Water is an Issue, But Not Because of Data Centers or AI

The near-hysteria about water consumption needs to be kept in proper perspective. In the water-short American West, including the Colorado River watershed, water is always an issue. 

In terms of water access, the United States is effectively divided by the historic 100th meridian, which runs roughly through Texas, Oklahoma, Kansas, Nebraska, and the Dakotas. East of that line, rainfall is generally sufficient to support agriculture without irrigation. West of it, irrigation is necessary.

Region	Typical Annual Precipitation
Pacific Northwest mountains	60–150+ inches
Eastern U.S. (most areas)	30–60 inches
Midwest	25–45 inches
Great Plains	15–35 inches
Intermountain West (Nevada, Utah, Wyoming interior, western Colorado)	5–20 inches
Desert Southwest	3–15 inches

But precipitation alone does not illustrate the issue as well as water runoff, which is the amount of liquid that remains available for use after evaporation and plant transpiration. In much of the Intermountain West and Great Plains, most precipitation evaporates or is consumed by vegetation before reaching streams.

Region	Typical Annual Runoff
Appalachian region	15–40+ inches
Upper Midwest	5–15 inches
Great Plains	0.5–5 inches
Intermountain West basins	Less than 1–3 inches
Desert Southwest	Often less than 0.5 inch

Relative to demand, west of the 100th meridian, water is always going to be an issue. 

Region	Water Supply Relative to Demand
Northeast	Large surplus
Southeast	Large surplus
Great Lakes	Very large surplus
High Plains	Small surplus
Southwest	Deficit
Colorado River Basin	Deficit

So it might be inevitable that water footprint becomes an issue for data centers, even if relative water consumption is quite low. Of course, a total water footprint would include the cost of generating electricity. 

Still, industry uses relatively little water, compared to other sectors of the economy. 

source: Axios 

But an argument can be made that the easiest gains might come from increasing agriculture efficiency where it comes to water consumption. 

And even if controversial, the easiest market encouragement might include shifting our subsidies for agricultural water pricing, as difficult as that will be for many farmers always on the brink of survival. 

As always, rights and values are in tension. Most people might say they believe in supporting family farms, just as much as they might say they value water conservation. But the numbers are clear. Small gains in agriculture will produce more efficiency, faster, than small gains in consumption in other sectors. 

Sector	Share of Water Consumption (Typical Western Basin)
Agriculture	70–80%
Municipal	10–20%
Industry	5–10%

Indeed, water pricing discourages efficiency because the users of 70 percent to 80 percent of the water pay the lowest prices for consumption. Again, values are in conflict. We might value food production and small farms as much as we value drinking water and electricity. 

But there is an order of magnitude difference between agricultural water prices and all urban uses of water. And as with all commodities and goods, low prices encourage consumption; higher prices encourage efficiency. 

Tradable water rights might be a preferred solution, shifting supply towards demand without expropriating or destroying farming. Also, it might make sense to encourage water-intensive agriculture only in regions with lots of water, while discouraging it in regions that are water scarce. 

Again, this will be controversial. 

User Type	Typical Economic Value of Water
Alfalfa irrigation	$50–$300 per acre-foot
Corn irrigation	$100–$500 per acre-foot
Municipal supply	$1,000–$5,000+ per acre-foot
Industrial/high-value uses	Often much higher

In other words, does it make good sense to grow water-intensive rice, almonds or alfalfa in water-scarce regions?

Crop	Acre-Feet per Acre
Wheat	1–2
Corn	2–3
Alfalfa	3–6
Almonds	3–4
Rice	4–5

As if that were not complicated enough, we also must balance protection of wetlands, fisheries, recreation and food sourcing. 

Data center water consumption might be an issue, but a relatively small one, overall. How we use and price use of a scarce resource is really the bigger issue. 

SpaceX Acquisition of Cursor is About the Stack

SpaceX is acquiring Cursor’s parent company Anysphere for $60 billion, and the valuation might be more a matter of strategic value than traditional software as a service metrics.

In other words, critics will question the cost of the deal to SpaceX.

Cursor already had become the fastest-scaling enterprise SaaS company ever, growing from zero revenue to $2 billion annualized run rate in three years. 

If SpaceX acquires Cursor, the biggest advantages would be faster entry into AI coding, access to Cursor’s loyal developer base, and a stronger product layer for monetizing its compute infrastructure. 

For Cursor, competition from Claude Code seems key. But Cursor also gains from SpaceX compute infrastructure, as compute resources have been an issue in the past. 

source: MarketWatch 

It should therefore provide a better competitive position against Anthropic in enterprise AI.

Cursor is a fork of Microsoft’s Visual Studio Code, the most-widely-used code editor in the world, with AI capabilities integrated at every level of the development workflow. 

It autocompletes code, suggests changes across multiple files, runs tests, iterates on errors, and increasingly operates as an autonomous agent that can execute multi-step coding tasks with minimal human intervention. 

But Cursor has been dependent on third-party AI providers for the underlying intelligence. 

As those model developers increasingly launched competing coding products, Cursor risked being squeezed by companies that controlled both the models and the computing infrastructure. 

For SpaceX, the deal adds a fast-growing software business to an AI platform built around xAI and the company's Colossus data center infrastructure.

The deal would also help SpaceX compete more directly in AI agents and coding tools, where practical developer utility matters as much as raw model quality.

Buying Cursor does not automatically close the gap with the best frontier-model labs, though. 

SpaceX would still need to keep improving model quality, not just own a popular application layer. But the move leads to a more integrated approach (energy, compute infrastructure, models, apps).

Mergers, Joint Ventures or Investments as Routes to Controlling AI Model Costs

Just how artificial intelligence model providers might improve their economics is a key business model issue. 

A shift to inference operations also emphasizes the importance of reducing cost per token at scale. 

source: McKinsey 

Where software often has marginal costs close to zero, use of AI models seem to have costs that scale almost linearly with usage, so marginal costs are high. 

A few key business issues are clear enough.

A greater shift of cost towards fixed cost rather than variable cost seems necessary. For example, model creators could move in the direction of owning their compute infrastructure rather than renting cloud capacity.

The problems with high variable cost are clear:

• Unit economics

• Cash burn and need for capital injections

• Competitive pressure

• Capital allocation

Right now, model queries do not show software-style marginal cost trends, where marginal costs are close to zero. 

Every additional user or query drives proportional costs for GPUs, power, and data center capacity. 

Reports suggest inference alone consumes 50 percent of revenue in some cases. Some suggest the problem is worse, with at least some model providers spending more than 100 percent of revenue on compute services. 

By itself, that might not be existential, as model providers are in the early stages of growth, meaning incremental revenues would not be expected to cover the full costs of creating and operating the models at scale.  

But observers do worry about marginal costs that seemingly do not have software-style economics with near-zero marginal costs.

Cash burn and capital intensity also are issues. Rapid revenue growth is offset by even faster cost scaling, which necessitates high investor funding requirements, borrowing, equity raises. 

Pricing issues also are an issue. When model suppliers raise prices, introduce usage-based tiers, or limit free access, they risk customer churn or slower adoption, even as they address revenue issues. 

Strategic vulnerabilities also exist when model suppliers are dependent on external suppliers for crucial computing services (operating costs, ability to manage surges of demand). 

Capital allocation is an issue as well. One might argue that model builders should divert capital into compute infrastructure, but that is hugely expensive and detracts from the job of developing the next generation of models.

So the issue is how to fix the problem. Revenue growth with scale obviously helps, but doesn't solve the marginal cost issue. Efficiency improvements, owned infrastructure and pricing innovations all will play a role.

Smaller or distilled models, sparse activations, better architectures, prompt caching, batching, quantization, and routing to cheaper models for simpler tasks will happen.

Inference costs per token have dropped significantly in some cases, allowing gross compute margins to  improve.

source: Beth Kindig 

Stranded assets always are a problem, so higher GPU utilization rates help. So do custom silicon and algorithmic advances.

Owned infrastructure is a partial answer. Model builders and compute suppliers are investing heavily in their own data centers and chips (Anthropic's $50 billion commitments for custom U.S. facilities with partners like Fluidstack; OpenAI's Stargate).

Revenue models also are adjusting. Higher enterprise pricing, usage-based tiers, value-based pricing (charge relative to delivered value, not just tokens) and premium features or apps are introduced. 

Given all that, one logical historical precedent is for mergers and acquisitions that place model building and compute functions under a single ownership. On the other hand, antitrust regulations will probably tend to restrict options for some of the most-likely buyers (Alphabet, AWS, Microsoft, SpaceX, for example).

So other forms of cooperation are likely to develop. 

Expect partnerships, joint ventures, dedicated capacity deals, and partial ownership rather than full-scale mergers and acquisitions, which will face antitrust opposition. 

When feasible, model builders are creating their own compute infrastructure. OpenAI's Stargate project with Oracle and SoftBank (up to $500B, multi-gigawatt scale) provides an example. Investments in neocloud suppliers is another example. 

But that might be the exception to the rule. Anthropic, for example, has chosen to sign big supply deals rather than build its own facilities.

Partnerships for Tensor Processing Units, Trainium, and other accelerators reduce reliance on expensive third-party GPUs and improve efficiency also are growing. 

But full vertical integration might not be the immediate or mid-term path forward, partly for regulatory scrutiny reasons; partly for capital intensity reasons and partly for business diversity reasons. Both model builders and compute infra providers prefer a diversity of partners. 

So joint ventures and consortia that also have the advantage of off-balance-sheet implications will happen. 

In summary, tight strategic integration and partial ownership rather than blockbuster mergers are the main approaches. That avoids regulatory opposition and also is capital efficient. 

Orbital AI Compute Seems to be Coming, but Not at Scale, Right Away

With SpaceX going public on June 12, 2026, lots of investors will be pondering the feasibility of creating orbital data centers at scale.

But space-based data centers are not an immediate replacement for terrestrial data center alternatives for reasons of initial cost and capacity. Launch costs remain substantial.  

Potential upsides center on lower ongoing costs offsetting high upfront costs, eventually, though initial total operating costs will probably not match terrestrial alternatives:

• Cheap/abundant power: Solar in orbit provides ~36% higher irradiance, near-constant supply (no night/clouds/weather), and very low marginal costs (projections ~$0.005/kWh vs. $0.04-0.08/kWh terrestrial wholesale). No grid connection, fuel, or large storage needed in ideal orbits.

• Lower OpEx: Projections include 97% lower operating costs in some models (energy + cooling). No land, permitting, property taxes, or water for cooling. Avoids terrestrial delays/queues for power infrastructure.

• Scalability and utilization: Unlimited "land" in orbit for expansion. High utilization from constant power. Falling launch costs could lead to cost parity or better for power-dominant workloads by late 2020s to 2030s.

Orbital systems could ease some important terrestrial obstacles:

• Energy and emissions: Relies on clean solar (potentially 10x lower CO₂ emissions). Reduces strain on terrestrial grids, which often use fossil backups for data centers.

• Resource Savings: No water consumption for evaporative cooling (a major terrestrial issue). Frees land for other uses; avoids local ecosystem/power price impacts from hyperscale farms.

• Overall footprint: Could lower terrestrial data center growth, helping with power queues, water scarcity, and NIMBY opposition.

Of course, environmental impact is still there. Launch emissions, space debris (cluttering orbits, potential Kessler syndrome risk), manufacturing impacts and end-of-life disposal remain issues. 

Some use cases might make more sense. Workloads tolerant of moderate latency (~100-500 ms round-trip) and benefiting from proximity to space data or constant power suggest suitability for:

• AI Inference: Querying trained models (chat, search, voice agents, video generation, back-office automation)

• Some telemetry use cases: Onboard near-source analysis of Earth observation, climate monitoring, disaster detection (wildfires/floods), maritime surveillance, sensor apps

• Some edge compute cases: Real-time processing for satellites, space cybersecurity or autonomous operations or resilience against terrestrial outages/disasters

• High-Security/ Sovereign Compute: Isolated environments for sensitive data, national security, or regions with poor terrestrial infrastructure.

AI Infra Financing Gets Creative

Financing of AI infrastructure has evolved into a complex, multi-layered financial architecture that extends well beyond traditional corporate balance sheets. 

External financing structures include:

• Strategic partnerships: Frontier model labs and hyperscalers are forming partnerships for regional development, power infrastructure, and equity contributions

• Public sector and sovereign support

• Captive markets: In some instances, state-owned enterprises or governments direct domestic demand toward local chip manufacturers.

Financing Model	Description	Example / Context	Source
Structured/Off-Balance Sheet	Using infrastructure funds and private credit to distribute risk across a layered set of claims.	General industry shift toward using private credit and structured vehicles to fund data center buildouts.	BIS
Community-First Partnerships	Joint commitments between developers and providers to share infrastructure costs and regional responsibilities.	Microsoft's "Community-First AI Infrastructure" plan and OpenAI's "Stargate Community" initiative.	HKS
National Sovereign Investment	Coordinating investments in data, compute, and algorithms through sovereign-backed frameworks.	Frameworks for "AI Triads" in low-to-middle-income countries using structured funding tranches.	Oxford
Captive Market Funding	Generating revenues through domestic mandated demand to fund internal R&D cycles.	Huawei’s AI chip revenue generation within the Chinese domestic ecosystem.	Bruegel

In many instances, the intent is to reduce capital investment requirements by moving to off balance sheet vehicles or “compute as payment” arrangements.

Hyperscaler	Model Supplier	Deal Type / Structure	Estimated Value / Capacity	Source
Google	Anthropic	Multi-year compute commitment + Equity investment	Up to $40B investment; 3.5GW TPU capacity (via Broadcom)	Silicon Republic
Amazon (AWS)	Anthropic	Compute credit + Equity investment	Up to $25B total commitment; multi-year cloud compute	Silicon Republic
Microsoft	OpenAI	Exclusive cloud provider + Multi-stage capital injection	~$10B+ in multi-year funding; 49% profit stake	Aranca
Meta	N/A (Self-build)	Structured finance (SPV) for data center buildout	~$30B "Hyperion" SPV (Blue Owl Capital led)	SoftwareSeni
Google/Anthropic	SpaceX	Compute infrastructure delivery contracts	Potentially >$70B over multi-year term	AA

As seen with Meta’s "Hyperion" transaction, hyperscalers are increasingly utilizing Special Purpose Vehicles (SPVs) and partnerships with private credit firms (e.g., Blue Owl Capital) to fund massive data center buildouts. This allows the companies to offload the capital intensity of the physical build while retaining operational control and capacity priority.

In many of these deals, "compute" has become a literal form of payment. The Google-Anthropic and Amazon-Anthropic deals are not merely cash-for-equity; they are deeply intertwined with multi-gigawatt (GW) capacity commitments and customized hardware access (such as Google’s TPUs).

Financing is no longer focused just on chips. The capital is increasingly directed toward the "AI Triad"—the integration of compute, dedicated energy infrastructure, and data center physical shells. This is evidenced by the trend of co-locating data centers with renewable energy sources and the invocation of national defense acts (as seen in the U.S. in early 2026) to prioritize grid expansion for AI.