Amazon, Google, Microsoft, and Meta are collectively spending nearly $700 billion on AI infrastructure in 2026 alone — a figure that rivals Sweden's entire GDP. The race to control the AI stack is entering its most expensive phase yet.
When historians look back at the 2020s, one number may define the decade more than any other: $700 billion.
That is the approximate combined capital expenditure that Amazon, Google's parent Alphabet, Microsoft, and Meta are planning for 2026 alone — an amount that rivals the annual GDP of entire nations, and represents a near-doubling from the roughly $365 billion these four companies spent in 2025. This is year three of a spending race that shows absolutely no sign of slowing down.
The vast majority of this capital — approximately 75%, or around $450 billion — is being directed at AI-specific infrastructure: Nvidia chips, custom silicon, data centers, networking, and liquid cooling systems. The question on every analyst's mind is the same: at what point does the spending generate returns that justify the scale?
| Stat | Value |
|---|---|
| Combined Big Four AI capex in 2026 | ~$700B — up ~74% year-over-year |
| Share directed at AI infrastructure | 75% of hyperscaler capex |
| Amazon's plan | $200B — most aggressive capex in company history |
Who Is Spending What
Amazon (AWS) — $200B
The largest capex plan of any company in history. Primarily earmarked for AWS AI workload infrastructure. Amazon may need to raise equity and debt to fund it — and faces projected negative free cash flow in 2026.
+52% YoY from $131.8B in 2025.
Alphabet (Google) — $175–185B
Doubling down on Gemini AI models, Vertex AI enterprise platform, and Google Cloud expansion. Nearly doubles 2025 expenditure, reflecting Google's determination to defend its AI leadership.
+96% YoY from $91B in 2025.
Microsoft — ~$145B
Already spent $37.5B in its most recent single quarter. Annualized, this puts Microsoft on track for $120–145B in fiscal 2026. OpenAI partnership remains central to its AI cloud strategy.
+45% YoY from ~$100B in 2025.
Meta — $115–135B
Most aggressive infrastructure buildout in company history. Includes a 1GW data center in Ohio and a Louisiana facility that could eventually scale to 5GW — powering its Llama open-source ecosystem and AI ad platform.
+75% YoY from $72B in 2025.
What Is All This Money Building?
The spending is not evenly distributed across AI projects. The largest single allocation is for AI compute — Nvidia H100 and H200 clusters, custom TPUs and Trainium chips, and the data centers to house them.
The second-largest allocation is networking: moving data between chips fast enough to train and serve frontier models requires custom silicon and fiber infrastructure that doesn't exist off the shelf.
A growing third allocation is going to power and cooling — liquid cooling systems and dedicated power infrastructure — as data center heat loads reach levels that air cooling cannot handle. In some cases, Big Tech companies are now directly negotiating with nuclear power plant operators and building dedicated generation capacity to fuel their AI ambitions.
| Company | 2026 Capex | 2025 Capex | YoY Change | Primary AI Focus |
|---|---|---|---|---|
| Amazon | ~$200B | $131.8B | +52% | AWS AI infrastructure |
| Alphabet | $175–185B | $91B | +96% | Gemini, Vertex AI, Cloud |
| Microsoft | ~$145B | ~$100B | +45% | Azure AI, OpenAI |
| Meta | $115–135B | $72B | +75% | Llama models, AI ads |
| Oracle | ~$50B | ~$30B | +67% | OCI AI cloud |
"The AI spending race has entered a third year with no obvious slowdown. This is about controlling the stack, securing distribution, and turning AI into a durable competitive advantage."
The Alarming Math: Can Returns Justify This?
That is the multi-trillion-dollar question. None of the Big Four can yet show a neat dollar-for-dollar payback on their AI investments, though all are already monetizing AI through faster cloud growth, higher ad yields, and new software subscriptions.
- Microsoft and Alphabet are furthest along in turning AI spend into visible profit
- Meta and Amazon are leaning more on engagement and growth metrics as proof points
The depreciation math is particularly alarming. The five largest hyperscalers had plans to add roughly $2 trillion of AI-related assets to their balance sheets by 2030. Given that AI assets depreciate at around 20% per year, that implies an annual depreciation expense of $400 billion — more than their combined profits in 2025.
Meanwhile, analysts at Barclays project Meta's free cash flow could drop by almost 90% this year alone.
The Competitive Logic: Big Tech companies have issued over $100 billion in bonds so far in 2026 to fund their AI capex. Investors demanded record protection against defaults via Credit Default Swaps. The spending is not irrational — it is the result of a classic coordination problem: any company that pulls back risks falling permanently behind. So everyone keeps spending, and the arms race accelerates.
Who Is Actually Winning?
The picture is nuanced.
Google generates most of its revenue from advertising and is betting that AI-enhanced search and Gemini-powered products will keep its ad business dominant.
Microsoft is furthest along in enterprise AI monetization, with Copilot products integrated across Office 365, GitHub, and Azure.
Amazon is playing a longer game, betting that AI will become the primary workload for cloud computing — and that whoever owns the best AI infrastructure wins the cloud war.
Meta, uniquely, is not a cloud company: it is betting that AI will make its advertising platform dramatically more effective, and that open-source Llama models will attract the developer ecosystem that eventually powers its next platform.
The winner of this arms race will not be declared in 2026. But the foundations being laid this year — the data centers, the chips, the networking — will determine who controls AI infrastructure for the next decade. That is why $700 billion is, somehow, not enough for these companies.
The Bottom Line
In the AI infrastructure race, falling behind is not an option any of these companies is willing to consider. Whoever controls the largest, most capable AI infrastructure will control the next generation of cloud customers, enterprise software contracts, and consumer AI applications.
The rewards for winning are enormous. The penalty for losing may be existential. That calculus is driving $700 billion of capital expenditure in a single year — and it is why, heading into year four, no one is hitting the brakes.
This report is based on Q1 2026 earnings calls, company guidance, and analyst research published in April 2026. Capex figures represent full-year 2026 guidance or annualized run rates where full-year guidance has not been issued.
