The Capital Expenditure Smoke Signal
The most direct, if blunt, signal of AI cost pressure is capital expenditure, or CapEx. This is the money a company spends on physical assets like servers, land, and buildings. In the age of AI, this line item is exploding. Microsoft's CapEx guidance
for calendar year 2026 is reportedly approaching an immense $190 billion, a figure driven almost entirely by the need to build and equip new data centers with power-hungry, specialized AI chips. When you see CapEx figures rising dramatically quarter after quarter, as they have for Microsoft, it’s the clearest indication that the company is in a massive investment cycle. Analysts watch this number obsessively because it’s a direct drain on free cash flow. The company is spending money now in the hope of generating revenue later, and the sheer scale of the spending shows just how much it costs to lay the foundation for the AI economy.
Reading the Cloud Margins
The next clue is more subtle and lives in the gross margin of Microsoft's Intelligent Cloud division, which is dominated by Azure. Gross margin is the percentage of revenue left after accounting for the cost of goods sold (COGS). For a cloud business, COGS includes the immense operational costs of running data centers—like power and cooling. While AI drives more Azure usage and revenue, it's also more expensive to run than traditional cloud services. This creates a tug-of-war. Microsoft's earnings reports have noted that cloud gross margin percentage has decreased due to the costs of scaling its AI infrastructure. Even a small dip of one or two percentage points, from 66% to 64% for instance, can translate to billions of dollars at Microsoft's scale and signals that the high cost of AI is eating into the profitability of its core cloud engine.
Listen for 'Capacity Constraints'
A key phrase to listen for during the investor call that follows an earnings report is “capacity constraints.” It’s a double-edged sword that both Microsoft’s CEO and CFO have used. On one hand, it sounds great—it means demand for their AI services is so high that it’s outstripping their available supply of computing power. But it's also an admission of the problem. It means that even with record-breaking capital expenditures, the company still can't build data centers fast enough. This signals that the spending spree isn’t over; in fact, it has to continue at a breakneck pace just to keep up. When executives talk about demand exceeding supply, they are telling investors that the massive CapEx numbers are not a temporary spike but a sustained reality for the foreseeable future.
The Shift in Pricing Models
Finally, pay attention to how the company talks about pricing for its AI products, like Copilot. Initially sold on a simple per-user subscription, the underlying economics are more complex. An employee asking an AI to summarize an email and an autonomous agent running a complex, multi-hour task consume vastly different amounts of expensive computing power. To manage these variable costs, vendors like Microsoft are subtly shifting toward consumption-based or credit-based billing, as seen with some GitHub Copilot plans. When you hear executives talking about evolving their pricing models, it’s another clue. It reflects the internal pressure to align the price customers pay more closely with the actual, highly variable cost of delivering the AI service, protecting their margins from runaway usage.
