For years, big technology companies projected a clean image: data centers powered by renewables and commitments to climate neutrality. But the explosion of artificial intelligence is putting that narrative to the test. Electricity demand is growing at a rate that the grid cannot keep up with, and the fuel that is covering the gap is not the wind or the sun. It is natural gas.
The contradiction is already visible in the numbers. Google and Microsoft consume around 24 terawatt hours (TWh) of electricity per year each, more than more than a hundred countries. And while they announce record clean energy contracts, their emissions continue to rise: Google has increased its emissions by 48% in the last five years and Microsoft by 31% since 2020. An independent analysis rated climate integrity of several technologies as “poor” or “very deficient” in the face of the energy boom of AI. The cloud is not ethereal. It’s physics. And for AI to work without interruptions, we are starting to burn more hydrocarbons.
The electron fever. The phenomenon is not marginal. A report from the Open Energy Outlook initiative—led by researchers at Carnegie Mellon and NC State— projects that electricity demand of data centers and crypto mining could grow by 350% between 2020 and 2030, going from representing 4% to 9% of total consumption in the United States. Goldman Sachs points in the same direction: Specific consumption of data centers could increase by 160% before the end of the decade.
The pressure has already broken market balances. In December 2024, in the PJM region—which supplies 13 states in the eastern United States and has the highest density of data centers in the world—capacity prices went from $30 to $270 per MW-day in a single auction. The extra cost will end up affecting the bills of some 67 million customers. John Ketchum, CEO of NextEra Energy, described it as a “golden era of energy demand”, but warned of a physical limit: “the new electrons cannot reach the grid quickly enough.” And in that void between explosive demand and insufficient supply is where gas reappears.
The tyranny of 24/7. If renewables are increasingly competitive, why not cover this demand with more wind and solar? The answer is technical. Artificial intelligence requires continuous, 24/7 supply. It cannot be turned off when the wind goes down or the sun goes down. As Manuel Losa, manager at Pictet Asset Management, explained, to the Financial Times: If demand grows and firm energy is needed 24 hours a day, “today, the only way to achieve this is with gas.”
The problem is not the marginal cost of renewables, it is firmness. Without massive storage or reinforced grids, solar and wind generation cannot guarantee constant supply. And the deployment of new transmission lines is slow and contentious. Furthermore, traditional electrical planning assumed growth of 1-2% annually; Now there are areas with increases of 20-30% annually linked to data centers.
The quickest solution today is to build or expand gas-fired generation. But even there there are limits. Gas turbines—critical equipment—have become a bottleneck. Just three years ago, Siemens Energy executives stated that the turbine market was “dead” in the face of renewable advancement. Today, the factories are overflowing. Global orders are expected to exceed 1,000 units this year, with the United States absorbing almost half. Delivery times can be extended up to five or even seven years in some cases. The bottleneck is no longer the chips. They are the turbines.
So what happens with renewables? Renewables do not disappear. In fact, they continue to expand. Google has signed agreements to purchase nearly 1.2 gigawatts of new wind and solar energy in the United States from Clearway Energy. Big tech companies continue to sign clean energy contracts in multiple regions. However, the problem is temporary and structural. Purchasing renewable electricity does not guarantee that hourly consumption is supported by clean generation at that same time and place.
In fact, there are solutions. Battery storage and grid upgrades can increase renewable integration. The Open Energy Outlook report shows which regions like Texas, with more investment in transmission, they manage to take better advantage of wind power to feed new demand. But deploying storage and hardening the network takes years, and AI is growing rapidly.
For this reason, even companies traditionally focused on renewables are expanding their portfolio in gas, How did you have access? Financial Times. NextEra has announced plans to develop up to an additional 8 gigawatts of gas-fired generation. Clearway builds hybrid data center campuses combining renewables and combustion turbines. It is not an explicit abandonment of renewables. It is an emergency solution.
But there is also nuclear. amazon tried to connect directly a data center to the Susquehanna nuclear power plant to ensure stable and clean supply. Federal regulators blocked the deal over potential effects on grid stability and the impact on other consumers.
Furthermore, Google has signed an agreement with Kairos Power to develop seven small modular reactors (SMR), with the goal of adding 500 MW emissions-free by 2030. Microsoft and other companies are exploring similar deals. But even in the most optimistic scenario, new nuclear capacity will not be operational on a relevant scale before the end of the decade. AI needs electricity now.
A clash of transitions. Five years ago, natural gas was presented as a retreating bridge fuel within the energy transition. Today it has become the structural support of artificial intelligence. A friction between two transitions that advance at different paces: the digital one, exponential; the energy, regulated and slow.
As the Open Energy Outlook initiative warnsthe choice should not be between digital progress and network stability. But if energy planning doesn’t adapt more quickly—more transmission, more storage, better market design—the expansion of AI could mean more gas, more emissions, and higher bills. Artificial intelligence promises efficiency and intelligent decarbonization. But for now, its massive expansion is prolonging the life of the fossil generation. The digital future is advancing at full speed and the energy transition cannot catch up. And in that gap, the gas burns again.
Xataka | Anthropic is growing so fast that OpenAI’s problem is growing at the same speed: losing the market that matters
GIPHY App Key not set. Please check settings