Meta wants to sell its AI computing power to the world — and take on AWS doing it
Meta is preparing to launch a commercial cloud service that would sell access to its AI computing infrastructure — putting the social media giant in direct competition with AWS, Google Cloud, and Microsoft Azure. Bloomberg broke the story on July 1, 2026, confirming the division behind it — called Meta Compute — was quietly created back in January. For anyone who relies on cloud services, this is the biggest new entrant to the market in years.
The plan
Meta Compute is targeting two distinct revenue streams. First, it plans to sell access to Meta's own AI models — including Muse Spark — running on its own hardware. Second, it wants to rent out raw GPU capacity from its data centers so other companies can train and run their own AI systems. That dual approach mirrors what AWS did when Amazon turned its internal retail infrastructure into the world's largest cloud business.
The financial scale here is hard to overstate. Meta has committed $600 billion in US infrastructure investment through 2028, with 2026 capital expenditure alone guided at $115–135 billion. Zuckerberg flagged cloud ambitions publicly at the company's May 2026 shareholder meeting, framing it as a way to monetize capacity that exceeds Meta's own internal AI needs.
Winners, losers, and open questions
The immediate losers are the so-called neocloud providers — specialist GPU rental companies like CoreWeave and Nebius — that currently supply Meta and count it among their biggest customers. Shares in both dropped sharply on the news: CoreWeave fell around 10%, Nebius around 12%, per TechCrunch. The concern is straightforward: if Meta starts selling compute directly, it no longer needs to buy it from them.
For businesses currently paying AWS or Google Cloud prices, a new competitor with this kind of infrastructure could eventually mean better deals — but that's not guaranteed or imminent. Meta has not announced pricing, availability timelines, or any details on data residency or compliance (a gap that matters especially for European customers under GDPR and the EU AI Act).
What's still unclear
Zuckerberg's "excess capacity" framing deserves some skepticism. Meta has historically struggled to forecast its own AI compute demand accurately, and whether genuine surplus capacity exists — rather than capacity earmarked for near-future model training — remains unverified. Antitrust watchers will also note the parallels with Amazon's own cloud expansion: using dominant market revenue from one business (social advertising, in Meta's case) to subsidize entry into another.
Meta's existing AI features already reach billions of users across Facebook, Instagram, WhatsApp, and the Meta AI app. A cloud business would add a new revenue line aimed squarely at developers and enterprises — a very different customer from its current base.
