Meta is making its plans to become a public cloud provider more concrete. The company would reportedly rent out both computing capacity and access to AI models.
Will we soon see a new hyperscaler with Meta? According to Bloomberg, Meta is working on establishing a new cloud business to compete with AWS, Microsoft, and Google Cloud. Under the not-so-original name Meta Compute, Meta would offer both computing capacity and access to its most advanced AI models for a fee.
Meta’s move is as logical as it is surprising. In recent years, Meta has invested billions of dollars in its own data centers to run its AI models itself. While many companies are hitting the limits of their capacity, Meta appears to still have plenty of free servers available.
With more data centers in the states of Louisiana and Ohio—which Zuckerberg describes as being ‘the size of Manhattan’—about to open, Meta should normally not lack server capacity in the coming years. However, in times of need, even Meta must sometimes resort to creative solutions to create extra capacity.
Recouping investments
Why not rent out some excess server capacity to offset those billion-dollar investments? In its most recent quarterly results, Meta stated it will invest up to $189 billion in AI infrastructure, which includes the development of chips alongside data centers. AI models and services have not yet proven profitable, so Meta has to find the money elsewhere.
That is why Meta would also offer paid access to its most advanced AI models through its brand-new cloud division. This, in turn, is a striking change of course. On paper, Meta offers its Llama models as open source, although the open-source community argues the licenses are not free of restrictions. With the introduction of Muse Spark, Meta had already moved away from its open-source policy.
SpaceX also previously announced its intention to rent out unused server capacity. The public cloud and AI are inextricably linked. AI companies are becoming cloud providers and vice versa. One thing is certain: warnings about overinvestment in AI are currently being completely ignored.