The AI Spending Boom Is Approaching $1 Trillion. Can Revenue Catch Up?

The AI boom has become an infrastructure race. Amazon, Google, Microsoft and their peers are building data centers, buying accelerators and memory, and securing power capacity before competitors can claim it. The assumption behind that spending is straightforward: demand for AI models, cloud computing and enterprise applications will eventually become large and profitable enough to support the cost of the build-out.

The concern is equally straightforward. Infrastructure spending is accelerating faster than the AI revenue that is clearly visible in company accounts. That makes the gap between investment and monetization the central question for investors.

The scale of the AI infrastructure build-out

The Economist estimates that America’s largest technology companies spent about $450 billion on infrastructure in 2025, much of it related to artificial intelligence. It puts spending at roughly $900 billion in 2026 and $1.4 trillion in 2027, covering chips, data centers, power and related capacity. These are broad infrastructure estimates, not a pure measure of AI-only capital expenditure.

A separate Goldman Sachs estimate puts global AI-related investment at about $1 trillion in 2026, including approximately $581 billion in the United States. That calculation is broader than a simple tally of hyperscaler capital expenditure because it also accounts for private companies, non-hyperscaler businesses and investment outside the United States.

Those figures are not contradictory. They measure different universes:

• The Economist’s estimate focuses on spending by America’s largest technology companies on the physical infrastructure supporting AI.
• Goldman Sachs’s estimate covers global AI-related investment and adjusts for spending that a hyperscaler-only measure omits.

Comparing the figures without accounting for those differences can make the investment cycle appear either overstated or understated.

What the major technology companies are spending

Company-level estimates are difficult to compare because fiscal calendars, accounting methods and the share of capital expenditure devoted to AI vary. Still, the direction is clear: guidance has been repeatedly revised upward as demand for computing capacity and memory remains strong.

Amazon, for example, raised its expected 2026 cash capital expenditure from about $200 billion to approximately $220 billion, saying that most of the spending would support AI and AWS. Amazon, Google and Microsoft together were reported to be approaching roughly $595 billion in 2026 capital expenditure, although the figures are not perfectly comparable across the three companies.

The broader group of hyperscalers—including Meta alongside Amazon, Alphabet and Microsoft—has been estimated at roughly $700 billion to $900 billion in 2026 capital expenditure, depending on which companies and accounting definitions are included. The safest conclusion is not a single precise company ranking, but that the leading cloud and technology platforms are committing hundreds of billions of dollars each year to capacity that is being built largely for AI workloads.

Borrowing is helping finance the race

The build-out is no longer being funded solely from operating cash flow. Reporting citing Goldman Sachs estimates that hyperscalers spent about $405 billion on capital expenditure in 2025, with the figure expected to reach approximately $750 billion in 2026 and nearly $1.2 trillion in 2027. Investment-grade bond issuance by hyperscalers totaled $108 billion in 2025. By the first half of 2026, issuance had reportedly reached $194 billion, with roughly $250 billion expected for the full year.

Debt does not automatically make the investment irrational. Large cloud companies generally have substantial cash generation and access to credit markets. But borrowing raises the cost of being wrong. If AI demand, pricing or utilization falls short of expectations, companies could be left servicing debt against data centers and hardware whose value depreciates quickly.

Why AI revenue has not yet closed the gap

AI-related revenue is growing rapidly, but The Economist’s central argument is that it is not growing fast enough to match the infrastructure spending surge. The comparison is also difficult because AI revenue is often embedded in broader businesses rather than reported as one clean line item.

AI-related sales may appear through cloud consumption, enterprise software subscriptions, productivity products, advertising improvements or early application deployments. That makes it difficult to determine how much revenue is directly attributable to AI, what margins it carries and whether it is recurring enough to support the next round of capital spending.

The relevant investor question is therefore not simply whether companies are selling AI products. It is whether the revenue generated per dollar of infrastructure investment can rise quickly enough to cover:

• the cost of new data centers and power capacity;
• accelerator, networking and memory purchases;
• the rapid depreciation and replacement of computing hardware; and
• the interest expense associated with additional borrowing.

The semiconductor industry is absorbing the shock

The infrastructure boom is flowing directly into semiconductors. AI systems require accelerators, advanced networking, storage and large quantities of high-bandwidth and conventional memory. Supply constraints in memory, advanced packaging and leading-edge manufacturing have made the chip market one of the clearest beneficiaries of the spending cycle.

WSTS’s earlier spring 2026 forecast projected global semiconductor sales of $1.51 trillion, up 90% from the prior year. It expected memory revenue to rise by roughly 250% and exceed $800 billion, driven by AI infrastructure, high-bandwidth memory and accelerated computing.

The forecast was later revised upward. After incorporating second-quarter 2026 data, WSTS projected a global semiconductor market of about $1.655 trillion in 2026—roughly 108% annual growth—and approximately $2.1 trillion in 2027, with growth of about 29%.

Omdia offered a separate forecast of 94.1% semiconductor-revenue growth in 2026, saying that AI demand was outpacing the industry’s ability to produce and package chips. It also projected that memory would account for more than half of total semiconductor revenue.

These estimates should not be blended into one definitive number. They are different forecast vintages and methodologies. But together they show how strongly AI infrastructure demand is influencing expectations for chips, memory and related manufacturing capacity.

The trillion-dollar test for AI investors

The AI investment thesis has two parts. The first is already visible: companies are spending aggressively because compute capacity, power and advanced chips are strategic constraints. Waiting may mean paying more later or losing customers to rivals with available capacity.

The second part remains unproven at the required scale: AI applications must generate durable revenue, attractive margins and measurable productivity gains quickly enough to justify continual infrastructure expansion.

If monetization catches up, the current spending cycle could prove rational—even if individual investments are eventually written down or redirected. If it does not, the consequences could include weaker cloud returns, excess capacity, greater debt-service pressure and a sharp reassessment of AI-linked assets.

That is why the most important number is not simply $1 trillion of investment. It is the return that investment ultimately produces.