# The Hidden Crisis Behind AI’s Growth: Why Water Could Become More Valuable Than Compute
Something unusual appeared in SpaceX’s amended IPO filing recently. Alongside standard financial disclosures and risk factors about rocket failures and satellite debris, the company warned investors about a threat that has nothing to do with rockets, satellites, or Starlink terminals: water scarcity.
Specifically, SpaceX flagged water access as a critical risk for operating and expanding large-scale AI infrastructure. This isn’t a minor footnote in a document about space technology. It’s a warning that the AI industry’s growth could soon be constrained by something far more fundamental than chip availability or electricity supply.
## The Unglamorous Reality of Data Centers
Every time you interact with an AI system—asking ChatGPT a question, having Claude help draft an email, running an image generation through Midjourney—you’re depending on vast data centers humming with servers. These facilities generate enormous heat. All that heat needs to go somewhere.
Traditional air cooling works for smaller deployments. But for hyperscale data centers running thousands of servers, liquid cooling has become essential. And liquid cooling, it turns out, requires significant amounts of water.
Microsoft’s latest data centers use evaporative cooling systems that consume millions of gallons annually. Google’s facilities in particularly water-stressed regions have faced criticism for their consumption. Amazon, Meta, and other major cloud providers operate similar systems at scale.
The math gets alarming quickly. A typical data center might use 1-5 million gallons of water per day. Large hyperscale facilities can exceed 20 million gallons daily. For context, an average American household uses about 300 gallons per day. So one large data center consumes as much water as 65,000 homes.
## Why SpaceX’s Warning Matters
SpaceX isn’t primarily an AI company. They build rockets and satellites. Their Starlink division competes with traditional internet providers, not OpenAI. So why is SpaceX warning about AI water consumption?
The answer lies in a peculiar intersection of industries. SpaceX needs computing infrastructure to support their operations—everything from satellite constellation management to launch trajectory calculations. They’re expanding into AI-adjacent applications. And they’re planning massive facility expansions.
But beyond their specific needs, SpaceX’s warning reflects something broader: even companies tangentially connected to AI infrastructure are recognizing the resource constraints. When a rocket company starts worrying about water policy, you know the issue has entered mainstream risk assessment.
In their IPO filing amendment, SpaceX cited several specific concerns:
– Droughts reducing water availability in key regions
– Local competition for water resources from agriculture, municipalities, and industry
– Regulatory restrictions on water usage
– Potential operational constraints at existing facilities
This isn’t theoretical risk management. These are real constraints already affecting decisions about where to build new facilities.
## The Geography Problem
AI data centers aren’t distributed evenly across the globe. They concentrate in regions with cheap electricity, favorable regulations, and existing fiber infrastructure. Many of these same regions face significant water stress.
Northern Virginia hosts more data center capacity than anywhere else in the world. It also happens to be experiencing increasing water competition. The area’s aquifers and surface water sources face demands from residential growth, agriculture, and industry. Data centers are a growing piece of that demand puzzle.
Similar tensions exist in Arizona, where both Google and Apple have major facilities. The desert Southwest has always been water-challenged, but climate change is intensifying those challenges. Summer temperatures break records with disturbing regularity, making evaporative cooling both more necessary and less effective.
Texas, another data center hotspot, has dealt with winter storms that disrupted water infrastructure and summer droughts that strained supplies. The state’s deregulated energy market occasionally produces crises; its water situation could become similarly volatile.
The implication is uncomfortable: the regions best suited for data center expansion often have the least water security. This creates a fundamental tension between optimal location and sustainable operations.
## The Carbon Calculus Gets Complicated
Here’s the irony that makes this issue so challenging: water and electricity aren’t independent variables. Some cooling systems use more water but less electricity. Others minimize water consumption but require more power for mechanical cooling.
Evaporative cooling is highly energy-efficient but water-intensive. Compressor-based cooling minimizes water usage but consumes more electricity. The choices interact in complex ways that vary by climate, season, and local resource availability.
As regions heat up due to climate change, the efficiency of data center cooling systems decreases. Air-cooled systems work less well when ambient temperatures rise. Evaporative systems need more water when humidity drops. This creates a feedback loop: climate change makes cooling harder, requiring more resources, while those resource demands contribute to climate change.
The mathematical relationship is unforgiving. A data center that was optimally sized for 2020 climate conditions might be undersized or over-consuming by 2030. Facilities being designed today need to account for climate projections decades into the future.
Some companies are responding by building facilities in historically cool locations—Scandinavia, Canada, New Zealand—where cooling is easier and water is more abundant. But these locations introduce latency challenges for users, and they create their own supply chain complications.
## Industry Responses and Innovations
Major tech companies aren’t ignoring these concerns. They’re investing in various approaches to address them, though none offers a complete solution.
Microsoft has explored underwater data center concepts that use ocean water for cooling. The Project Natick experiment demonstrated that submerging servers was technically feasible, though the approach faces obvious practical limitations for widespread deployment.
Google has committed to replenishing more water than they consume through various restoration projects. The company funds river restoration, wetland creation, and groundwater recharge initiatives. Critics note this doesn’t address local consumption impacts—the replenishment might occur in different watersheds than the usage.
Amazon has announced similar commitments and is investing in water-efficient cooling technologies. But their data center fleet keeps growing faster than efficiency improvements can offset.
Several startups are developing more efficient cooling technologies. Immersion cooling—submerging servers in dielectric fluids—can dramatically reduce water consumption. These systems use closed-loop cooling where water never touches the servers directly. However, they have higher capital costs and introduce new maintenance requirements.
Some companies are building facilities in locations chosen specifically for water availability, even if electricity costs are higher. Iceland, with its abundant geothermal energy and glacial water, has attracted data center investments partly for these reasons. Northern Scotland, Norway, and Sweden offer similar advantages.
## The Regulatory Landscape Shifts
Water rights regulations vary dramatically by jurisdiction. Some regions treat water as a public resource subject to government allocation. Others have robust private water rights markets where water can be bought and sold like any other commodity.
AI data center operators are discovering that securing water access can be as challenging as securing power or land. In some jurisdictions, industrial water rights require lengthy permitting processes or face political opposition.
California recently enacted restrictions on data center water usage in certain watersheds. The state has dealt with severe droughts and isn’t eager to approve large new water consumers without clear mitigation plans. Other Western states are considering similar measures.
The federal government has begun studying data center water consumption patterns and potential regulatory responses. No comprehensive federal framework exists yet, but that could change if consumption continues growing.
Dr. Sarah Kondolf, a water resource expert at UC Berkeley, noted: “We’re seeing the emergence of water as a critical infrastructure constraint in a way we haven’t experienced since the early industrial era. The difference is that water constraints today affect digital infrastructure, not just manufacturing.”
The regulatory uncertainty adds another dimension to data center planning. Companies must now consider not just current regulations but potential future restrictions when deciding where to invest.
## The Intersection With AI’s Resource Appetite
The timing of this emerging crisis is particularly interesting. AI workloads have dramatically higher resource demands than traditional computing tasks.
Training a large language model requires exponentially more computing power than running inference queries. The energy and water consumption scales with model size and usage volume. As AI adoption grows, these demands multiply.
Consider the trajectory. GPT-3 training consumed an estimated 1,287 megawatt-hours of electricity. GPT-4 was significantly larger. Systems now in development are larger still. The trend line points toward ever-greater resource consumption.
Some researchers are working on more efficient model architectures that accomplish more with less computation. Mixture-of-experts approaches, pruning techniques, and quantization all offer improvements. But these gains are being overwhelmed by the growth in model sizes and usage volumes.
The industry is in a race between efficiency improvements and demand growth. So far, demand is winning.
## What This Means For the AI Industry
The water constraint adds a new dimension to AI industry planning. Where companies build, how they design facilities, and what technologies they adopt will increasingly reflect water considerations.
This might accelerate geographic diversification of AI infrastructure. Companies might build more smaller facilities in water-secure regions rather than massive campuses in water-stressed areas. The hyperscale model—giant facilities serving huge geographic areas—may give way to more distributed architectures.
It could also drive investment in alternative cooling technologies. While immersion cooling and other approaches have higher upfront costs, those costs might become more acceptable when the alternative is inability to expand operations due to water restrictions.
The relationship between AI companies and local communities may also evolve. Data center operators have traditionally negotiated for electricity rate concessions and regulatory flexibility. Water agreements may become similarly contentious.
Companies that secure water rights early might gain competitive advantages. Those that don’t might find themselves constrained when they want to expand.
## The Bigger Picture
Water is weirdly easy to take for granted. It falls from the sky (in most places). It comes out of taps reliably. Until it doesn’t.
We think of digital technology as dematerialized, ethereal, weightless. But every AI interaction has a physical footprint. Every query, every generation, every conversation requires energy, water, chips manufactured from rare earths extracted through environmentally disruptive processes, and cooling systems that consume resources.
This doesn’t mean AI is bad or that we should use it less. It means we should be clearer-eyed about what sustainable technology development actually requires.
SpaceX’s warning in an IPO filing might seem like a small thing. But it’s a sign that the industry’s leaders are beginning to internalize an uncomfortable truth: the AI boom has physical limits, and water might be the constraint we least expected to face.
The future of AI isn’t just about algorithms and compute. It’s also about whether we can find enough water to keep the servers cool.
—
## Sources
– [IMFounder – 7 Explosive Tech News Stories That Could Change the Future of Technology](https://imfounder.com/science-tech/explosive-tech-news-june-2026-ai-lawsuits-ipos-cyberattacks/)
– [UC Berkeley Water Resources Group](https://nature.berkeley.edu/groups/kondolf)
– [IEA – Data Centres and Data Transmission Networks](https://www.iea.org/reports/data-centres-and-data-transmission-networks)
– [Nature – The Water Footprint of Data Centers](https://www.nature.com/articles/d41586-023-00069-6)
– [Google Environmental Report – Water](https://sustainability.google/reports/water/)
– [Microsoft Project Natick](https://azure.microsoft.com/en-us/solutions/sustainability/project-natick/)