Meta's shares rose more than 8 percent on July 1 as it pushes a new cloud business that sells spare AI computing resources to outside customers.

CNBC reported that Meta is preparing a plan to sell computing capacity that it does not use on its own infrastructure to external customers.

The plan is drawing attention because it suggests the market could see a path to recoup Meta's large investments in AI infrastructure. Meta told investors in April it plans to spend up to $145 billion in capital expenditure this year as it continues to expand data centres and secure graphics processing units to train AI models. With concerns that the timing of monetisation was unclear, some investors welcomed the prospect of generating revenue by selling spare compute resources.

Meta is reportedly reviewing what form of product it will offer. Discussions are under way on whether it will provide access to AI models running on Meta infrastructure or sell the raw compute resources themselves. Meta's business character could change depending on the service format. If it focuses on model access, it would be closer to an AI platform business, while a focus on selling compute resources would put it closer to an infrastructure cloud business.

If the business materialises, Meta would enter a new competitive market. Amazon, Microsoft, Google and CoreWeave currently lead the cloud infrastructure market. If Meta expands beyond an advertising-centred business structure into selling AI infrastructure, competition among big tech companies could become more direct.

The move also aligns with an intensifying race to secure computing resources since the AI boom. After OpenAI launched ChatGPT in 2022, model developers have stepped up efforts to secure computing power, but demand far outstrips supply. Meta has also expanded data centres and secured equipment to handle large workloads in that trend.

Some companies moved earlier to sell spare compute resources. Elon Musk's SpaceX began selling surplus computing capacity this year. SpaceX signed a contract with Anthropic worth $1.25 billion per month and also reached a contract with Google worth $920 million per month. For Meta, that amounts to an earlier case showing that a model of supplying idle resources to outsiders can function as a real revenue business.

Still, Meta's AI business has yet to clearly secure a decisive edge. Meta spent $14 billion last year to recruit Alexander Wang (알렉산더 왕) of Scale AI. In April, it unveiled its first model, Muse Spark, under Wang's leadership, but Meta presented it not as a cutting-edge model but as a "strong foundation". It still faces the task of organising both its AI model competitiveness and its cloud product strategy.

Market attention is focused on what form of service Meta will actually offer and at what pace it will turn its massive AI infrastructure investments into revenue.