US Data Centres Could Burn More Natural Gas Than Germany and Japan Combined by 2035

A new analysis has put a number on something that has been worrying energy economists for a while: US data centres, driven primarily by the AI infrastructure boom, could consume more natural gas than the entirety of Germany and Japan combined by 2035. That is not a marginal increase. That is a seismic shift in global energy demand, and it has implications that go well beyond the tech industry.

To understand the scale of what is being described here, Germany and Japan together represent two of the world's largest industrial economies. Germany consumes roughly 80 billion cubic metres of natural gas per year, and Japan consumes around 100 billion cubic metres. If US data centres are on a trajectory to exceed both combined, we are talking about an additional 180-plus billion cubic metres of annual gas demand from a single sector within a decade.

Where Is All This Energy Going?

The answer is AI. Training large language models, running inference at scale for hundreds of millions of users, and storing the data those systems generate all require enormous amounts of electricity. Modern AI chips like Nvidia's H100 and B200 series draw hundreds of watts each, and the data centres housing them are packed with thousands of these chips running continuously.

Microsoft, Google, Amazon, Meta, and a host of newer AI companies have all announced multi-billion-dollar data centre construction programmes in the past two years. These facilities are not small. A hyperscale data centre can draw 100 megawatts of power or more, the equivalent of a small city. And natural gas, which can be turned on quickly to meet demand spikes, has become the preferred backup and increasingly primary fuel for the grid capacity serving these facilities.

It is worth noting that many tech companies have made ambitious net-zero commitments. Microsoft pledged to be carbon negative by 2030. Google claimed to run on 100 percent renewable energy matched. But the sheer pace of AI infrastructure build-out has, in the bluntest possible terms, outrun those commitments. Microsoft's own emissions have reportedly risen sharply since it began its AI push, not fallen.

The Grid Problem

The US electricity grid was not built for this. Upgrading transmission infrastructure is a slow, expensive, and politically complicated process. In the interim, the fastest way to get power to a new data centre is often to connect it to gas-fired generation. This is exactly what is happening in states like Virginia, Texas, and Georgia, which have become major data centre hubs.

The analysis suggests that the AI industry's energy demand is not just a climate story, it is an infrastructure story. Grid operators are already flagging concerns about reliability as demand surges faster than new supply can be brought online.

The Political Dimension

This is landing in an awkward political moment. The current US administration has pulled back on climate commitments, and the AI industry has broadly benefited from that relaxation. But even setting climate aside, gas consumption at this scale has supply chain, geopolitical, and price implications. Natural gas markets are global. A surge in US domestic demand affects prices for consumers, for industrial users, and for countries that import US liquefied natural gas.

AI companies have started to invest in nuclear power as a longer-term solution, with several striking deals with nuclear operators to secure dedicated capacity. But nuclear takes years to bring online, and the 2035 projection is not that far away.

The uncomfortable reality is that the AI boom and the energy transition are on a collision course, and right now the AI boom appears to be winning.