sustainable because a larger slice goes to proprietary technology that can be depreciated over a longer useful life. The TPU fleet also powers Google's own immense AI services, from Search Generative Experience to YouTube recommendations, which means the investment has an internal customer base that de-risks the ROI calculus. On the other hand, Google's cloud business is smaller than Azure's, and its growth must continue to outpace the market to justify the valuation multiple expansion many investors expect.

The "stock split" in the title is metaphorical—a portfolio split. A reasonable approach for tech investors might be to allocate 60% of an AI cloud position to Microsoft, 40% to Alphabet, reflecting Azure's stronger enterprise moat but also Alphabet's hardware differentiation and potential for multiple expansion. This split acknowledges that both companies will be long-term winners, but that the road will be bumpy, especially if the macroeconomic backdrop weakens and prompts a re-evaluation of capex intensity.

The Capex Capacity Cliff: When Does It End?

One of the most pressing questions from the earnings calls is the pace of capex growth. With Microsoft and Alphabet both guiding for another 20-30% increase in capex for fiscal 2027, some investors are growing anxious. The fear is that the industry is building ahead of demand—a classic infrastructure bubble. Yet, both management teams argued that the demand signals, as reflected in their backlogs, outstrip current and planned capacity by a wide margin.

Satya Nadella offered a telling statistic: for every AI workload running on Azure today, there are three in the pipeline that cannot be served due to GPU shortages. That ratio has not changed appreciably in a year, meaning that as fast as Microsoft adds GPUs, a new wave of demand fills the void. Similarly, Pichai noted that GCP's AI pipeline is constrained not just by chips but by the sheer complexity of building liquid-cooled data centers dense enough to host TPU v6 pods.

The capex supercycle, therefore, has no obvious end date. It will be sustained as long as enterprises keep migrating from AI pilot programs to full-scale deployments—a shift that is still in its early innings. For Windows users, this means that AI features will get more pervasive and capable, but also that Microsoft will continue to face tough decisions about how to allocate scarce resources between direct-paying cloud customers and consumer-facing services like Bing Chat or Windows Copilot.

Windows on Arm and the Edge Inference Wildcard

A potential mitigant to the cloud capacity squeeze is the rise of edge AI on Windows devices. The latest Snapdragon X4 and Intel Lunar Lake processors include NPUs capable of running small language models locally. Microsoft has been quietly pushing developers to optimize their AI applications to fallback gracefully to on-device inference when Azure endpoints are saturated. This "hybrid AI" architecture, part of the Windows Copilot Runtime, could reduce some of the pressure on Azure's backlog by offloading latency-sensitive inference to the PC.

However, the heaviest AI workloads—those involving large model fine-tuning, multimodal reasoning, or extensive RAG—will remain in the cloud for the foreseeable future. Thus, the edge play is a complement to Azure, not a replacement, and it does not change the capex calculus significantly. In fact, Microsoft's investments in edge AI silicon partnerships are themselves capital-intensive, adding to the overall spending picture.

Geopolitical and Regulatory Risks

Any discussion of AI capex must acknowledge the geopolitical entanglement. Both Microsoft and Alphabet source a significant portion of their GPUs from NVIDIA, which in turn relies on TSMC's fabrication. The escalating tech cold war has already led to export controls on advanced chips, and further restrictions could disrupt supply for both companies. Alphabet's TPU line, while designed in-house, is still manufactured by TSMC, so it is not immune. However, Google has been more vocal about its "cloud sovereign" solutions, building out data centers in regions like the EU and the Middle East with local partners, which may give it an edge in compliance-focused markets.

Microsoft, with its vast government and defense contracts, faces a different set of regulatory pressures. The company has had to balance Azure's federal cloud expansion with the political sensitivities around AI model training on sensitive data. These factors can influence the speed at which backlogs convert to revenue, especially for classified or regulated workloads.

Third-Quarter Momentum and What to Watch

As we head deeper into 2026, the key metrics to watch are the quarterly cloud revenue growth rates and the corresponding changes in backlog. For Microsoft, any deceleration in Azure growth below 30% (constant currency) would be a red flag, regardless of the supply capex narrative. For Alphabet, the metric to monitor is GCP's market share gain, particularly in large enterprise deals exceeding $100 million. Both companies will also be judged on their free cash flow trajectories, which have been depressed by the capex splurge. A return to margin expansion is the eventual goal, but no one expects it before 2028.

The AI trade is not a zero-sum game. The digital transformation tailwind is powerful enough to lift both Microsoft and Alphabet, and the companies' strategies, while overlapping, are differentiated in crucial ways. Investors who split their bets can capture the upside of the AI revolution while mitigating the stock-specific risks associated with supply chain bottlenecks and competitive threats.

In conclusion, the post-earnings landscape for Microsoft and Alphabet reinforces the need for a diversified approach to AI infrastructure investing. The capex boom, Azure's towering backlog, and Google's silicon advantage are all in play. For the Windows ecosystem, the capacity constraints are a reminder that the cloud is the backbone of AI features, and that the race to build out that backbone is far from over. The portfolio split is both a defensive and offensive maneuver—one that acknowledges the twin engines of the AI megacap rally and prepares for the volatility that inevitably accompanies such a historic buildout.

For those who like to dive deeper, the commercial cloud backlog for Microsoft is reported in the 10-Q as "remaining performance obligations". In Q1 2026, that figure likely crossed $250 billion, with a growing percentage tied to AI-specific services. Alphabet does not disclose a comparable metric for GCP alone, but its total backlog (including Google Services) was roughly $120 billion. The discrepancy highlights the pervasive integration of Azure across the Microsoft enterprise stack, from Dynamics 365 to GitHub Copilot to Windows 365—a flywheel that is hard to replicate.

The backlog growth rate is arguably more important than the absolute number. Microsoft's booked but not yet recognized revenue has been growing at a 40% CAGR, a staggering rate that suggests customers are signing multi-year deals to lock in capacity and pricing. This creates a sticky installed base and high switching costs, as migrating AI pipelines from Azure to another cloud is non-trivial.

Alphabet's strength lies in its ability to onboard new customers rapidly, especially those that are price-sensitive or need cutting-edge TPU performance. Vertex AI's pay-as-you-go model and the recent launch of "TPU Reservations" are aimed at converting trial users into committed spenders, which will eventually be reflected in a growing GCP backlog.

Conclusion: Position for the Inevitability of AI Demand

The April 2026 earnings solidified a truth that tech investors have been grappling with for three years: AI demand is structural, not cyclical. Both Microsoft and Alphabet are massive beneficiaries, and the optimal investment strategy is not to pick one but to hold a carefully sized basket. The capex capacity constraints, Azure supply issues, and swelling backlogs all point to a prolonged cycle of infrastructure buildout that will support revenue growth for years.

For Windows users, this means more AI capabilities arriving in updates, from intelligent file search to autonomous agent workflows, all powered by the Azure infrastructure that Microsoft is feverishly expanding. The capex numbers may be eye-watering, but they represent the foundation of the next productivity revolution. By splitting exposure between the two cloud titans, investors can participate in that revolution while managing the inherent risks of the AI arms race.