Mark Zuckerberg didn’t mince words. Selling AI computing power “makes sense.”
The Meta CEO’s blunt assessment, delivered this week, confirms months of speculation. The social media giant is building plans for a cloud-style business. It would sell excess capacity from its ballooning AI infrastructure to outside customers. And it would do so while still one of the largest buyers from the very startups it now threatens to disrupt.
This marks a sharp turn. For years Meta poured cash into data centers and chips almost exclusively to feed its own models. Advertising still delivers nearly all its profit. Yet the scale of spending has grown so vast that leaving silicon idle no longer adds up. Analysts project Meta’s capital expenditures could top $145 billion in 2026 alone, according to a Forbes report. Much of that builds the very compute now eyed for resale.
From Cost Center to Revenue Stream
The logic appears straightforward on paper. Meta has committed more than $180 billion to AI infrastructure in recent filings. It operates massive projects in Louisiana and Ohio. One site, dubbed Hyperion, could eventually reach five gigawatts. That is enough power to rival entire cities. Zuckerberg himself described plans for “tens of gigawatts this decade, and hundreds of gigawatts or more over time,” as reported in a January Data Center Dynamics article.
Idle capacity wastes money. Renting it out turns a colossal expense into income. Bloomberg first detailed the internal project on July 1, revealing Meta’s work on what some call “Meta Compute.” The company would offer raw processing power. It might also host its own AI models, much like Amazon Web Services does with Bedrock. One model under consideration: the closed-weight Muse Spark. Reuters picked up the story the same day, noting developers could access those models and pay for the underlying compute.
But execution won’t come easy. Cloud operations demand more than hardware. They require service level agreements, customer support, and uptime guarantees. Meta has never run such a business at scale. Its advertising machine prints high-margin cash. Cloud services typically deliver thinner returns. CNBC highlighted Wall Street’s mixed reaction. Investors cheered the diversification potential even as they weighed those lower margins.
Shares of Meta jumped roughly 9% after the initial reports. Meanwhile CoreWeave, Nebius, and IREN each dropped double digits. The irony runs deep. Meta has signed enormous contracts with these neocloud providers. One deal with CoreWeave alone reached $35 billion, according to The Next Web’s coverage on July 9. The social giant buys the very capacity it now intends to resell. That dependency won’t vanish overnight.
Zuckerberg planted seeds for this shift earlier. At Meta’s May shareholder meeting he called a cloud business “definitely on the table.” Companies approach the firm “almost every week” seeking access to its models or spare compute, he said then. CNBC captured those remarks. Seven months before that, on an earnings call, he noted similar inbound interest. The message has stayed consistent. Excess capacity should not sit unused.
Meta’s technical foundation gives it advantages. The company designs its own MTIA chips and placed them into production last year. It extended a key deal with Broadcom through 2029. These moves grant tighter control over the stack than many rivals enjoy. Yet building a customer-facing service from scratch introduces new risks. Neocloud specialists have honed sales, billing, and support for years. Meta starts closer to the beginning on those fronts.
And the market keeps growing. Goldman Sachs once projected the AI cloud sector could hit $2 trillion by 2030. That figure draws new entrants. It also intensifies pressure on incumbents. Amazon, Microsoft, and Google already dominate. A fourth hyperscaler with Meta’s balance sheet and custom silicon would reshape the math. Smaller players face immediate pressure. Larger ones must defend their turf.
Recent moves show Meta’s ambition hasn’t slowed. The firm announced a nearly $13 billion AI data center investment in Alberta, Canada, according to investor chatter on X this week. That project targets gigawatt-scale capacity. It directly counters any notion that Meta plans to dial back spending. Instead the company appears to double down while searching for ways to offset costs.
Its broader AI push extends beyond infrastructure. Consumer chatbots, enterprise agents, and API access to models like Muse Spark 1.1 all aim to generate fresh revenue. Pricing on the new model adopts an aggressive stance to grab share from higher-margin rivals such as OpenAI and Anthropic. Chartlyst noted the strategy on X, pointing to Meta’s shift from indirect ad improvements toward direct commercialization.
Still, questions linger about returns. Meta’s advertising business improved thanks to better targeting fueled by AI. Yet 98% of revenue still comes from ads. Diversification matters. So does proving that hundreds of billions in capex deliver more than internal gains. Selling compute offers one path. Whether customers bite at scale remains unproven.
Zuckerberg’s comments this week remove ambiguity. The exploration has moved beyond internal debate. Teams are now shaping the offering. Details on launch timing, exact capabilities, and target customers stay under wraps. But the direction is set. A company sitting on tens of gigawatts of AI silicon has powerful reasons to make that capacity pay.
Wall Street will watch closely. So will the cloud incumbents and the nimble neoclouds that suddenly find a new giant in their rearview mirror. Meta’s bet carries risks. Its potential rewards could reshape not just its own financials but the entire market for AI infrastructure. The era of hyperscalers renting to one another has arrived. And Zuckerberg just signaled Meta wants in.
WebProNews is an iEntry Publication