AI is starting to look profitable, but the infrastructure bet still has to pay off

There are two different AI businesses being discussed at once.

The first is AI attached to existing platforms: cloud computing, Microsoft 365-style productivity software, enterprise contracts and ad systems. This is where monetization is clearest because customers and billing channels already exist ^[17][45][56].

The second is AI infrastructure built ahead of demand: GPUs, servers, data centers and networking. That is still a capital-heavy bet. Liontrust, citing Omdia and Futurum Group estimates, said nearly $500 billion of 2026 spending is directly tied to AI infrastructure ^[2]. Business Insider reported that Amazon, Microsoft, Google and Meta were planning up to $725 billion in 2026 capital expenditures after first-quarter earnings updates ^[8].

That split matters. AI can be profitable in one part of the stack while the broader infrastructure race still has an uncertain payback period.

Where the profit signal is strongest

For Microsoft, the public signal is indirect rather than a clean AI profit-and-loss statement. Its FY2025 performance release reported revenue up $36.6 billion, or 15%, with Intelligent Cloud revenue driven by Azure and Productivity and Business Processes revenue driven by Microsoft 365 Commercial cloud ^[17]. Microsoft also reported gross margin dollars up $22.9 billion, or 13%, even as cost of revenue increased $13.7 billion, or 19%, driven by Microsoft Cloud growth ^[17].

Alphabet’s signal is more explicit in the cited investor materials. In its Q4 2025 earnings call, Alphabet said annual revenue exceeded $400 billion for the first time, Google Cloud grew 48% to an annual run rate above $70 billion, and Cloud backlog rose 55% quarter over quarter to $240 billion, driven by demand for AI products ^[56]. In Q1 2026, Alphabet’s total revenue rose 22% to $109.9 billion, while Google Cloud revenue grew 63% to about $20 billion ^[51]. MarketBeat’s Q1 2026 summary said Google Cloud operating income tripled and operating margin reached 32.9% ^[49].

Meta shows a different route: AI improving a core advertising engine rather than primarily selling cloud capacity. Business Insider reported that Meta’s ad revenue remained its powerhouse and that AI was improving ad targeting, while also reporting that Meta expected up to $135 billion of 2026 capex, including AI infrastructure ^[45]. Meta’s own 2025 results put full-year capital expenditures, including principal payments on finance leases, at $72.22 billion and said future expense growth would be driven mainly by infrastructure costs, including third-party cloud spend, depreciation and infrastructure operating expenses ^[38].

The pattern is consistent: AI monetizes fastest when it strengthens a business that already has scale.

Why this does not prove all AI is profitable

The clearest earnings evidence still comes through cloud, software and advertising segments, not through a standalone AI segment. Microsoft’s cited release reports Azure, Microsoft 365 Commercial cloud and Microsoft Cloud cost dynamics, not a separate AI P&L ^[17]. Alphabet’s disclosures show AI demand inside Cloud growth and backlog, but they do not prove that every model, feature or inference workload is independently profitable ^[56].

That distinction is crucial for investors and operators. Revenue growth can be real while infrastructure return on investment remains unsettled. A cloud provider can be compute-constrained today and still face a harder question later: whether the new capacity earns attractive margins once it is deployed.

The $500 billion question is absorption

The capex boom changes the question from whether Big Tech can afford AI to whether customers can absorb enough AI compute at profitable prices. One recent analysis framed the core constraint as absorption: the largest platforms can fund the buildout for now, but enterprise demand has to consume the new compute profitably ^[1].

If demand stays ahead of capacity, the infrastructure buildout can support cloud revenue growth and backlog conversion. If capacity arrives faster than paid workloads, the same assets become depreciation and operating-cost pressure. Microsoft already reported cloud-driven cost of revenue growth ^[17], and Meta said infrastructure costs would drive the majority of future expense growth ^[38].

This is why strong AI-linked revenue and investor concern can coexist. The revenue evidence is getting better, but the denominator is also getting much larger.

What to watch next

• Cloud growth versus capex growth. Google Cloud grew 63% to about $20 billion in Q1 2026, but Big Tech’s 2026 capex plans were still being revised higher ^[51][8].
• Margins, not just sales. MarketBeat reported Google Cloud operating margin at 32.9%, while Microsoft reported higher gross margin dollars alongside higher cloud-related cost of revenue ^[49][17].
• Backlog conversion. Alphabet’s Q4 2025 Cloud backlog reached $240 billion after a 55% quarter-over-quarter increase; the next test is how much of that backlog converts into profitable revenue ^[56].
• Infrastructure cost pressure. Meta and Microsoft both point to rising infrastructure or cloud costs, including depreciation, operating expenses and cloud cost of revenue ^[38][17].
• Capex guidance. Liontrust noted that Alphabet’s 2026 capex guidance had been revised upward to $180 billion, while Business Insider reported that the combined 2026 capex plan for Amazon, Microsoft, Google and Meta had climbed toward $725 billion ^[2][8].

Bottom line

AI is starting to pay off where it is sold through businesses that already know how to monetize compute, software and advertising. Microsoft and Alphabet provide the clearest public evidence because AI demand is showing up inside Azure, Microsoft 365 Commercial cloud and Google Cloud ^[17][51][56]. Meta’s case is different, but still fits the pattern: AI can improve advertising performance while the company absorbs a much larger infrastructure bill ^[38][45].

The sector-wide verdict is still pending. Big Tech can show real AI-linked revenue momentum, but it has not yet proven that the 2026 infrastructure buildout, including nearly $500 billion directly tied to AI infrastructure in one estimate, will earn attractive returns across the cycle ^[2][8].