Powering AI Requires More Energy; Meta Compute Has the Task of Solving the Problem

The race for artificial intelligence is no longer measured solely in teraflops or parameters of language models, but in gigawatts.

In a strategic move signaling a radical shift in corporate priorities, Facebook’s parent company announced the creation of Meta Compute, a new division wholly dedicated to the supervision, planning, and operational management of its growing fleet of data centers for AI.

Mark Zuckerberg’s stated objective is as ambitious as it is complex: to build infrastructures capable of managing tens of gigawatts by this decade, aiming for hundreds of gigawatts in the long term.

The AI race isn’t won by GPUs and memory alone

In a post published on Monday on Threads, the Meta CEO highlighted how engineering, investments, and partnerships necessary to build this infrastructure will become a crucial strategic advantage for the company.

This statement outlines Meta’s transformation into an entity that will have to manage energy resources comparable to those of entire nations to power the dream of the “personal superintelligence“.

A new leadership for a global challenge

The birth of Meta Compute comes alongside a significant reorganization at the top of the company. A few hours before the announcement, Zuckerberg revealed the arrival of Dina Powell McCormick in the role of President and Vice Chairman.

With a sixteen-year tenure at Goldman Sachs and a role as an adviser to the Trump administration, McCormick brings a high-profile financial and geopolitical experience. Her task will be crucial: to collaborate with governments and sovereign funds to build, finance and deploy Meta’s infrastructure, a necessary step when it comes to energy projects of such scale.

The operational leadership of Meta Compute will instead be entrusted to a duo of executives. Santosh Janardhan, already at the helm of global infrastructure, will maintain control over the technical architecture, the software stack, the development program for proprietary chips, and the physical management of the data-center fleet.

Next to him will work Daniel Gross, who joined Meta’s Superintelligence team in mid-2025, which will lead a new group focused on the long-term capacity strategy, supplier partnerships and industrial analysis.

This dual-headed structure, supported by McCormick’s diplomatic and financial capabilities, suggests that Meta sees the infrastructural challenge as much a logistical problem as a geopolitical puzzle.

Monumental investments and course changes

The initiative sits within a context of unprecedented capital expenditures. For fiscal year 2025, Meta has forecast investments of $72 billion, with the prospect of further increasing spending in the following year.

However, this financial “generosity” must contend with recent results not exactly stellar. The tepid reception of open-source Llama 4, combined with a fierce talent war with OpenAI and the departure of machine-learning guru Yann LeCun, has forced the company to rethink its strategy.

Although Meta continues to release open-source tools like the Segment Anything (SAM) series, rumors suggest a shift in direction for core models. Zuckerberg would have abandoned further development of Llama to focus on proprietary models, developed under code names “Avocado” and “Mango”.

These projects, shrouded in mystery, represent Meta’s new attempt to not lose ground in the race toward next-generation generative AI, despite the difficulties faced and the criticism received.

The nuclear gamble

To prevent this infrastructural expansion from hitting the physical limits of the electrical grid, Meta is looking decisively at nuclear energy.

The company currently has gigawatt-scale data center projects in Ohio, Louisiana and Texas in the works, but the real challenge is powering them. To that end, last week Meta signed three new long-term contracts with TerraPower, Oklo, and Vistra for nuclear energy supply.

These agreements, added to the previous commitments with Constellation Energy, bring the social network’s contracted atomic energy portfolio to about 6.6 gigawatts.

This move highlights how the future of artificial intelligence is inextricably linked to the ability to secure stable, clean and massive energy sources, transforming Big Tech into some of the leading players in the global energy market.