AI's $7.6 Trillion Buildout Has an Environmental Bill Nobody Budgeted For

2026-06-20 · Nia

There are two numbers that define the AI industry right now, and they're on a collision course.

The first: Goldman Sachs projects $7.6 trillion in cumulative AI capital expenditure between 2026 and 2031. Annual spending is expected to more than double, from $765 billion this year to $1.6 trillion by 2031. Hyperscaler capex alone could hit $1.1 trillion by 2027.

The second: A United Nations University study published on June 3 found that by 2030, the water consumed by AI infrastructure will equal the basic domestic water needs of 1.3 billion people in Sub-Saharan Africa. Data centers powering AI will consume 945 terawatt-hours of electricity annually — nearly triple the combined electricity use of Pakistan, Bangladesh, and Nigeria.

Read those two numbers side by side. One industry is preparing to spend more money than the GDP of Japan. The environmental cost of that spending will consume resources equivalent to what hundreds of millions of people need to survive.

This isn't a hypothetical tension. It's here.

The UN Report: Beyond Carbon

The UNU-INWEH report, titled "Environmental Cost of AI's Energy Use: Carbon, Water and Land Footprints," makes a crucial argument that most sustainability discussions around AI miss: carbon is only one part of the story.

Here's what rarely makes the headlines:

Water footprint. Mid-sized data centers consume up to 300,000 gallons of water daily for cooling. Larger facilities — the kind being built for frontier AI training — can use 5 million gallons per day, comparable to a small town's entire water supply. And here's the catch: switching to renewable energy sources to reduce carbon emissions can actually increase water consumption, because many renewables require water-intensive cooling or production processes.

Land footprint. AI-related infrastructure — data centers, power plants, transmission lines, supply chains — could consume over 14,500 square kilometers by 2030. That's roughly twice the size of the Jakarta metropolitan area. This isn't abstract. It means real communities losing real land.

The justice dimension. The environmental burden concentrates geographically. Data centers are built where power is cheap and regulation is light. The communities hosting these facilities bear the water stress, land use changes, and noise pollution, while the benefits of AI diffuse globally. The UN report explicitly frames this as a governance and justice challenge, not just an engineering problem.

We've covered the broader environmental crisis building around AI infrastructure before — but the UN report makes the scale impossible to ignore.

Goldman's $7.6 Trillion and the Sustainability Blindspot

Goldman Sachs' projection of cumulative AI capex reaching $7.6 trillion from 2026 to 2031 comes with detailed analysis of compute needs, power capacity, and financing structures. They estimate big tech companies alone will spend $5.3 trillion on AI buildout from 2025 through 2030 — revised upward from their earlier estimate of $4.5 trillion.

What's conspicuously absent from these projections? Environmental cost accounting.

When Goldman talks about infrastructure, they mean chips, data centers, and power contracts. The water those data centers will consume, the land they'll occupy, the communities they'll displace — those appear nowhere in the investment thesis. They're "externalities." Which is corporate finance's way of saying "someone else's problem."

This isn't unique to Goldman. The AI infrastructure race we've been tracking has been almost entirely a story about compute supply and demand. Environmental constraints barely register in the boardroom conversations driving these decisions.

But they will. Because physics doesn't care about earnings calls.

Where Things Get Real: The Water Wars

Let me make this concrete.

Northern Virginia — the epicenter of US data center construction — has been fighting over water allocation for years. The region's data centers already consume so much electricity that Dominion Energy has had to delay coal plant retirements. Now add water. Each new hyperscale facility needs millions of gallons per day from a watershed that's also serving millions of people.

Similar conflicts are emerging in Arizona, Texas, Ireland, the Netherlands, and parts of Southeast Asia — anywhere data centers cluster. The pattern is consistent: tech companies move in, water and power prices rise, local communities push back until the political pressure becomes unmanageable.

This is the bottleneck nobody on Wall Street is modeling. You can finance $7.6 trillion in new infrastructure. You can manufacture enough chips. But you cannot manufacture water. And in a warming world, the places with cheap power and abundant water are a shrinking list.

The "Renewable" Illusion

Here's the part that frustrates me most.

Every major AI company claims renewable energy commitments. Microsoft, Google, Amazon, Meta — all have net-zero pledges and massive renewable energy purchases. But here's what the UN report exposes: "renewable" doesn't mean "sustainable."

Many renewable energy sources — hydroelectric, certain solar configurations, even some wind installations — have significant water and land requirements of their own. When a data center switches from natural gas to solar + battery storage, the carbon footprint drops. The water and land footprints can increase.

This isn't an argument against renewables. It's an argument against treating carbon as the only metric that matters. Real sustainability means accounting for all the resources an industry consumes, and right now, the AI industry's sustainability reporting is largely carbon-only.

Companies like OpenAI are even looking at unconventional energy sources — we covered OpenAI's fusion energy deal with Helion — which signals recognition that current energy infrastructure can't sustain the buildout. But fusion is years away from commercial viability, and today's data centers are being built today.

What Needs to Change

The UN report offers a framework based on seven principles: transparency, efficiency by design, equity, environmental justice, lifecycle responsibility, global cooperation, and sustainable use. That's sensible but insufficient without enforcement mechanisms.

Here's what I think actually has to happen:

Mandatory environmental impact disclosure. AI companies should be required to report water consumption, land use, and full lifecycle environmental costs — not just carbon. The EU's AI Act doesn't require this. It should. The SEC is circling similar requirements for US companies but hasn't pulled the trigger.

Water-aware compute pricing. Cloud providers should price compute partly based on the environmental cost at each data center location. If a facility is in a water-stressed region, the cost should reflect that. Market mechanisms work when they account for real costs.

Investment in cooling innovation. The AI industry needs to treat data center cooling as a frontier research problem, not an operational detail. Some companies are experimenting with liquid cooling, waste heat recapture, and even underwater data centers. This needs to be an industry-wide priority, not a collection of pilot projects.

Local community veto power. Communities affected by data center construction should have meaningful input — including the ability to block projects when water or land impacts are too severe. The current model, where tech companies negotiate directly with state governments and bypass local concerns, isn't sustainable politically or environmentally.

The Paradox of AI for Climate

There's a genuine irony here. AI is being positioned — legitimately — as a tool for addressing climate change. AI models optimize energy grids, predict weather patterns, accelerate materials science for better batteries and solar cells, and improve agricultural efficiency.

But the infrastructure required to run those models is itself becoming a significant environmental burden. The trillion-dollar AI spending boom creates its own climate problem while promising to solve everyone else's.

This isn't an argument against AI. It's an argument for being honest about the tradeoffs. The industry has collectively decided that the benefits justify the environmental costs. Maybe they do. But that calculation should be explicit, transparent, and subjected to democratic accountability — not buried in corporate sustainability reports that count carbon and ignore water.

The Bottom Line

Goldman's $7.6 trillion number is going to get a lot of attention from investors. The UN's water and land numbers should get just as much attention from everyone else.

We're building the most powerful technology in human history. We're doing it at unprecedented speed. And we're doing it without a clear-eyed accounting of what it costs — not in dollars, but in the water, land, and livable environment that every community on Earth depends on.

That's not a technology problem. It's a governance problem. And right now, governance is losing the race.