With Little AI Adoption, Meta Eyes Selling Inference at a Premium

For AI practitioners, Meta's problem is product-market fit: its models have little external use despite massive frontier research spend. Selling inference and training capacity at a premium could give developers another path to Meta infrastructure, but it may not fix model adoption.

Meta Platforms enters late August 2026 at a crucial crossroads. The stock is trading below $600, even as artificial-intelligence enthusiasm reaccelerated across the broader market this month, according to a Motley Fool analysis published August 16. The disconnect, the author argues, stems from the company's enormous capital spending and the market's inability to see a near-term return. In the most recent quarter, Meta spent $30 billion on capital expenditures, nearly doubling spending year over year. Management has guided to $130 billion to $145 billion in full-year 2026 capex, with the majority directed toward artificial-intelligence data centers, frontier model research, and inference capacity. Some of that spend supports the advertising platform, but the bulk is going toward compute that does not yet have a clear external revenue stream.

“

Management has guided to $130 billion to $145 billion in full-year 2026 capex, with the majority directed toward artificial-intelligence data centers, frontier model research, and inference capacity.

That is where a potential AI cloud business comes in. CEO Mark Zuckerberg has said other AI companies have reached out to Meta to buy access to its compute capacity at a premium over the purchase price. The size and pricing of those arrangements remain unclear, but the inquiries suggest Meta is sitting on a scarce resource: large-scale, already-built GPU and data center capacity. If Meta were to sell excess inference and training capacity to third parties, it could begin monetizing capital investments that are currently underutilized. The Motley Fool piece likens the concept to recent deals signed by SpaceX, though the article does not provide Meta-specific financial details and the comparison is incomplete. The broader strategic logic is the same used by Amazon when it turned internal infrastructure into AWS, though Meta would be entering a market already dominated by AWS, Microsoft Azure, and Google Cloud.

Near term, the financial picture is less comfortable. Meta's AI models currently see little external use, so the company is building more capacity than its own software services require. That excess capacity adds depreciation, energy, and staffing costs without offsetting revenue. The Motley Fool analysis expects earnings to decline in the near future, even as the stock appears statistically cheap. Investors must weigh a cheap equity multiple against the possibility that capital intensity will continue rising while adoption of Meta's AI tools remains limited. The company has failed to see material AI adoption across most of its consumer and business software, a problem the potential cloud offering may only partially solve because it competes with established hyperscalers that already have deep sales relationships and developer ecosystems.

The market reaction tells a story about patience. Meta did not join the August AI rally, indicating investors are increasingly focused on cash generation and return on invested capital rather than raw AI ambition. This is a shift from prior AI-driven equity narratives. If Meta can sign even a few meaningful compute deals, sentiment could improve, but the terms and scale will matter. Selling capacity at a premium could offset some depreciation and validate the company's infrastructure buildout. Selling at commodity pricing, however, would underscore the risk that AI compute is becoming abundant.