OpenAI, Google, and Meta: The Hidden Energy Battle of AI

OpenAI, Google, and Meta: The Hidden Energy Battle of AI

Artificial intelligence is often portrayed as a race for models. OpenAI, Anthropic, Google, Meta, and Mistral AI compete through benchmarks, reasoning capabilities, and billions of parameters. However, behind this competition lies a battle that takes place not in laboratories or research centers, but in electrical grids, substations, and the trading floors of infrastructure funds.

The AI economy is entering a new phase. After the race for data and then for GPUs, we now have the race for gigawatts.

This evolution profoundly changes the very nature of the sector. For over twenty years, the digital economy has thrived on an implicit assumption: electricity was available and abundant. Infrastructure was a secondary concern compared to software, but AI reverses this logic. Every technological advancement requires more computation, more servers, and more energy. The limits are no longer just algorithmic; they are becoming physical.

According to the International Energy Agency, data centers consumed approximately 460 TWh of electricity worldwide in 2024. This figure could exceed 1,000 TWh before the end of the decade due to generative AI. For comparison, this would represent a consumption higher than that of Japan today.

This dynamic is already visible among hyperscalers. Meta plans to invest between $125 billion and $145 billion in its AI infrastructure by 2026. Microsoft could allocate between $115 billion and $135 billion to its computing and cloud capabilities. Alphabet is following a similar trajectory with an investment program of around $80 billion. Behind these amounts lie data centers whose energy needs now reach several hundred megawatts.

The Stargate project in the United States illustrates this shift in scale. Its ambition is to develop several gigawatts of computing capacity dedicated to artificial intelligence. At this level, a data center becomes an energy consumer comparable to a large metropolitan area or a major industrial complex.

This escalation reveals a new form of scarcity. For the past two years, the industry has focused on advanced semiconductors and NVIDIA GPUs. However, many players are gradually discovering that the real bottleneck is access to energy.

While capital is available, securing several hundred megawatts connected to the grid within timelines compatible with the ambitions of AI players is becoming an increasing challenge. In some regions, connection delays can reach several years.

This reality is now evident in regulatory debates. Mississippi has become one of the first laboratories for this new energy economy of AI. The state has attracted several AWS projects representing over $13 billion in announced investments. To support this growth, operator Entergy is developing several new production capacities, including three gas plants totaling over 2,200 MW of installed power. The estimated cost of these infrastructures exceeds $3.8 billion.

A report published by Synapse Energy Economics estimates that residential consumers have already contributed approximately $38 million to the investments associated with the infrastructure needed to accommodate these new data centers, with an amount potentially reaching $74 million by the end of 2026. The authors particularly highlight the impossibility of precisely verifying the distribution of costs due to the confidentiality of contracts between large consumers and the electric operator.

Mississippi is not an isolated case. Several U.S. states are beginning to create specific rate classes for very large electricity consumers. Virginia, Ohio, Kansas, and Pennsylvania are working on mechanisms that impose long-term commitments, financial guarantees, or minimum consumption levels to prevent infrastructure costs from being transferred to other network users.

Europe has not yet opened this debate with the same intensity; however, similar tensions are gradually emerging.

France aims to become one of the leading European hubs for AI. Announcements of projects are multiplying around Mistral AI, Data4, OpCore, and consortiums supported by international investors. The country has a clear comparative advantage with a largely decarbonized and relatively abundant electricity production thanks to nuclear power, but this abundance is less evident when it comes to connecting several hundred megawatts in the same territory.

The stakes are not limited to data centers. The automotive industry is demanding more electricity for its gigafactories, while hydrogen producers seek to secure significant capacities. The decarbonization of heavy industry also requires massive electrification. Transportation and residential uses are following the same trajectory. For the first time in several decades, multiple public policies are converging on a single resource: electricity.

The question then becomes less technological than industrial, as each gigawatt allocated to an AI campus is a gigawatt that will not be immediately available for other uses. The debate over financing is gradually masking a new question: who should be prioritized in the allocation of electrical capacities?

This evolution also transforms the role of hyperscalers. AWS, Microsoft, Google, and Meta are no longer just technology companies. Their investment decisions now influence the energy strategies of entire regions. Like the large steel or automotive groups of the last century, they are becoming players capable of directing investments in networks and the economic development of territories.

Another player is emerging in this equation: infrastructure funds. Brookfield, BlackRock, KKR, Macquarie, Global Infrastructure Partners, and Gulf sovereign funds are investing heavily in data centers, electrical networks, energy infrastructures, and production capacities. AI creates a new market for patient capital; the infrastructures necessary for its development will be amortized over twenty, thirty, or forty years, while models evolve every six months.

This paradox is perhaps one of the most symptomatic of the current period: the fastest technology industry in history now depends on the slowest infrastructures to build.

The question "Who will pay for the gigawatts of artificial intelligence?" does not call for a single answer. Consumers, hyperscalers, network operators, infrastructure investors, and public authorities will all participate, in one way or another, in financing this new layer of infrastructure.

The real question is who will decide on their allocation? Which projects will be considered priorities? What place does Europe wish to give to AI compared to reindustrialization, transportation, or the energy transition?

For two years, the AI industry has been dominated by the battle of models. The next decade could be dominated by the battle for infrastructure. And in this new economy, the decisive factor may no longer be the quality of algorithms, but the ability to sustainably mobilize gigawatts.