- Meta is reportedly building a cloud business to sell excess compute capacity
- CEO Mark Zuckerberg said Meta was considering building a cloud business to compete with AWS, Azure and other cloud giants
- Meta could sell access to its AI models or sell neoclouds bare metal capacity
Meta is reportedly building a cloud business to sell excess compute capacity from its Meta Compute initiative.
Launched in January, Meta Compute will see the company build tens of gigawatts of compute capacity this decade and hundreds more over time.
Bloomberg reported that Meta is forming a plan to sell compute capacity which is not used for Meta’s internal needs. How exactly it plans to monetize the compute is not entirely clear. Options reportedly include selling access to AI models hosted on its infrastructure and selling bare metal capacity to neocloud companies.
The report comes after CEO Mark Zuckerberg said in May that it was considering building a cloud business to compete with AWS, Azure and other cloud giants.
Fierce Network analyzes Meta's cloud moves
A move into the cloud realm would put Meta into competition with AWS, Microsoft and Alphabet, plus neoclouds CoreWeave, Nebius, Lambda, Crusoe and Vultr.
The big difference between the big three hyperscalers and neoclouds is that hyperscalers have robust stack of software tools, services and AI models alongside raw compute capabilities.
Sid Nag, president and chief research officer at research and consulting firm Tekonyx, told Fierce the AI infrastructure market is likely large enough to support a fourth hyperscaler, “but not a fourth full stack cloud.”
“Meta already has hyperscale economics, world-class networking, custom silicon investments and one of the largest GPU footprints on the planet. That gives it every capability to monetize excess AI capacity, particularly for large-scale training, inference, and sovereign AI deployments,” he said.
Cloud customers today tend to buy from platforms that combine infrastructure with tools and services like security, databases, observability, governance, developer tooling and global support, he said.
Rather than chasing mainstream enterprise workloads, Meta’s most successful path could be selling bare metal compute capacity and AI infrastructure to sophisticated buyers, Nag continued.
“The bigger strategic question is whether bare-metal AI compute becomes its own durable market segment. In my view the answer is yes,” he concluded.
Meta’s compute ambitions go hand-in-hand with its AI efforts. On Wednesday, it appointed its former Chief Marketing Officer Alex Schulz as its first-ever Chief Data Officer.
More about Meta
Opinion: Meta’s $100B house of GPU cards
Amazon and Corning announce multi-billion-dollar deal for fiber supply
Meta is training up a new fiber workforce for its data center boom
Hyperscalers flex fiber muscle as telcos watch supply lines shift