Version 2

Meta is building a cloud business to sell excess AI compute capacity and hosted AI models to outside customers, per a Bloomberg report on July 1, 2026, which sent META shares up approximately 10% while CoreWeave fell ~12.7% and Nebius fell ~12.4%. [1][3] The planned service includes per-token access to Llama and Muse Spark models and raw GPU compute rental, structured like AWS Bedrock. [1] SemiAnalysis reported on July 2 that Meta contracted over 5GW of external cloud and colo capacity in the first half of 2026 alone and is in final talks with Anthropic for private Claude instances — a scale of internal demand that directly complicates the 'excess capacity' framing. [6] Whether Meta's internal compute needs will leave a meaningful surplus to sell, and whether it can build enterprise sales infrastructure from scratch, remain the core unresolved questions.

Meta would be the first major US social media company in enterprise cloud, converting $125-145 billion in planned capex into a direct B2B revenue line. Wells Fargo models $20 billion in revenue per gigawatt at 85% operating margin, but SemiAnalysis argues Meta will pursue premium SpaceX-style contracts rather than commodity compute resale — a model with better margins but a narrower addressable market. [6][2] The split market reaction (META up, neoclouds down) and Milk Road AI's counter-argument that Meta's implied $20B/GW pricing validates CoreWeave and Nebius at twice their current valuations point to a genuine unresolved debate about who benefits from Meta's entry. [7]

On July 1, 2026, Bloomberg reported that Meta is building a cloud business to sell excess AI compute capacity to outside developers and enterprises. The planned offering resembles AWS Bedrock — per-token access to Meta's Llama and Muse Spark models alongside raw GPU compute rental — with Meta charging developers directly for infrastructure built primarily for its own AI systems. [1][2] Meta shares rose approximately 10% on the announcement. CoreWeave fell approximately 10-13% on the day, with options data showing $2.3 million in 2026 calls sold into the decline; Nebius fell approximately 12.4%. [1][3] Mark Zuckerberg had signaled the direction in May 2026, noting that outside companies ask Meta for compute almost every week, and a January 2026 infrastructure initiative had established the groundwork. [4][5][1]

SemiAnalysis reported on July 2 that the prevailing 'excess capacity' framing may misread the situation. Meta contracted over 5GW of external cloud and colo capacity in the first six months of 2026 — not counting its own self-build activity — and is in final talks with Anthropic to access private instances of Claude, analogous to Bedrock and Vertex offerings from other hyperscalers. [6] Meta also plans to scale its advertising recommendation systems by more than 10x in complexity, and Meta Superintelligence Labs continues to receive the bulk of incremental compute. SemiAnalysis argues Meta will pursue SpaceX-style premium on-demand compute contracts — potentially including a deal with Anthropic valued at $10 billion or more — and that Meta will not operate as a commodity IaaS vendor at 30% gross margins: 'all its options are high value.' [6] If internal demand grows as fast as capacity, the surplus available for resale may be structurally narrower than the financial models assume.

The neocloud market reaction on July 1 reflected the conventional competitive-threat read. CoreWeave, to which Meta has committed more than $35 billion across take-or-pay contracts, and Nebius both fell sharply despite Meta's deep financial ties to both. [7][3] Milk Road AI countered on July 2 that Meta's cloud entry implies compute is valued at $20 billion per gigawatt, while CoreWeave and Nebius trade at roughly $10 billion per gigawatt — meaning Meta's move validates the incumbents' pricing rather than undercutting it. [7] Nebius's own fundamentals support a growth narrative independent of Meta: $399 million in Q1 2026 revenue (up 684% year-over-year), $46 billion in contracted backlog, 3.5 GW of contracted power across seven sites, and a target of $7-9 billion in annualized revenue by year-end 2026. [8] Milk Road AI frames Nebius and CoreWeave as strategic beneficiaries of Nvidia's deliberate effort — funded by roughly $40 billion in ecosystem investments — to build a multipolar compute market that keeps GPU demand high outside hyperscaler control. [8]

The skeptical read comes from Bloomberg analyst commentary summarized by Mandeep Singh: 'Meta can rent compute faster than it can become AWS, Azure, or Google Cloud.' Becoming a credible cloud provider requires billing systems, uptime guarantees, security infrastructure, and developer tooling that Meta does not have at enterprise scale, and building those capabilities could dilute the margin thesis. [1] Singh characterizes the move as 'fallback revenue' that funds AI rather than frontier leadership — a framing in direct tension with SemiAnalysis's view that Meta is pursuing high-value strategic partnerships. Neither a launch date nor pricing for Meta's cloud service has been publicly disclosed.