Written by Xavier Blot, Associate Professor at emlyon business school
On his first day, Trump announced the launch of the Stargate Joint Venture promising $500 billion in investment for the deployment of data centers for AI. The announcement is symptomatic of a frenetic period where the race for power requires an enormous supply of energy. Amid the effects of the announcement, trends are emerging.
The question is whether we can produce enough electricity, ideally low carbon, to meet growing demand. Today, this is not the case and energy constraints are likely to appear in Saudi Arabia or Ireland. In the United States, this challenge is translated locally, as in the minnesota where the projects will add the equivalent of the consumption of 2.3 million households. At the country level, Data from the Lawrence Berkeley National Lab make it possible to estimate that 525 TWh of renewable energy could be produced by 2030. Knowing that other uses must be electrified, 7 to 12% would be available for data centers, or about 50 TWh when 200 to 400 would be needed!
Tech players are already investing heavily in renewable energies, particularly through PPAs. Microsoft announced a partnership with Brookfield to build 10.5 GW of renewable capacity and a commitment for 12 GW of Qcells solar modules of which 1.5 GW will be deployed per year until 2032. These intermittent energies are suitable for data centers for inference of models (use) that have varying loads. Despite these investments, additional capabilities are needed.
In addition, for data centers used to train AI models, the supply must be more stable. This is why some GAFAM rely on nuclear power to, for example, Rerun a slice shutting down or get supplies from an existing reactor. However, the available capacities are not endless and are limited by the regulator. It is to develop new abilities that investments are made in SMRs. But it will be necessary to wait at least ten years to consider large-scale deployment. On the same level, the geothermal energy Presents ofobvious qualities but deploying it will take time.
Faced with this impatience, gas appears to be an “ideal” short-term solution. Reliable (at base and peak) and quick to deploy (A little over 2 years). This is the option chosen by xAI to launch its data center in record time with 14 gas power plants. Three others for a total power of 2.3 GW will serve a new Meta data center. On the developer side, the signals are also green. Exxon and TotalEnergies plan to deploy each of the 1.5 GW gas power plants for the sector. Goldman Sachs estimates that they will meet 60% of data center demand. This forecast should be taken with perspective but it is not illusory to imagine a real rebound in the sector with the Trump presidency, which wants to make gas a central element of his energy strategy. THELicensing is already facilitated and other levers are added, such as the conversion of coal-fired power plants to gas and the extension of the lifespan of existing power plants.
Finally, the sector is facing a final challenge with deployment times that are getting longer. Google noticed this recently : it takes 4 years today to connect a new solar power plant, compared to 2 previously. Transmission constraints and Interconnection times are limits that are difficult to overcome and no tech player is tackling them. Circumvention strategies are developed instead. The first is to connect production plants directly to data centers, without going through the network (behind-the-meter). The second would take advantage of new designs such as Microgrids and Energy Parks. The sizes and modalities vary but the idea remains the same: joint planning of data centers and power plants, “intelligent” local balancing (facilitated by AI) and a single connection to the network.
The approaches are therefore rich and it is difficult to predict which ones will become dominant. Political and economic interests intersect and gas may have little room for manoeuvre given Trump's export ambitions. Will the demand for energy linked to AI evolve differently from the scenarios? DeepSeek recalls that the sector can benefit from significant gains in energy efficiency at the hardware and software level. In any case, tech companies are determined to invest where needed and they have the resources to do so.