AI Has an Appetite Like an Industrial Powerhouse

The figures published by the IEA in its special report Energy and AI are alarming. In 2024, data centers worldwide consumed 415 TWh of electricity, which is roughly 1.5% of global consumption. Since 2017, their consumption has grown at a rate of 12% per year — more than four times faster than total electricity consumption. And the acceleration is just beginning. The main driver is artificial intelligence. Training large language models like GPT-5, Claude, or Gemini requires enormous computing power. A typical AI data center consumes as much electricity as 100,000 households, and the largest facilities under construction have demands up to twenty times higher. For comparison: such a facility has higher energy requirements than an aluminum smelter.

Panthalassa: The Data Center That Floats

Startup Panthalassa, based in Oregon and employing former engineers from SpaceX, Google, Blue Origin, Apple, and Boeing, is betting on an unconventional solution. Instead of fighting for capacity in overloaded land-based power grids, it is building autonomous floating platforms equipped with computing hardware directly on the open sea. The first pilot project named Ocean-2 is already being tested in the Pacific Ocean. The floating data center is meant to be fully self-sufficient — it gets its energy from ocean waves, and all operations are autonomous, with no human crew needed on board. The company describes itself as a "planetary energy platform" and its goal is to offer ultra-low-cost, emission-free energy for computing infrastructure anywhere in the world.

Why the Ocean?

The answer is simple: land-based energy infrastructure can't keep up. According to the IEA, up to 20% of planned data centers face the risk of delay due to insufficient grid capacity. Waiting times for key components like transformers and cables have doubled over the last three years. Building new transmission lines takes four to eight years in developed economies. The ocean, on the other hand, offers practically unlimited space and a constant source of renewable energy. Unlike solar or wind, waves are more predictable and available 24 hours a day. Moreover, platforms can be placed far from densely populated areas, eliminating objections from local communities against the construction of giant data halls.

Salt Water, Rust, and Other Obstacles

But not everything is rosy. Experts contacted by New Scientist point to a fundamental problem: maintenance. The marine environment is extremely aggressive — salt water, humidity, and corrosion can destroy sensitive electronics faster than anyone would like. Repairing a server in the middle of the ocean is a logistical nightmare. Another question mark hangs over latency. Data centers located hundreds of kilometers from the coast mean longer response times for users. For model training, this doesn't matter — raw performance is what counts there. But for real-time applications like voice assistants or autonomous driving, the delay can be critical.

Space, Nuclear, and Other Alternatives

Panthalassa is not the only one looking for new energy sources for AI. Some companies are seriously exploring the idea of data centers in orbit, where solar panels would have access to constant solar radiation without atmospheric losses. However, the cost of launching hardware into space is still astronomical. A more realistic path is small modular reactors (SMRs), compact nuclear power plants that could directly power data centers. The IEA expects the first SMRs to come online around 2030. Microsoft has already signed an agreement to purchase energy from a fusion reactor, and Google is investing in geothermal projects.

What Does This Mean for Europe and the Czech Republic

Europe accounts for roughly 15% of the growth in data centers, and this share will increase. The EU AI Act also requires that high-risk AI systems meet strict transparency and energy efficiency requirements — this indirectly pushes for greener power for European data centers. In the Czech Republic, we won't be building floating data centers on the sea, but the topic of AI's energy intensity concerns us directly. The Czech Republic hosts a growing number of data centers (for example for O2, T-Mobile, or multinational clouds), and the Czech National Bank recently purchased its own AI hardware from Nvidia for financial market oversight. With growing consumption, the Czech power grid will face the same pressure as grids in Western Europe and the USA.

Are Floating Data Centers Science Fiction or Near Reality?

Panthalassa has not yet announced any commercial customers, and the Ocean-2 project is still in the pilot phase. But the speed at which demand for AI computing power is growing suggests that unconventional solutions will be needed — and sooner than most of us think. Whether the future belongs to floating platforms, small reactors, or a combination of both, one thing is certain: without massive investment in energy infrastructure, the AI boom will hit a hard ceiling. And that ceiling is rapidly approaching.