In June 2026, SpaceX revised its prospectus before going public.

The revisions don't concern rocket technology, satellite internet, or Mars colonization plans. The new risk warning points to something more fundamental—water. The document states that water shortages, droughts, local water competition, or regulatory restrictions could prevent companies from obtaining sufficient cooling water, thus slowing down data center expansion and even forcing them to adopt more expensive alternative cooling solutions. Electricity, processors, and water resources are listed side-by-side in the document, constituting the core resource constraints for expanding AI computing power.

This marks the first time SpaceX has systematically emphasized water resource risks in a public filing. A company known for its Starship and Falcon rockets is reminding potential shareholders to pay attention to the stability of water taps.

The original risk warning read: "Water scarcity, drought conditions, competition for local water resources, or regulatory restrictions on water use could limit our ability to obtain sufficient water for cooling... delay or limit expansion... or require us to implement alternative cooling techniques that may be more costly." The wording is plain, typical of restrained legal writing. But the fact that it appeared in an IPO filing is itself a signal.

SpaceX's core AI business is xAI. According to a TechCrunch report in May, xAI is projected to incur $6.4 billion in operating losses by 2025, with annual revenue of $3.2 billion and continuously soaring capital expenditures. This rapid cash burn corresponds to the frenzied expansion of data centers, servers, and computing clusters. When billions of dollars in hardware and infrastructure investment become the norm each quarter, fluctuations in the supply of any physical resources will no longer be merely a cost item that operations departments need to optimize; it will become a risk that must be accounted for to investors.

The shift of water from operating costs to risk disclosure framework is noteworthy.

Operating costs are something a company can control; it can save money or switch to a different technology. But risks are different. Risks are external variables, things a company may not be able to control. Drought is a weather problem, local government tightening water permits is a policy problem, and community opposition is a political problem. These are problems that no amount of money a company can necessarily solve.

TechCrunch's report points out that this revision reflects renewed regulatory and investment concerns about the AI ​​industry's reliance on natural resources. Analyzing this assessment requires first answering a more fundamental question: How much water do AI data centers actually use?

17 billion gallons, that's just the portion that was directly cooled.

Lawrence Berkeley National Laboratory has provided a set of estimates. In 2023, the water consumption for direct cooling of U.S. data centers was approximately 17 billion gallons, or about 64 billion liters. This only accounts for direct cooling. Data center operations rely on electricity, and the power generation process itself consumes a significant amount of water. The combined indirect water consumption—cooling of thermal and nuclear power plants, and evaporation from hydroelectric power generation—is estimated at a staggering 211 billion gallons.

17 billion cubic meters of water was used directly, and 211 billion cubic meters was used indirectly. The latter figure is more than 12 times that of the former. When discussing the water footprint of AI, the direct cooling data we see is only the tip of the iceberg.

This estimate also predicts a trend: by 2028, direct cooling water usage in US data centers could increase two to four times. The reason for this wide range is that it depends on the rate of expansion of AI computing power, the choice of cooling technology, and the location distribution of new data centers. A doubling is the most conservative scenario, while a fourfold increase represents aggressive expansion. Regardless of the scenario, the trend is steeply upward.

These numbers are abstract in themselves. When you look at specific companies, the quantifiable feeling becomes clearer.

Google disclosed in its sustainability report that it consumed 6.4 billion gallons of water in 2023, 95% of which was used in its data centers. This means that Google's data centers alone consumed approximately 6 billion gallons of water that year. One of its sites, the Council Bluffs data center in Iowa, consumed an additional 1 billion gallons of drinking water in 2024.

Meta's data is slightly smaller, but still considerable. In 2023, Meta consumed 813 million gallons of water globally, 95% of which was in data centers.

Putting these figures together, the water consumption of a single Google data center is roughly equivalent to more than a third of the direct cooling water consumption of all data centers in the United States, as estimated by Lawrence Berkeley Lab. The Council Bluffs site in Iowa alone could support a medium-sized city's annual water consumption.

Where did all this water go?

Most large data centers use evaporative cooling technology. The principle is simple: water comes into contact with hot air in the cooling tower, evaporates, takes away heat, and turns into water vapor, which is then released into the atmosphere. This process is called "consumable water use." The water is used and does not return to rivers, lakes, or groundwater aquifers. This is different from residential water use, where bathwater or dishwater can be treated and returned. Data center cooling towers discharge steam. It is consumed, literally.

A 2021 study published in *Nature* journal *npj Clean Water* provided a technological scale: a typical 1-megawatt IT load data center, using traditional evaporative cooling technology, consumes approximately 25.5 million liters of water annually. One megawatt of IT load roughly corresponds to the computing power of several hundred servers. Large data centers often reach tens or hundreds of megawatts. Multiplying this by a factor of two, a 50-megawatt data center cluster could easily consume over 1 billion liters of water annually for cooling.

In arid regions, the implications of this consumption level are self-evident.

Building a water tiger on the edge of the desert

In April 2025, an investigative report by the British newspaper The Guardian revealed that Amazon, Microsoft, and Google were operating and expanding data centers in some of the world's driest regions, with the combined size of the three companies' data centers projected to expand by 78%. Behind these figures lies a series of ongoing conflicts.

In Querétaro state, central Mexico, 17 out of 18 municipalities are suffering from severe drought. Meanwhile, the state has become home to a cluster of data centers for international tech giants. Local residents held signs outside the data centers: "No queremos centros de datos, queremos agua"—we don't want data centers, we want water. The BBC provided detailed coverage of this conflict.

In Mesa, Arizona, according to a Business Insider report in June 2025, Meta reached a water agreement allowing its facilities to use up to 4 million gallons of water per day. What does 4 million gallons mean? Based on the average daily water consumption of approximately 82 gallons for a U.S. resident, this is equivalent to the daily water consumption of nearly 49,000 people. Arizona is one of the most water-stressed states in the U.S., with the Colorado River's water level declining year after year, leading to ongoing disputes among states over water allocation quotas. While a data center drawing 4 million gallons of water daily may be legal and compliant, it is not without controversy.

Similar concerns have emerged in Australia. A December 2025 report in The Guardian Australia pointed out that with the accelerated construction of large-scale data centers, drinking water supplies in some areas are facing direct competition. In developed countries with relatively mature water resource planning systems, the impact of a sudden surge in data center water consumption is equally severe, indicating that this is not an isolated case due to the governance level of a particular region, but rather a widespread contradiction between scale expansion and limited resources.

The common thread in these controversies isn't that tech companies are "illegally using water." They haven't violated any regulations. Every water agreement has been legally approved, and every water fee has been paid on time. The root of the problem lies in the fact that the existing water allocation framework was established before data centers became major water consumers. When a single data center's daily water consumption equals that of an entire town, compliance itself becomes an issue. The system hasn't kept pace with the growth of these water hogs.

A Guardian investigative report in October 2025 revealed another dimension. Amazon's long-standing refusal to disclose detailed water usage for its data centers has been accused of strategically concealing its complete water footprint. Google discloses data for individual sites, Meta releases aggregated global data, while Amazon provides the least information. This divergence in disclosure is beginning to be seen by analysts as a variable in risk assessment. The less a company is willing to tell you how much water it uses, the more likely its water consumption is to spark controversy.

The project was halted, water being the reason.

The water resource dispute has moved beyond mere public discourse; it is now substantially hindering projects from being implemented.

An industry report by Data Center Watch reveals that in the two years since mid-2024, approximately $64 billion worth of data center projects in the United States have been blocked or delayed due to local community opposition. Water consumption is one of the core reasons for protest, alongside electricity usage and noise pollution. The report documents 142 bipartisan grassroots opposition groups across different states and with diverse political backgrounds, but they have reached a rare consensus on opposing mega-data center projects.

Water is becoming a new weapon in the NIMBY (Not In My Backyard) movement. In the past, NIMBY movements mainly revolved around substations, waste treatment plants, and highways. Now, data centers have joined the list. The reasons have changed, but the logic remains the same. Residents' logic is simple: you say your data center contributes to the economy, but if the cost is lower water pressure, higher water bills, and a drop in well water levels, I won't accept that cost.

Once this kind of opposition takes hold, it cannot be resolved by a few community presentations or promises of jobs from businesses. Electricity can be supplied by building new power plants, fiber optic cables can be laid, and land can be acquired at a premium. But water has no substitute in the eyes of residents. And with no substitutes, there is very little room for negotiation.

According to industry statistics, in 2025, approximately half of the data center projects originally planned for launch in 2026 were canceled or delayed. This proportion is enough to make any company planning to expand its AI infrastructure re-evaluate its site selection logic. Previously, the ranking of data center sites was: electricity, fiber optics, land price, climate. Now, water is gaining weight.

In February 2026, the University of California, Berkeley's Center for Law and Energy released a dedicated report exploring how to regulate water use for data centers in California. This marked the first time academia had addressed this issue directly in a dedicated report. The report's release itself is an indicator: when top law schools and energy policy think tanks begin systematically studying regulatory frameworks for data center water use, it signifies that this issue has moved beyond internal industry discussions and entered the public policy agenda.

Investors are starting to do the math.

The capital market is also following suit.

In April 2026, the Journal Record reported that investors formally urged Amazon, Microsoft, and Google to disclose more data on their data center water consumption. The report also cited macroeconomic data: North American data centers had already used nearly one trillion liters of water in 2025.

One trillion liters is a number that's difficult to grasp intuitively. To put it another way: it's roughly equivalent to the volume of a large freshwater lake. Lawrence Berkeley Lab's 2023 estimate was already quite large, but considering actual consumption in 2025, it may be considered conservative.

The shift in the investment community's attitude is not without precedent. Previously, water resources appeared in ESG reports alongside other environmental indicators, primarily as part of a form filled out by corporate social responsibility departments. Now, things are different. Water resources have jumped from the "corporate image" section to the "operational risk" section. Shareholders are no longer concerned with environmental protection; they are concerned with whether there is enough water to keep the servers running. When the stability of water supply begins to affect revenue expectations, it is no longer an ESG topic, but a financial one.

The response strategies of different companies have diverged significantly. Google continues to release water usage data for individual sites and claimed in 2024 to have returned 4.5 billion gallons of water through a water replenishment program. Meta releases aggregated data. Amazon, despite the Guardian investigation, has not yet disclosed detailed water usage figures for each site. This divergence further reinforces the view that transparency of water usage data is itself becoming a variable for analysts assessing the risk exposure of AI infrastructure companies.

Enterprises are also exploring technological solutions. Switching to air cooling can reduce direct water consumption, but often increases power consumption. Liquid cooling technology can use water at higher temperatures (NVIDIA's Vera Rubin platform supports 45°C water cooling), but system deployment costs are higher. Each technological approach involves a trade-off between water and power consumption; there is no perfect, universal solution. Ultimately, what determines the cooling solution for a data center may not be the technologically optimal solution, but rather local water and electricity prices and policy tolerance. Technological choices have become compromises under resource constraints.

An ironic contrast

In March 2026, OpenAI CEO Sam Altman made a widely circulated statement during a public speech. According to Business Insider, he said, "We see a future where intelligence is like electricity or water, a utility that people buy from us on a meter."

This statement sparked considerable discussion regarding copyright and business models, but its deeper meaning is more relatable. While Altman likened AI to water and electricity, the actual physical operation of AI is consuming real water in the real world. The industry's vision for a business model is to package AI as an inexhaustible infrastructure, billed as you go, like turning on a tap. Meanwhile, SpaceX's prospectus is frankly admitting: without enough water, AI might not even be able to run.

Before a service is compared to water and electricity, its infrastructure already incurs huge bills for those utilities. This analogy itself is the most accurate description of the current state of understanding of the AI ​​industry in 2026.

Looking back at the timeline, the narrative path is very clear.

From 2023 to 2024, the annual water consumption data of leading global cloud vendors were gradually made public through sustainability reports. Lawrence Berkeley National Laboratory released estimates, providing the first macro-level picture of data center water consumption across the United States. Community conflicts in Querétaro, Mexico, and Mesa, Arizona, began to enter the mainstream media spotlight.

In 2025, The Guardian and the BBC, among other organizations, conducted systematic investigative reports, bringing the link between data center expansion in arid regions and local water pressures into the public discourse. Data Center Watch released quantitative statistics on the stalled $64 billion projects. Investors began to formally demand greater transparency regarding water footprints.

In 2026, SpaceX removed this issue from public discussion and industry reports and placed it in the "Risk Factors" section of its IPO prospectus. This marked the formal transformation of the water problem from a public opinion issue to a factor in investment pricing. Before an investor subscribes to SpaceX stock, they need to sign a confirmation that they are aware that the company's AI business may encounter problems due to water shortages.

This is how the capital market prices resource constraints. It doesn't care about sentiment, corporate promises, or sustainable development visions in press releases. It only cares about one thing: what, and under what circumstances, will cause expected returns to fall short. Water supply is affected by weather, water prices by policy, and water access by community opposition—these are three things no company can control. What cannot be controlled is risk. Risk needs to be documented and communicated to investors.

This mechanism itself is reshaping the logic of AI infrastructure expansion.

For the past few years, the main narrative of the AI ​​race has been an arms race in computing power. Chips, electricity, and talent were the three key elements. Water was an implicit condition, assumed to be available locally. Now, this assumed premise is being shaken. In arid regions, in cities where local water is already scarce, and in areas where regulations are tightening water quotas, "available local water" is no longer an automatically valid assumption.

The expansion of AI infrastructure is no longer just a game of technology and capital. It is entering a phase where resource allocation needs to be negotiated simultaneously with four groups: local residents, local governments, regulatory agencies, and investors. The speed of the computing power race may not be determined by the fastest company, but by the slowest one.