[Edaily Reporter Kim Hyun-ah ] Speculation has arisen that Meta may set out to build its own AI cloud platform based on its artificial intelligence (AI) infrastructure. Analysts suggest this is a strategy to expand beyond simply investing in computing power for AI model development and into a platform business that provides AI models to external companies.

Market research firm SemiAnalysis predicted in a recent report that “Meta’s investments in data centers and AI computing will not only not slow down but will actually accelerate further.”

According to the report, Meta secured more than 5 GW (gigawatts) of power capacity through its cloud and colocation data centers in the first half of this year alone. This figure excludes the massive data centers the company is currently building itself.

This is on par with the plan for a 5 GW AI data center (AIDC) by 2029 recently announced by SKTelecom(017670), and is seen as evidence that Meta has already embarked on securing massive AI infrastructure.

Inc.

SemiAnalysis particularly noted that Meta is in the final stages of negotiations with AI startup Anthropic to adopt a “Private Instance” of the AI model “Claude.”

A “Private Instance” refers to operating Claude in a dedicated environment within Meta’s own data centers, rather than on a public cloud. Since the AI model runs in an environment isolated from the outside world, this approach enhances security and data control while allowing Meta to independently optimize performance and costs.

This goes beyond simply using AI; it also signifies that Meta is securing the foundation to host Claude on its own infrastructure.

SemiAnalysis interpreted this not merely as the adoption of an AI model, but as a signal that Meta intends to expand into the AI model platform business, similar to AWS’s “Bedrock” or Google’s “Vertex AI.”

The analysis suggests that, initially, Meta will utilize Cloud internally for research and development, advertising, and workflow automation, and that it may later offer the AI model to enterprise customers via an API (Application Programming Interface) or as a “Token as a Service.”

It was also suggested that, in the long term, the platform could evolve into a multi-model platform offering not only Anthropic’s models but also those from OpenAI and other providers.

Four potential uses for Meta’s AI infrastructure were identified: △ development of next-generation AI models, △ enhancement of advertising recommendation systems, △ construction of an AI model platform, and △ high-performance computing rentals.

SemiAnalysis predicted, “Meta possesses a vast array of GPUs, data centers, a global base of advertising clients, and a powerful distribution network through its social platforms,” adding, “By combining AI models, Agent AI, and enterprise SaaS (Software as a Service), Meta could leap forward from being a mere purchaser of GPUs to becoming a key player in the AI platform and application distribution market.”