Antimatter: Energy, a New Crucial Challenge for AI

For years, computing power has been seen as the main driver of artificial intelligence. However, in the era of inference, the true limiting factor is now access to abundant, sovereign, and available energy. This is highlighted by David Gurlé, co-founder and CEO of Antimatter, in a recent op-ed.

Three years ago, Elon Musk stated that GPUs were "much harder to obtain than drugs." Since then, the tech world has plunged even deeper into an obsession with raw computing power, making GPUs one of the most sought-after resources on the planet. Yet, the industry is focusing on the tree without seeing the forest looming ahead: as we shift from training models to inference, the limiting factor is no longer the chip, but the electricity that powers it.

Demand is Exploding, and This is Just the Beginning

A request to a LLM consumes up to ten times more electricity than a Google search. Decision-makers who spoke of FLOPS and gigabytes two years ago are now talking about megawatts. Technological sovereignty depends, to a large extent, on energy sovereignty. And this reality is becoming more pronounced as the geopolitical instability from the Middle East to the Ural Mountains plunges energy markets into structural uncertainty.

In the face of this, the traditional model of hyperscalers is showing its limits in real-time. The cloud, as it has been built so far, relies on enormous centralized data centers, a model structurally incapable of keeping pace with the demands of the inference era. The latest connections are being made at powers of 100 to 200 MW according to RTE, illustrating a race for power that they cannot win. Hyperscalers are facing three worsening problems that they cannot remedy with technical solutions:

• Construction timelines measured in years rather than months
• Infrastructure costs skyrocketing as hardware components face supply constraints
• Data centralization leading to latency, sovereignty risks, and systemic fragility in the event of crises related to networks, geopolitics, or climate

A New Paradigm is Emerging

Given the constraints of cost, scale, and energy consumption, AI infrastructure must become modular and distributed. Computing capacity must be brought closer to the primary energy source that powers it, as well as to the user who utilizes it. This is not an architectural preference. It is a structural necessity.

The solution lies in the vertical integration of three areas that have historically operated in isolation from one another:

• Energy infrastructures
• Physical hardware of data centers
• Cloud software

Merge these three elements, and you get what the AI era truly needs: infrastructure that can be deployed in months rather than years, powered by existing electricity rather than electrical grids that take a decade to build, with data sovereignty integrated from the outset rather than added later through regulation.

This is not just a theory we will explore. We have succeeded in integrating these three areas within a single company: Antimatter. This is the company we launched today. And we advise all tech players to consider this new paradigm that will revolutionize infrastructure, with or without the participation of European actors.

The Next Battle in the AI Sector

The next battle in the AI sector will not be won by the company with the most GPUs in the largest data center. It will be won by the company that identifies where available energy is located, builds resilient infrastructure nearby, and provides computing power to organizations that need sovereignty, resilience, and efficiency.

So stop counting computing power and GPUs. Start counting megawatts. And in France, this may be a real opportunity for us, thanks to the nuclear sector.

The decentralized neo-cloud is not just an option. It is a vital necessity. And it is already here. The question remains whether you want to participate in its construction…

David Gurlé is an entrepreneur and a recognized pioneer in the global technology industry. He helped create real-time communication tools at Microsoft and Skype, founded Perzo before selling it to a consortium of fourteen global banks for a valuation of $100 million, and turned Symphony into a secure communications platform valued at $1.8 billion used by nearly all major financial institutions worldwide. He is co-founder and CEO of Antimatter — the world's first full-stack neo-cloud, combining distributed computing, sovereign hardware infrastructure, and flexible energy management within an integrated platform.