Meta reportedly building a cloud compute business — but is it a sign of AI strategy drift?

Meta's Cloud Play Signals Trouble for Its AI Strategy

Meta is building a cloud computing business to monetize excess AI infrastructure capacity—and the move reads less as strategic expansion than as an admission that the company's flagship AI ambitions lack near-term commercial traction.

The plan, reported by Bloomberg, would let Meta sell access to its AI models and raw computing power to external customers, competing directly with AWS and Google Cloud. On the surface, this is sound business: Meta has spent hundreds of billions on data centers and chips. Idle capacity generates no return. Selling it does.

But the real story sits underneath. Meta's stated goal is personal super intelligence—AI agents that live on Meta's own products and drive engagement. The company has built massive compute infrastructure specifically to power that vision. Turning around now to sell that capacity to rivals and startups signals Meta doesn't have a clear path to using it internally anytime soon.

The product gap

Meta has tried to integrate AI into its apps. Muse and Spark are competent models but haven't gained traction. Meta Vibes was a Midjourney wrapper. Instagram AI features remain generic and disconnected from user data. None of it feels like a killer application.

The clearest failure is what Meta hasn't attempted. Instagram creators have asked for personalized analytics powered by AI—which models to post more of, what drives conversion. Meta has granular data on every piece of content. The company could wire its AI models directly to that data and deliver exactly what creators need. It hasn't. Instead, when prompted for creator advice, Meta AI returns generic blog-post recommendations pulled from social media management platforms.

There's also agentic shopping—looking at a pair of shoes through Ray-Bans and saying "order me those." Meta has the glasses, the AI, and the infrastructure. It hasn't shipped even a one-click checkout experience powered by AI.

The pattern suggests not that Meta can't build these things, but that building them isn't a priority. That matters because it undermines the justification for the compute spend itself.

Market reaction tells the story

Neo cloud companies are selling off. Meta has existing contracts with providers like CoreWeave and Nebius worth tens of billions of dollars. Now Meta is competing with them. The company's framing—that it's simply getting ROI on excess capacity in the meantime—is reasonable. But the market reads it differently: if Meta has excess compute to sell, it either overbuilt or doesn't have products to consume what it built.

One counter-argument circulating among investors: Meta might be forced to do this. Google told Meta in March it was using too much compute and cut allocations. Meta then signed massive contracts with neo cloud providers. Selling capacity now might be a bridge while internal products scale up. But that's a generous read, and it rests on future products Meta hasn't shown.

The VR parallel

The precedent at Meta is not encouraging. The company invested heavily in the metaverse and VR. It eventually pivoted, and the only residual consumer win was Ray-Ban smart glasses—a narrow, hardware-focused product that barely resembles the original vision. If the super intelligence play follows the same trajectory, the cloud business might be the thing that actually survives while personal AI agents fade into the roadmap.

The unforced error

What sharpens the concern is timing. SpaceX handled this differently. When it built relationships with Anthropic and Google, it announced those deals as wins on the way into the public markets—concrete evidence that the compute business works. Meta is leaking plans. The company is sitting in limbo while speculation spreads that it either doesn't believe in its own AI products or has given up on making them work.

Meta doesn't move quickly on product experimentation. The company lacks the organizational momentum to productize AI features at the speed required to absorb the infrastructure it's built. Waiting for the obvious killer feature, as the Apple comparison suggests, might be the right long-term play. But it leaves Meta in the awkward position of spending like it has urgent AI applications while building like it doesn't.