Big Tech Will Burn $725 Billion on AI in 2026 — And 69 US Cities Just Said No

AI infrastructure spending 2026 has reached an unprecedented scale. Microsoft, Google, Amazon, and Meta are about to spend more money on AI infrastructure in a single year than the entire GDP of Sweden. The combined capital expenditure of these four tech giants is projected to hit $725 billion in 2026 — a staggering 77% increase over 2025’s already record-breaking $410 billion. And roughly 75% of that — over $545 billion — is going directly into AI-specific infrastructure: GPUs, data centers, cooling systems, and power generation.

But here’s the twist nobody’s talking about enough: while Silicon Valley races to build the most expensive infrastructure project in human history, communities across America are slamming the door shut. As of May 2026, 69 US jurisdictions have blocked new AI data center construction, with four implementing permanent bans. Town council members who approved these projects are being voted out. Residents are organizing mass protests. And the backlash is accelerating faster than the builds themselves.

AI Infrastructure Spending 2026: The Numbers Are Incomprehensible

Let’s break down what each company is committing to spend in 2026. Amazon leads the pack with approximately $200 billion in projected capital expenditure. Microsoft follows closely at $190 billion, with Alphabet (Google’s parent) doubling its spending to a jaw-dropping $180 billion. Meta rounds out the group with $115-135 billion earmarked for AI infrastructure.

To put these numbers in context: the entire Apollo space program cost about $257 billion in today’s dollars. Big Tech is spending nearly three times that amount — in a single calendar year. Goldman Sachs’ baseline model actually puts the total even higher, at $765 billion when including Oracle and other hyperscalers.

And this isn’t the ceiling. CNBC reports that Big Tech capital expenditures are now projected to top $1 trillion in 2027. We’re witnessing the most aggressive infrastructure buildout since the transcontinental railroad — except it’s happening in about 36 months instead of six years.

Where Is All This Money Actually Going?

The spending falls into several massive categories. First, there’s the hardware: Nvidia’s H100 and B200 GPUs aren’t cheap, and the demand has created a global semiconductor shortage that’s driving prices through the roof. Microsoft alone attributed $25 billion of its 2026 budget specifically to increased memory and chip costs. DRAM prices have surged from $3.76 per gigabyte in 2025 to $9.71 per gigabyte in 2026 — a 158% increase that’s eating into every company’s bottom line.

Then there’s the physical infrastructure. Data center IT capacity under construction now tops 23 gigawatts globally. Goldman Sachs Research predicts total data center demand will rise by roughly 50% to 92 gigawatts by 2027. Each large AI data center costs between $500 million and $1 billion+ depending on size, power capacity, and computing hardware. These aren’t warehouses with servers anymore — they’re industrial-scale facilities requiring dedicated power plants, advanced cooling systems, and massive water supplies.

The energy requirements alone are reshaping the US power grid. The Department of Energy projects that data centers will consume 12% of US electricity by 2028, up from just 4% in 2023. That’s why tech companies are cutting nuclear power deals: Microsoft restarted Three Mile Island at a cost of $1.6 billion (837 MW), Amazon signed a Susquehanna agreement, and Google contracted with Kairos Power for small modular reactors. US utilities are planning $1.4 trillion in infrastructure upgrades specifically to support AI data center demand.

The ROI Question Nobody Can Answer

Here’s where things get uncomfortable for investors. Capital expenditure is expanding far faster than revenues. According to financial analysts, there’s a 46% growth gap between investment and sales — a divergence that actually exceeds the 32% gap observed during the 2001 telecom excess cycle. For those who don’t remember, that cycle ended with hundreds of billions in stranded assets and some of the largest corporate bankruptcies in history.

The financial strain is already visible. Amazon is looking at negative free cash flow of $17-28 billion in 2026, depending on which analyst you ask (Morgan Stanley says $17B, Bank of America says $28B). Hyperscalers collectively raised $108 billion in debt during 2025 alone, with projections suggesting $1.5 trillion in total debt issuance over the coming years to fund this buildout.

Bulls point to the roughly $2 trillion backlog in cloud contracts as evidence that demand justifies the spend. Jefferies analysts argue that ROI is self-evident given accelerating cloud growth. But bears counter that these backlogs don’t guarantee profitability — they just guarantee that the spending will continue regardless of returns. US technology and AI equity valuations sit near historical extremes at roughly 25x EV/EBITDA, exceeding even the telecom valuations that preceded the 2000 dot-com crash.

69 US Cities Say: Not in Our Backyard

While Wall Street debates ROI, ordinary Americans are fighting a ground war against AI data center expansion. The numbers tell a dramatic story: in May 2025, just eight moratoriums existed across the entire country. One year later, 78 measures have been enacted across the nation, including 50 active bans and four permanent prohibitions on new data center construction.

Maine is on track to become the first state to implement a statewide data center construction moratorium, pausing all new projects until November 2027. Virginia — home to “Data Center Alley” in Loudoun County, which houses the largest concentration of data centers on Earth — is seeing organized resistance from residents in Loudoun and Prince William counties who cite noise pollution, visual blight, and enormous water consumption for cooling systems.

The opposition isn’t just NIMBYism. Communities are facing real economic consequences. Wholesale electricity prices have surged by up to 267% over the past five years, with upgrade costs being passed directly to consumers. In New York, more than 350 residents packed a hearing about a proposed data center moratorium in Lysander — so many showed up that attendees had to wait outside. Town council members who approved data center projects have been resigning or getting voted out as constituents revolt against rising utility bills and environmental degradation.

Critics make a compelling case: these facilities create relatively few permanent jobs while consuming enormous quantities of electricity and water. A single large data center can use as much water as a small city for cooling. And unlike manufacturing plants or office buildings, data centers generate minimal tax revenue relative to their resource consumption — many jurisdictions offered generous tax incentives to attract them, only to discover the costs far exceeded the benefits.

The Federal vs. Local Showdown

This creates a fascinating political collision. The federal government, under both parties, has pushed aggressively to maintain US AI leadership — and that requires data centers. But state and local governments, responding to constituent pressure, are increasingly blocking the very infrastructure that federal policy demands. According to MultiState’s 2026 tracker, the conflict between federal AI ambitions and state-level resistance is intensifying with no clear resolution in sight.

The political dynamics are splitting in unpredictable ways. Some red states are competing aggressively for data center investment by offering massive tax incentives. Some blue states are imposing moratoriums. But the opposition doesn’t break cleanly along partisan lines — it’s driven primarily by local impacts, particularly energy costs and environmental concerns, that affect voters regardless of political affiliation.

Is This the Telecom Bubble 2.0?

The comparisons to the late-1990s telecom infrastructure boom are impossible to ignore. During that era, companies laid millions of miles of fiber optic cable based on exponential growth projections. When demand failed to materialize at projected rates, the result was catastrophic: WorldCom’s $11 billion fraud, Global Crossing’s bankruptcy, and hundreds of billions in destroyed shareholder value. The capex-to-revenue gap today (46%) is actually worse than the telecom bubble’s peak divergence (32%).

But there are important differences. AI is already generating substantial revenue — unlike the telecom buildout, where revenue projections were largely theoretical. Cloud computing revenues are genuinely accelerating. Microsoft’s Azure AI revenue is growing at over 40% year-over-year. Google Cloud, Amazon Web Services, and Meta’s advertising business all show tangible AI-driven revenue growth.

The question isn’t whether AI will generate returns — it will. The question is whether the returns will justify this level of spending. And increasingly, the answer from Wall Street is nuanced: investors are selectively rewarding companies with clear AI revenue paths while punishing those where spending appears speculative. The AI trade is no longer a rising tide lifting all boats — it’s becoming a stock-picker’s market where execution matters more than ambition.

What Happens If the Music Stops?

The construction boom is creating its own dependencies. Hundreds of thousands of construction workers, engineers, and contractors are now employed in AI data center buildouts. Supply chains have been reconfigured around this demand. If spending slows significantly — whether due to recession, regulatory pressure, or simply a reassessment of ROI — the economic ripple effects would extend far beyond Silicon Valley.

There’s also a geopolitical dimension. China, the EU, and the Middle East are all building competing AI infrastructure. If US companies pull back while international competitors accelerate, the strategic implications for AI leadership could be lasting. This is partly why the federal government is reluctant to do anything that might slow the buildout — even as communities push back.

The most likely outcome? The spending continues at roughly current levels through 2027, with increasing geographic concentration in jurisdictions that welcome data centers (primarily in the South and Southwest) and growing political friction in regions that don’t. The community backlash will redirect investment rather than stop it. And the ROI question will be answered — just not as quickly as either bulls or bears hope.

The Bottom Line

We’re living through the largest infrastructure investment in human history, and it’s happening at a pace that’s simultaneously thrilling and terrifying. $725 billion in a single year. 69 cities saying no. A capex-to-revenue gap worse than the telecom bubble. Nuclear power plant restarts. Community revolts. And the spending is projected to exceed $1 trillion next year.

Whether this ends as the greatest infrastructure investment ever made or as the most expensive lesson in corporate excess depends entirely on one thing: whether AI revenue growth can catch up to AI infrastructure spending before investors and communities lose patience. Right now, it’s a race — and the finish line keeps moving.

Related reading: Cerebras IPO: $3.5B Raised at $26.6B Valuation | OpenAI Hit $25B Revenue but Its IPO Isn’t Ready | Pentagon AI Deals 2026: 8 Tech Giants In, Anthropic Out | Google Gemini 3.1 Pro & Gemma 4 Launch

Related Reading

If you found this article interesting, check out these related stories: OpenAI’s $25B revenue trajectory, Nvidia’s $40B AI investment portfolio, 76% of companies now have Chief AI Officers. Also worth reading: Cloudflare’s 1,100 AI-driven layoffs and Cerebras challenging Nvidia with its IPO.