AI Capex Boom 2026: Hyperscalers Spend $420 Billion as Investors Question Returns

Artificial intelligence infrastructure has become the largest corporate capital expenditure cycle in modern history. The four largest US hyperscalers — Microsoft, Amazon, Alphabet and Meta — are on track to spend a combined $420 billion on capital projects in 2026, up 38% from $305 billion in 2025.

The scale is unprecedented. For context, that figure exceeds the annual GDP of economies such as Belgium or Sweden, and it represents roughly 1.4% of total US gross domestic product.

Yet the spending boom is now colliding with a harder question: will it pay off? Depreciation expenses are rising sharply, data center power demand is straining electricity grids, and some investors worry that the returns may not justify the outlay.

Key Takeaways

• Hyperscaler capex is projected to reach $420 billion in 2026, up 38% from 2025.
• Microsoft alone plans to spend more than $80 billion on AI-enabled data centers in fiscal 2026.
• Data centers are projected to consume 9% of US electricity by 2030, up from about 4% today.
• Depreciation from AI infrastructure is expected to add $40 billion to $50 billion in annual costs across the hyperscalers by 2027.
• Data center construction spending reached a record $41 billion annualized rate in mid-2026.

Why Are Hyperscalers Spending So Much?

The answer is demand for AI compute. Training and running large language models requires massive amounts of processing power, memory and networking — all housed in data centers that must be built, powered and cooled.

Microsoft has announced plans to invest more than $80 billion in AI-enabled data centers during fiscal 2026 alone. Amazon, Alphabet and Meta have each announced similarly ambitious programs. Collectively, the four companies are building the physical backbone of the AI economy.

The spending is not limited to the US. Investment is flowing into data centers in Europe, the Middle East and Asia, driven by the same demand for compute capacity and by a desire to localize AI infrastructure for regulatory and latency reasons.

The Supply Chain Behind the Boom

The capex surge has created a powerful secondary boom for suppliers. Semiconductor makers, networking equipment providers, power management companies, and construction firms are all benefiting.

Nvidia remains the most visible beneficiary, but the spending extends far beyond chips. Transformers, switchgear, cooling systems, backup generators and fiber-optic cable are all in high demand, and lead times for some components have stretched to more than a year.

How Does the Spending Compare Over Time?

The trajectory is remarkable. In four years, hyperscaler capex has nearly tripled. No comparable investment cycle in recent corporate history has been this large or this fast.

What Are the Risks of the AI Capex Boom?

Three risks stand out.

First, depreciation. The servers and infrastructure being built today have useful lives of perhaps five to six years. As those assets age, depreciation charges will rise. Analysts estimate that AI infrastructure depreciation could add $40 billion to $50 billion in annual costs across the hyperscalers by 2027, pressuring earnings.

Second, power constraints. Data centers are projected to consume 9% of US electricity by 2030, up from about 4% today. In some regions, grid operators are already warning that new data center connections may be delayed for years. Power availability could become the binding constraint on AI expansion.

Third, return uncertainty. The revenue from AI services is growing, but it is not yet clear whether it will grow fast enough to justify the capital deployed. If AI monetization disappoints, the industry could face a painful period of overcapacity.

The Depreciation Question

Depreciation is a slow-moving but powerful force. Unlike a one-time cost, it hits earnings every quarter for years. As the 2024-2026 capex wave ages, the drag on reported profits will grow.

Some companies are extending the assumed useful lives of their AI equipment to spread the cost, a practice that analysts are watching closely. If useful lives prove shorter than assumed, write-downs could follow.

How Does This Affect Small Businesses and SMEs?

The AI capex boom affects small businesses in two ways.

First, through costs. Energy-intensive businesses in regions with heavy data center concentration may face higher electricity prices as demand rises. In some markets, data centers are already the largest single source of new power demand.

Second, through opportunities. SMEs that supply components, logistics, construction services or specialized labor to the data center supply chain are seeing strong demand. Companies that can help data centers operate more efficiently — particularly in cooling and power management — are finding a growing market.

But for most SMEs, the AI capex boom is an indirect story. They benefit from the productivity gains of AI tools, but they are not direct participants in the infrastructure build-out.

What Do Investors Need to Watch?

Three metrics will reveal whether the AI capex cycle is sustainable.

• Revenue growth from AI services. Cloud and AI revenue must grow fast enough to justify the capital deployed.
• Depreciation trends. Rising depreciation charges will pressure margins; watch for changes in useful-life assumptions.
• Power availability and pricing. Electricity constraints could cap growth in some regions, and rising power costs add to operating expenses.

Investors are also watching free cash flow. Heavy capex reduces free cash flow in the near term, and some companies have seen their free cash flow decline even as revenues rise. If that trend continues, dividend growth and buyback capacity could be affected.

Conclusion: A Historic Investment Cycle With Uncertain Payoff

The AI capex boom is one of the defining business stories of 2026. It is reshaping corporate balance sheets, supply chains, energy markets and investor expectations.

The spending is real and the demand for AI compute appears durable. But the returns are not yet proven, and the risks — depreciation, power constraints, overcapacity — are significant.

For investors, the question is not whether AI will matter; it is whether today's spending will generate returns that justify the price. For businesses and workers, the boom is creating opportunities in the supply chain, but also raising costs and reshaping the competitive landscape.

Frequently Asked Questions (FAQ)

How much are hyperscalers spending on AI in 2026?

Microsoft, Amazon, Alphabet and Meta are projected to spend a combined $420 billion on capital expenditure in 2026, up 38% from $305 billion in 2025. Microsoft alone plans to invest more than $80 billion in AI-enabled data centers in fiscal 2026.

Will the AI capex boom generate positive returns?

That is the central question for investors. AI service revenues are growing, but depreciation costs are rising and power constraints are emerging. If AI monetization grows fast enough, the investment will pay off; if not, the industry could face overcapacity and write-downs.

How does the AI boom affect electricity prices and power grids?

Data centers are projected to consume 9% of US electricity by 2030, up from about 4% today. In some regions, grid operators warn that new data center connections may be delayed for years. Rising power demand could increase electricity prices for other users in affected markets.

What does this mean for small businesses and SMEs?

Most SMEs benefit indirectly through AI productivity tools rather than direct participation. However, SMEs in the data center supply chain — construction, logistics, components, specialized labor — are seeing strong demand. Some energy-intensive businesses may face higher electricity costs in regions with heavy data center concentration.

What should investors watch in the AI capex cycle?

Three metrics matter most: revenue growth from AI services, depreciation trends, and power availability and pricing. Investors should also watch free cash flow, since heavy capex reduces near-term cash generation and can affect dividend growth and buyback capacity.