META IS BUILDIND A CLOUD BUSINESS JUST TO SELL ITS EXCESS COMPUTING CAPACITY

Meta Platforms is building a cloud business to sell its surplus AI computing capacity, a move that would pit it directly against Amazon, Microsoft, and Alphabet in the infrastructure market. The plans are still in development and could change, but the market reacted immediately: Meta shares rose more than 10%, while CoreWeave and Nebius shares fell 10.8% and 12.4%, respectively, on the fear that Meta may no longer need their services.

At Meta’s shareholder meeting in May, Mark Zuckerberg said entering cloud computing was "definitely on the table" and that companies approach Meta almost every week to buy access to AI models. Now the company appears to be moving toward turning that demand into a revenue stream.

you may like

MICROSOFT TURNS TO AWS FOR GITHUB CAPACITY AFTER AI SURGE

EU LAUNCHES TECH SOVEREIGNTY PACKAGE, CUTS ON EU AND CHINESE TECH

MICROSOFT CUTS 3200 XBOX JOBS, CEO SAYS BUSINESS NOT HEALTHY

THE SHAPE OF THE PLAN

According to the reported details, the cloud service Meta is planning would let developers access AI models hosted on Meta's infrastructure, including Muse Spark, and pay for computing power. That model is similar to Amazon Web Services' Bedrock, which offers foundation models from multiple providers on top of AWS's compute. Meta is also considering selling raw AI computing capacity much like neoclouds such as CoreWeave and Nebius do. The dual-track approach would let Meta serve both developers who want managed models and those who want raw GPU cycles.

Meta declined to comment on the report. Reuters could not independently verify it, but the pattern aligns with what the industry has seen before. Elon Musk's SpaceX recently struck deals to rent out access to its Memphis data centre to Anthropic and Google, offering a proof point that large infrastructure owners can profit from spare capacity.

WHY THIS MATTERS FOR META

The calculus for Meta is straightforward. The company is projected to spend as much as $145 billion on AI infrastructure this year, a figure that sits inside the $700 billion total Big Tech outlay on AI technology. That scale of build-out produces an unavoidable quantity of idle compute when training and inference cycles fluctuate. Selling excess capacity turns a fixed cost into a variable revenue stream, and at Meta's scale even a small utilisation improvement moves the needle.

There is a strategic dimension too. Reducing dependence on advertising revenue has been a long-running goal at Meta. A cloud business would add a second major revenue leg, even if it takes years to reach material size. Zuckerberg's shareholder meeting comments confirm the company has been weighing the option seriously.

THE NEOCLOUD SHOCK

The sharpest market reaction came from CoreWeave and Nebius, and for good reason. The move is more likely to impact neoclouds than big hyperscalers, because CoreWeave and Nebius rely on Meta for a meaningful portion of their growth. If Meta no longer needs to rent their capacity, the neoclouds lose a major customer. If Meta starts selling its own capacity, it becomes a direct competitor for the same mid-market AI developers.

The numbers tell the story. CoreWeave dropped 10.8%, and Nebius fell 12.4% in a single session. That is the market pricing in a structurally weaker competitive position for both companies. The big three hyperscalers, Amazon, Microsoft, and Alphabet, will feel less direct pain because their scale and existing customer relationships provide a moat that Meta will have to work years to cross.

THE CATCH: MUSE SPARK

The specific AI model Meta plans to anchor its cloud service with is Muse Spark, unveiled in April. It is the first model from a costly team Meta assembled, but it has yet to be released and has no scheduled launch date. That gap between announcement and availability is a real risk for a cloud service that needs to offer production-ready capabilities immediately.

Analysts said the move deepened doubts about Meta's efforts to catch up with leading AI labs such as Anthropic. If Meta's flagship model is still unreleased while Anthropic and OpenAI are shipping regularly, the cloud service might launch with a compelling compute backend but a weak model portfolio. The Bedrock analogy cuts both ways: AWS Bedrock works because it aggregates many models. Meta would need to either open its platform to competitors' models or rely solely on its own, which would narrow its appeal.

WHAT COMES NEXT

The cloud business is still being built, and the final strategy may look different from the current shape. Three things will determine its success. First, whether Meta can ship Muse Spark and follow-on models on a competitive cadence. Second, whether the $145 billion infrastructure spend yields capacity that is truly surplus and not needed for Meta's own products like Facebook, Instagram, and Llama. Third, whether the neoclouds can differentiate themselves on service, specialisation, or pricing fast enough to keep Meta from eating their lunch.

what to read next

NOTEBOOKLM NOW GEMINI NOTEBOOK, ADDS CLOUD CODE EXECUTION

AMAZON'S LEO SATELLITE INTERNET TO LAUNCH LATER THIS YEAR

NICE GUY MICROSOFT EXTENDS WINDOWS 10 SUPPORT FOR ANOTHER YEAR

For now, the market sees a new entrant with massive capital, an existing developer ecosystem, and a CEO who has publicly signalled intent. That combination is why Meta shares moved up while the neoclouds bled. The narrative has shifted from Meta as a buyer of cloud capacity to Meta as a seller, and the competitive dynamics of AI infrastructure just got a lot more interesting.