As a percentage of GDP, we are spending more than ten times as much on AI infrastructure as we did on the Apollo space program.
- Big Tech’s AI Push Is Costing a Lot More Than the Moon Landing
- AI Investment Will Hit 2% of U.S. GDP This Year… Nearing Defense Spending Levels
Looking deeper at the Apollo program, there are several lessons for AI, starting with President Kennedy’s Special Message to Congress on Urgent National Needs.
"I believe that this nation should commit itself to achieving the goal, before this decade is out, of landing a man on the Moon and returning him safely to the Earth.”
At the time of that declaration in 1961, we were incapable of carrying out the mission. A vast number of challenges had to be solved, and multiple required technologies had not yet been invented.
- Advanced computing
- Radiation protection
- Life support systems
The idea of sending a man to the moon before solving these problems would be preposterous. Instead, we systematically created the technology to solve legitimate problems. Many of those technologies are still used today.
Today's moonshot is the pursuit of Artificial General Intelligence (AGI). On our current trajectory, we may well achieve orbit, but it is not clear that we will return safely to Earth. Chiefly, that’s because we are focusing on extractive value and consumption rather than creative value and production.
Were we to learn from the Apollo program, these are some of the life-sustaining challenges we would seek to address before continuing on our current path:
- Energy Consumption
Data centers consume so much energy that municipalities are forced to build more generation capacity in response. Utilities then increase the rates consumers pay to offset additional costs, while municipalities offer tax breaks to data centers to lure them in. It’s a case of distributing costs while privatizing any profits. - Water Consumption
Clean water is one of the few life-sustaining planetary resources for which there is no substitute. It helps cool the tremendous heat produced by data centers. There are alternative cooling methods, but water is the most cost-effective for big tech. So by consuming up to 5 million gallons per day, watersheds are being impacted, and water quality for human consumption is being degraded. People then incur the additional expense of finding alternative sources of clean water, rather than data centers paying the true cost of their usage. Even worse are cases where the data centers inadvertently increase the levels of toxic bacteria in the water that they discharge. The risk here is that we are setting up another Flint, Michigan Water Crisis, which cost well over $350M as of 2020, and that is a risk for every single data center. - Land Use
Regarding energy production, utilities are taking people's property and houses through eminent domain. This is to make way for power distribution lines to reach the data centers. Specifically, were it not for the data centers, people would still have their homes.
Planning for Reentry
Addressing the challenges of energy, water, and land use on the level demanded by AI is a wicked problem for which we do not have ready-made solutions. Charging ahead before addressing them is either corporate welfare, irresponsible planning, or both.
I am not opposed to data centers, nor to the proliferation of AI. Instead, I am opposed to the rapid proliferation of resource-intensive data centers before we have addressed their massive consumptive needs.
Imagine the possibilities if we solved the sustainability problem before AI ran humanity off a cliff. The alternative has us trying to create more energy and water afterward, from a position of desperation.
These unresolved resource needs represent the modern equivalent of designing life-support systems and heat shields for the early space program. We did not have answers then, and we don’t have them now. But back then, we took it upon ourselves to systematically address the challenges as part of the effort to land on the moon. Today, we are standing on the launch pad on a voyage to achieving AGI, and we have no plan for reentry. We need to solve multiple challenges before launch, not afterward.
Energy
For energy production, we need a massive addition of renewable energy streams - both central energy like wind farms and distributed energy like solar panels. Also, despite it being a political hot-button, we may also need to bring on significant nuclear power production and address the risks to make it viable. There won’t be a silver bullet here - it’s going to take a broad number of independent efforts, all of which face a War Against Renewable Energy and cuts to nearly $8B in clean energy projects.
It is undeniable that data centers require a phenomenal amount of energy. It should be undeniable that energy needs to be produced (or saved) elsewhere to supply it. If a data center project requires an energy equivalent of powering a small city, then it should be required to bring an equivalent amount of renewable energy online as a prerequisite. With only some loss from transmission, energy is fungible. Data centers that bring renewable energy online in one area could use it in another in a zero-sum game.
A similar model is already in place with the Department of Natural Resources for wetland development. Before a developer can build on an existing wetland, they need to identify and replace it with another one using a 2:1 swap. If permitting for data centers were to require bringing additional energy production online, then communities and societies would be net beneficiaries of these projects rather than subsidizing their operations. Unfortunately, data centers are consuming energy now, claiming that replacement sources will be available within 18 months, despite conservative estimates pointing to closer to seven years.
Data centers could also save energy by directing the heat into other industrial processes (or homes) that require it, thus saving energy for those other systems.
- Heat from data centers and factories to warm thousands of homes
- In Belgium, a factory is heating homes
- How to Put the Heat from Data Centers to Good Use
- Can Data Centers Heat Our Buildings?
None of these options is a definitive solution to the energy needs. I don’t pretend to have the answers, but I assert that finding answers before launch is better than praying for them in orbit. Regardless of the eventual solutions, the most viable options are ones where data centers shift to being value creators rather than merely extractors. An additional incentive for solving the challenge of energy production is that we would simultaneously position ourselves to solve for water as well.
Water
For cooling, we may invent chips that require far less energy or produce far less heat. But until the laws of thermodynamics change, we will be forced to innovate with water. That could include cooling with gray or recycled water, including the infrastructure needed to make this viable.
Another solution for more fresh water is desalination. Presently, according to Desalination Cost Per Gallon: Price Guide for U.S. Buyers 2026, the cost per gallon ranges from $0.03 to $0.12. For a data center using 5MM gallons/day, that would cost between $150,000 and $600,000 per day. However, the highest cost in desalination is energy consumption, so by solving for production above, we also gain the leverage needed to solve for water.
With numbers that large, one might conclude that there’s no way data centers would ever be remotely profitable. If that is true, then the only way they can be operating now is at a tremendous loss or with massive subsidies currently provided by governments. We find ourselves in a pay-now-or-pay-more-later scenario akin to that of the payday lender. By building out power and water solutions now, we have the luxury of choice, even if we don’t like any of the choices. By building solutions later, we will find ourselves in a lifeboat-ethics position, forced to choose who will be allocated an ever-dwindling share of critical resources, including the water needed for life.
By analogy, today’s AI data centers are like a patient who needs to be intubated long enough to survive on their own. Here, intubation comes in the form of subsidies afforded to the industry. The decision to intubate is predicated on two assumptions: that the patient will grow stronger, and that if it does not, then we will subsequently withdraw care.
Except that every NIMBY resisting a data center is crying out for a DNI order (Do-Not-Intubate) while municipalities are intubating funds as fast as they can in a race to the bottom. And if the AI industry fails to become self-sufficient later, it will be exponentially more difficult to withdraw support from a series of tools and technologies that will have become deeply integrated with human industrial systems like Mr. Smith in the Matrix.
Not only are these problems avoidable if we solve for water first, but doing so opens the possibility of phenomenal human thriving in new areas, which brings us to the issue of land.
Land
Presently, data centers are being built in areas where they already have access to power and water. They essentially become competitors with human beings for these resources. But imagine what becomes possible if we solve for energy and water.
- We would not need to build data centers so close to human population centers.
- The extra water production could also be used to make arid lands livable and farmable.
- Rather than having people compete with data centers for resources, bringing new resources online first to serve data centers could vastly expand the areas where it becomes viable for humans to live.
Costs
The most obvious counterargument to addressing energy, water, and land issues is cost. However, the only thing more expensive than creating solutions is the cost of doing nothing.
By most accounts, politically, financially, and militarily, nothing is going to stop the pursuit of AGI.
- Politically, we are in a story that AI is the most transformative technology since the Industrial Revolution, and to stop it (or even slow it down) would risk being left behind.
- Financially, we have already invested nearly 2% of GDP, and even if we are sending good money after bad, the astronomical expectations of returns make turning around a near impossibility unless (or until) a massive crash.
- Militarily, the pursuit of an AI is already recognized as a modern AI Arms Race, and no country can afford to lose, even if the real loser is the whole of humanity or the planet.
When President Kennedy made his bold declaration about the moonshot, he also directed Congress to allocate the resources to make it possible. If we are declaring a future in which the US is a leader in AI development, then we must also make a corresponding investment in the resources that make it possible.
This means substantial federal investments to bring new resources online and vastly expand what is possible, starting with energy, water, and land. It may also require that municipalities stop subsidizing data centers through tax breaks. Instead, local and federal governments could use their financial resources to buy ownership in technology companies so that future benefits can funnel back to the people who really bore the costs.
Not long ago, that idea would trigger alarms of Nationalization. Perhaps it still should. However, the US government is already a shareholder in 26 companies but lacks equity investment in AI. It is undeniable that AI is going to play an increasing role in society. It is creeping into almost every application available to us.
If the idea of the government owning even part of this increasingly critical technology still seems foreign, then ask yourself:
Who should own AI?
Its very existence depends on being trained on the contributions made by all of society, often without permission, acknowledgment, or compensation. The real foreign concept should be how private ownership of such a tool is even imaginable.
Right now, communities are subsidizing AI through utility rates, tax breaks, and sacrificing resources, but they are not reaping the benefits. Equity ownership at this point would not be nationalization. Instead, it would be fair compensation for the sacrifices being forced upon communities.
Even if AI is unstoppable, that does not mean that our future is unshapable. Let us demand that it be created in a world dominated by human thriving. That means solving for energy, water, and land as a prerequisite of data center expansion, so that we don’t find ourselves competing with AI for the prerequisites of our survival.
We solved the impossible before.
We can do it again.
But only if we insist on planning before prompts.
This is part 3 of a three-part series
1. AI Is a Payday Lender
2. Wolves of Yellowstone
3. Reentry: Lessons from Apollo