Environmental Moonshot Data Centers

Driving the industry towards sustainability

There is a growing opposition to data centers across the United States for good reason: they are harmful to local communities in myriad ways, including extreme overuse of water and power, as well as generation of noise and pollution, while generating fairly minimal jobs.

Given that these data centers aim to power a revolution in intelligence, why not make their design a primary measure and focus of our intelligence?

What could a data center look like if it was built with the most leading edge thinking on sustainable design, so that they do not use external water, generate all their own power from solar, wind or geothermal, and are visually appealing and sonically neutral, with no downstream pollution in their neighborhoods?

What would the equivalent of gold standard environmentally certified data center look like? The good news is that innovation is happening globally but hasn’t yet become standardized for the industry as a whole. Here’s a listing of top 10 sustainable data centers in the world, ranging from 100% solar arrays at the massive Moro Hub in UAE to Europe’s first gigascale Start campus in Portugal pioneering the use of sea water. It’s clear that more companies are seeing this as an area for competitive differentiation that also leaves the world a better place.

How do we help these innovations becomes standardized?

Our current trajectory of massive, rapid implementation of dirty, wasteful centers is short-sighted and ultimately more expensive for society, with the risk of very real backlash. AI companies will become known as the “enemy of the people” if the trajectory doesn’t change. With climate change taking on increasingly dramatic and dangerous forms, we need to ensure that future generations benefit from today’s innovations rather than inherit a desiccated, denuded landscape populated with data centers and diminished human culture.

Here’s how we could approach the acceleration of positive change.

First, we could have a national competition for the most sustainable and beneficial data center design every year (or even every six months) - a moonshot prize of perhaps $100 million dollars - that inspires engineering talent around the world to take a shot at winning.

Next, we need to require the winning designs be open-source, so they are not proprietary or owned by any particular company. After each design demonstrates its innovations, that is integrated into the new standards for approving the next generation of data centers. National permits are then issued for a limited time for a data center to operate and those permits are contingent on continuing to improve to meet the national standards. Every data center gets an eco score for how sustainable it is (based on energy efficiency, use of water, and pollution) and the minimum scores keep going up.

That drives change at the design level. But focusing purely on the building side would not be enough. Every AI model that is available as a subscription to U.S. consumers could be required to publish their model score showing the average eco score of all of its data centers. Users who care about the impacts of an AI system would thus have a clear signal at the point of use, showing which models and companies are more committed to creating a healthy world and therefore warrant our business. Companies might choose to only pay for models that have scores above a certain level. This creates market pressure for companies to migrate their computer towards more and more ecologically efficient data centers as they come on line. This gives a carrot to justify higher spend on the factors that reduce environmental damage. Use of outdated, inefficient, environmentally harmful data centers would drag the AI model score down and undermine the company’s brand.

With sufficient open-source design incentives, paired with environmental standards and consumer pressure, this could drive data centers towards total resource self-sufficiency quickly. Instead of resource-draining islands of pollution, data centers could become greener neighbors, with self-sustaining power, water, and operations. The best in the world are already demonstrating it is possible. Now the task is to accelerate adoption of the innovations.

Right now, compute is one of the most expensive parts of using an AI, which means that there are strong incentives to cut corners. We need to create an incentive structure so that instead of cutting corners, the companies are embedding forward-looking, sustainable values into their products.

As we move deeper into the AI era, we’d have an evolving set of national standards that push data centers to becomes self-sufficient resource islands that are perceived as good neighbors, a policy framework that encourages rapid innovation, and a market-based ratings system in which consumers themselves can help motivate companies to move in the right direction.

The alternative is a world where data centers are incredibly wasteful of our most precious, shared resources, especially water and thus become a target for societal backlash. The only real way for AI to make a real contribution to society is to think holistically about all impacts. Setting up the incentives for them to do so at the foundational level of compute is essential. That is the foundation for truly wise AI.