For more than fifteen years, Meta Platforms built one of the most massive IT infrastructures in the world without ever seeking to commercialize it. Unlike Amazon with AWS, Microsoft with Azure or Alphabet with Google Cloud, Meta used its data centers only to power Facebook, Instagram, WhatsApp and its advertising machine.
But artificial intelligence is changing this historical logic, during the group’s annual general meeting yesterday, Mark Zuckerberg recognized that Meta could ultimately enter the cloud computing market if its massive investments in AI infrastructure generated excess capacity. A hypothesis still presented as secondary, but which reveals that Meta is starting to consider its infrastructure no longer just as a strategic cost, but as a monetizable asset.
The change would be major, Meta has also raised its forecasts for investment spending linked to artificial intelligence to between 125 and 145 billion dollars for 2026. Amounts which now bring the company closer to the major computing manufacturers.
For several months, Silicon Valley has witnessed a rush for GPUs and computing capacities. Generative AI models require gigantic volumes of training and inference, while laboratories seek to secure computing capabilities that have become scarce. In this context, data centers are becoming strategic economic assets.
The Meta manager also confided that companies were already approaching the group to buy computing or rent certain capacities.
A situation reminiscent of the beginnings of AWS, Amazon had built an internal infrastructure intended to absorb peaks of activity in its e-commerce activity. Faced with the growing needs of developers and startups, the group had gradually begun to rent out its excess resources. What was initially just an internal tool has become one of the most profitable businesses in tech history.
A situation which occurs even though the AI cloud market no longer resembles the traditional cloud of the 2010s. Needs are exploding, infrastructures are enormously more expensive and energy constraints are becoming critical. Excluding building an AI cluster today involves tens of thousands of GPUs, electrical capacities comparable to those of industrial sites, advanced cooling systems, supply chains dependent on NVIDIA, and investments of several tens of billions of dollars, and few players bring together these capacities. Especially since in this new economy, the main risk for Meta becomes that of underutilization of infrastructure.
Large AI labs are currently building gigantic capacities to avoid any future shortages. But usage cycles remain extremely variable. Massive training phases alternate with periods of lower load. The risk of overcapacity therefore becomes structural.
This is precisely where the cloud appears as a strategic shock absorber. By opening up the possibility of renting part of its infrastructure, Meta could develop a new narrative justifying its AI spending to investors.
Especially since the latest signals show a gradual convergence towards the economic models of the rest of the AI industry. Meta will test paid subscriptions for Meta AI, with offerings ranging from $7.99 to $19.99 per month in several pilot markets. Zuckerberg also mentions premium versions requiring more computing.
Meta now seems to consider that computing power can become a business in its own right. And behind this evolution lies a broader transformation of the technology industry: the digital giants are gradually ceasing to be solely software platforms to become industrial operators of very large-scale calculations.
