Meta’s AI Push Moves From Models to Megawatts

As artificial intelligence models grow larger, more energy-hungry, and harder to scale, the competitive edge is shifting away from algorithms alone. Increasingly, it is about who controls compute at scale. Meta’s latest announcement makes that shift explicit.

On January 13, Mark Zuckerberg unveiled Meta Compute, a top-level initiative designed to build the large-scale computing infrastructure that will underpin the company’s long-term AI ambitions. The plan is unapologetically ambitious: Meta intends to build “tens of gigawatts” of computing capacity this decade, with the potential to scale to “hundreds of gigawatts or more over time.”

In Zuckerberg’s framing, infrastructure is no longer a backend function. It is strategy.

“How we engineer, invest, and partner to build this infrastructure will become a strategic advantage,” Zuckerberg wrote, positioning compute as a foundational pillar alongside software and AI model development.

From Models to Megawatts: Why Compute Now Matters

Meta’s move comes at a moment when AI’s growth is colliding with physical constraints. Training and running large-scale AI systems demand enormous power, specialised silicon, and globally distributed data centres. Compute availability is rapidly becoming a gating factor for innovation.

By elevating Meta Compute to a standalone initiative, the company is signalling that long-term AI leadership will depend as much on capacity planning and energy security as on breakthroughs in model architecture.

This is not a short-term infrastructure refresh. Meta is planning for decades, treating compute the way hyperscalers once treated cloud: as a platform advantage that compounds over time.

Who’s Building Meta Compute

Leadership of Meta Compute reflects the initiative’s scope, spanning engineering, energy, geopolitics, and long-term capital planning.

• Santosh Janardhan, Head of Global Infrastructure and Co-Head of Engineering, Meta, will continue overseeing technical architecture, software systems, silicon efforts, developer productivity, and Meta’s global data centre and network operations.

• Daniel Gross, who joined Meta last year and previously co-founded Safe Superintelligence alongside former OpenAI Chief Scientist Ilya Sutskever, will lead a newly formed group focused on long-term capacity planning, supplier partnerships, industry analysis, planning, and business modelling.

• Dina Powell McCormick, President and Vice Chairman, Meta, will work with governments and sovereign partners to help build, deploy, invest in, and finance Meta’s infrastructure globally.

The structure underscores that Meta Compute is not only an engineering programme but also a coordination effort across policy, finance, and global energy markets.

Energy Becomes the Silent AI Constraint

One of the clearest signals behind Meta Compute is Meta’s growing focus on energy availability. The company has announced agreements linked to nuclear energy projects capable of supporting up to 6.6 GW of electricity capacity in the United States by 2035.

These agreements span extended operations at existing nuclear plants, development of advanced nuclear reactors, and long-term energy procurement. According to Meta, the electricity will be supplied into regional grids that support its operations, including its Prometheus AI supercluster in New Albany, Ohio.

Meta has emphasised that it pays the full cost of the energy consumed by its data centres, while the projects themselves are expected to generate thousands of construction jobs and hundreds of long-term operational roles, particularly in Ohio and Pennsylvania.

Support for advanced nuclear developers such as TerraPower and Oklo, alongside power purchases from facilities owned by Vistra, points to how seriously Meta is treating power stability as a prerequisite for AI scale.

Compute as Competitive Moat

Meta’s infrastructure push also reflects intensifying competition across Big Tech. AI-ready cloud environments are becoming the next battleground, with peers aggressively expanding capacity and securing long-term energy sources.

Meta CFO Susan Li had previously framed this direction during an earnings call, stating:
“We expect that developing leading AI infrastructure will be a core advantage in developing the best AI models and product experiences.”

Meta Compute operationalises that thesis. Rather than relying entirely on external capacity or short-term expansion, the company is building a deeply integrated compute stack, spanning silicon, data centres, networks, and energy procurement.

For enterprises watching closely, the message is clear: AI scale is no longer just about access to models. It is about ownership of the physical layers beneath them.

Meta’s announcement reinforces a broader industry reality. As AI adoption accelerates, compute scarcity and energy availability could reshape pricing, partnerships, and innovation timelines.

For startups, this may widen the gap between those with access to hyperscaler infrastructure and those without. For governments, it raises new questions around energy policy, grid resilience, and sovereign partnerships. And for Meta, it marks a decisive bet that infrastructure control will determine who leads the next phase of AI.

As Zuckerberg put it, the goal is to “deliver personal superintelligence to billions of people around the world.” Meta Compute is the foundation on which that vision now rests.