Major Bank Rating｜Bank of America: Meta’s Cloud Business Plans “Dual Track,” Key Still Lies in Large Language Model Development, Maintains “Buy” Rating

Glonghui, July 3rd | Bank of America analysts indicated in their latest report that Meta’s cloud business plans to adopt a “dual-track approach”: first, by providing AI model hosting services (such as its own Muse Spark series), charging based on usage, with a model similar to AWS Bedrock; second, by directly renting out raw GPU computing power, positioning itself close to emerging compute rental companies like CoreWeave. Bank of America maintained a “Buy” rating on Meta with a target price of $835. The bank also reminded that Meta’s in-house chip development lags behind cloud giants such as Amazon, Microsoft, and Google, and it still needs to purchase up to 1.6 GW of computing power from third parties like Crusoe. The logic of reselling computing power and Meta’s competitiveness in the cloud market remains under debate. According to analysts, whether Meta’s cloud business can gain market recognition crucially depends on the development of large language model (LLM) capabilities—the stronger the model, the higher the external demand for Meta’s computing power and the more solid the business logic. Additionally, Bank of America mentioned that OpenAI recently reduced inference costs for certain models by about half through system-level optimization. If such technologies are widely adopted in the industry, it would expand effective output without additional hardware investment, improve the unit economics of AI businesses, and have an overall positive effect on large internet companies.