Recently, news that Meta Platforms (META) has signed a “multi-year” agreement to secure flash memory supply from SanDisk (SNDK) triggered a chain reaction in the market. Following the release of this news, SanDisk’s stock price surged sharply, while Meta’s stock came under pressure, experiencing a decline of approximately 4%. However, as of 11:40 a.m. Eastern Time, Meta’s stock had reversed its losses and turned positive, ultimately recording a gain of about 0.5%.
According to Reuters, citing internal Meta documents, the social media giant plans to purchase NAND flash memory from SanDisk, DRAM from Samsung, and fiber optic products from Sumitomo Electric Industries as part of its effort to build its own artificial intelligence computing infrastructure. In addition, Broadcom (AVGO) is assisting Meta in designing an AI semiconductor named Iris, while Taiwan Semiconductor Manufacturing Company (TSM) will be responsible for the foundry manufacturing of the chip. Although none of the above companies have officially confirmed the details of the reports, investors have acted based on expectations, with the U.S.-traded stocks of SanDisk, Broadcom, and TSMC all rising on the day.
Regarding the short-term decline in Meta’s stock price, market analysis points to its massive capital expenditure plan. Meta plans to invest up to $145 billion in AI infrastructure this year, and today’s procurement news confirms that this plan is steadily advancing. Although this investment is intended to provide Meta with powerful computing support in its competition with rivals such as Alphabet and OpenAI, according to data from S&P Global Market Intelligence, this amount even exceeds Meta’s estimated cash inflow of approximately $136.6 billion from operating activities this year. This high-risk financial outlook is considered the primary reason for triggering nervousness among some investors and causing the initial decline in the stock price.
While fully committing to internal research and development, Meta is also exploring ways to enhance the return on its AI infrastructure. The company’s Chief Executive Officer, Mark Zuckerberg, stated that the current demand for computing power in the AI industry is strong, and that Meta’s existing computing capacity is almost entirely used for internal R&D, with the company itself still requiring substantial resources. However, he also noted that if external leasing offers are sufficiently attractive, leasing some computing resources to external clients could be more commercially viable than using them entirely for internal projects.
It is understood that Meta is evaluating multiple cloud service models, including services similar to Amazon AWS’s Bedrock platform, which provides developers with access to AI models hosted on Meta’s infrastructure and charges based on usage. The company has announced that it will make its next-generation large language model, Muse Spark 1.1, available to developers via API. In addition, Meta is also considering launching a “bare computing” leasing service similar to that offered by CoreWeave (CRWV). Zuckerberg stated that even if Meta does not fully utilize all its computing power in the future, the company could still lease resources on a long-term basis, much like Amazon AWS, Microsoft Azure, and Google Cloud. He specifically mentioned SpaceX’s model of leasing data center computing power, considering it a valuable reference, and noted that Meta has already received many similar collaboration invitations. However, he emphasized that at this stage, the company will continue to prioritize its own AI R&D and the computing resources required for product deployment.
