Meta Platforms is developing a cloud infrastructure business to sell access to AI computing power and AI models to external customers, the company confirmed in early July 2026 [1, 2, 3, 4, 5]. This business aims to compete directly with major cloud providers Amazon Web Services, Microsoft Azure, and Google Cloud [1, 2, 3, 4, 5].
The offering could include access to hosted AI models like Meta’s Muse Spark, similar to Amazon’s Bedrock service, alongside "raw" or "bare-metal" computing capacity sold directly to businesses, resembling neocloud providers such as CoreWeave [1, 2, 3, 4, 5, 6]. Meta’s internal initiative behind the plan is named Meta Compute and is led by executives Santosh Janardhan, Daniel Gross, and Dina Powell McCormick [1, 2, 4, 5].
Meta has invested hundreds of billions of dollars in building data centers, AI infrastructure, and AI chips. Its 2026 capital expenditure guidance rose sharply to $135-145 billion, up from $69.6 billion in 2025 and $37.2 billion in 2024 [1, 2, 3, 7, 4, 8, 5, 6, 9]. This scale of investment reflects the company’s significant AI compute buildup, including ongoing data center projects in Louisiana and Ohio [3].
On July 1, 2026, news of Meta’s cloud business plans pushed its share price up 8-9%, closing at $612.91, as investors welcomed the prospect of monetizing excess AI compute [1, 2, 7, 4, 8, 5]. CNBC’s Jim Cramer said, "Until today, our feeling was, what the heck is Meta doing? Now, they're going to use that [compute] power to offer a profitable enterprise to their customers" [8]. Post Oak Group’s Karan Ramchandani called it "a no-brainer to compete in the market, to sell compute power to other B2B players" [9].
Meta plans to sell older generation hardware, such as H series GPUs, to outside customers while continuing to acquire newer high-end chips for its own AI research efforts [6]. This approach aims to generate revenue from unused data center capacity and address investor concerns about the company’s high AI spending [1, 3, 7, 8, 9].
However, some market participants view Meta’s move as a sign of AI compute oversupply that has prompted a sharp decline in Chinese tech and semiconductor stocks [10, 11]. Wang Chen of Xufang Investment Partners commented, "Meta's news indicates a tendency toward compute oversupply; there have been concerns that AI capital expenditures might fall short of expectations at some point" [10]. Yet other industry insiders argue this interpretation is mistaken and see the cloud business as a sign of maturation in AI infrastructure business models [11].
Meta follows SpaceX’s xAI, which recently began selling excess AI compute capacity through deals with companies like Anthropic and Google, with monthly spend estimated at $1.25 billion and $920 million respectively [3, 7]. In mid-to-late June, Meta signed agreements with Crusoe to supply about 1.6 GW of AI compute capacity from its Texas and Missouri data centers [6].
Meta’s cloud business launch is expected to continue developing throughout 2026 as it deploys its capital spending plan, expands data center operations, and looks to grow external demand for its AI compute and model services.
