Everyone Wants To Be A Neocloud

Key Points:

• Meta is rapidly expanding its datacenter and compute capacity, having contracted over 5GW in the first half of 2024, with expectations for capital expenditure in 2027 to be significantly high, contradicting market concerns about overcapacity.
• Meta is pursuing multiple high-value use cases for its compute resources, including training frontier AI models through its Superintelligence Labs, scaling ads recommendation systems (RecSys) to boost revenue, and potentially offering private AI model instances via partnerships such as with Anthropic.
• The company is exploring innovative business models similar to SpaceX's high-revenue, on-demand compute deals, which could generate billions in annual revenue by leasing compute capacity with flexible contract terms.
• Meta aims to leverage its compute infrastructure to create a "Bedrock-type" AI service by partnering with frontier AI labs to provide private access to models like Anthropic's Claude, enabling both internal use and external commercialization opportunities.
• The scaling of Meta's ads recommendation systems has been a key driver of renewed revenue growth, with increased GPU investments improving ad targeting and content engagement, suggesting sustainable acceleration in core revenue streams.