Meta is weighing in on a new line of business: renting out its AI computing power. During the company’s second-quarter earnings call, CEO Mark Zuckerberg said Meta has fielded offers from companies keen to rent its cloud software at a “meaningful premium” over what the tech firm paid to build that compute. That is a rare acknowledgment from a company that has spent years insisting its AI infrastructure exists to serve its own apps.   Meta’s capital expenditure is guided at $130 billion to $145 billion for 2026, per Q2 earnings. Free cash flow dropped sharply in its latest quarter as the company builds out data centers, AI chips, and tech developments like smart glasses, ADWEEK reported earlier. Renting out unused cloud capacity could help Meta recoup some of that spending.  Renting out compute would also open Meta to an entirely new customer base of AI frontier labs and scaling tech startups, businesses that won't necessarily have a history with Meta. But turning a consumer-facing business into a cloud business has its hurdles, and it's no secret that AI ambitions are racking up astronomical bills. Meta declined to comment beyond what the executives disclosed during earnings calls, and did not address questions about operational or financial details. “Meta would have a challenge, being a large company coming into a new innovative market, where some of the smaller innovative companies coming out from universities might be more agile and more innovative than Meta,” said Eric Newcomer, CTO & principal analyst, Intellyx. Analysts point to three challenges Meta faces if it wants to sell a service that’s genuinely competitive rather than a narrative point to justify high AI capex to skittish investors. Lacking an enterprise sales machine   Meta knows how to sell advertising to millions of businesses. Selling infrastructure to those businesses is a different proposition. The tech firm is lacking an "enterprise B2B engine,” according to Forrester’s principal analyst Naveen Chhabra.  “A true cloud computing provider requires enterprise-grade sales teams, global customer support pipelines, rigid service level agreements (SLAs), and complex billing systems,” Chhabra said. While these are not insurmountable, they require near-term focus.  Meta is unlikely to compete in the traditional cloud market against rivals like Amazon Web Services, Microsoft Azure, and Google Gemini, Newcomer said. "That competition is over. You've already got the winners, the incumbents, the dominant players." Instead, Meta is more likely to focus on becoming a neocloud provider, built specifically for the demands of AI. These providers "are still learning some of these enterprise requirements,” Gartner’s distinguished VP analyst Ed Anderson said.  Zuckerberg told Bloomberg in July that the company is developing plans to sell access to “raw” computing capacity, a model closer to neocloud business. A trust deficit with data privacy That enterprise-sales problem is further complicated by Meta's track record on data privacy, including how it collects, uses, and shares user data across Facebook, Instagram, and its AI systems. “[Meta] hasn't always had the best track record when it comes to data privacy," said Dave McCarthy, group vice president, cloud and datacenter infrastructure, IDC. "Public concerns around how data is being used is going to be an inhibitor for customers to want to jump right in." Chhabra described enterprise sensitivity around data privacy as a “massive psychological hurdle for corporates,” particularly for a company whose most visible products are consumer social networks. The economics of AI compute are rising  Then there is the problem that affects every neocloud provider: AI compute is getting more expensive. And while Meta's cloud ambitions could bring in revenue, the road to get there won't be cheap. Many neocloud providers today are built around a single type of accelerator, typically Nvidia's GPUs. But as the cost of running AI workloads rises, companies are looking for ways to route the workloads across different chips depending on the task.  Whether Meta can ultimately drive down the cost of running AI workloads would determine whether its infrastructure business becomes economically compelling, according to Newcomer. “There's a cost factor of AI that's a challenge,” he said. “People are starting to get sensitive to costs. If anyone could deliver a neocloud with cost optimization for the applications that would drive the price down compared to the foundation models, that could be an opportunity.” Ultimately, Chhabra cautioned, the economics of it all could be difficult to make it work. “Financial models tracking the tech industry indicate that building public cloud infrastructure from scratch yields poor initial returns,” he said.