Meta Platforms Inc. will sell up to $95 billion worth of stock to finance its data center projects.

The fundraising initiative, which the company announced today, will kick off with a series of transactions cumulatively worth $35 billion. About two-thirds of that sum will come from the sale of common stock. Meta said Nvidia Corp., Advanced Micro Devices Inc. and Broadcom Corp. had expressed interest in buying up to $5 billion of the shares.

The company will sell the common stock alongside $12.5 billion worth of securities it calls Series A depositary shares. Each such instrument corresponds to 1/20th of a so-called Series A mandatory convertible preferred share. The latter securities will turn into Meta common stock on Aug. 15, 2029. Until then, they will trade under a separate ticker symbol on the Nasdaq. 

Meta’s funding haul could increase if the banks underwriting the transactions decide to buy more stock. The company will give them the option to buy 37.5 million additional depositary shares and 3.375 billion additional shares of common stock.

Meta intends to follow up the deals with a stock sale that is expected to raise as much as $60 billion. It will be structured as an at-the-market offering, which means that the company will sell shares incrementally over a relatively long time period. It will begin issuing the shares early next year.

The deals will help Meta finance its expensive data center construction program. The initiative is headlined by four so-called AI superclusters that are each expected to provide more than 1 gigawatt of computing capacity. The largest development, a Louisiana campus dubbed Hyperion, is set to provide 5 gigawatts. Meta will commission 10 new power plants to meet Hyperion’s energy requirements.

The company is not relying on stock sales alone to finance its data center investments. Hyperion is a joint venture between Meta and investment firm Blue Owl Capital, which has committed $7 billion to the project. Meta plans to build its Texas supercluster through a similar partnership with BlackRock Inc., the world’s largest asset manager.

The company’s data center network powers its Muse family of large language models. The newest addition to the series, Muse Wonder Full, is one of the most advanced LLMs on the market. It outperformed GPT-6 Astra and Claude Opus 5.5 on multiple popular benchmarks including the DeepSWE collection of coding challenges.

Meta stated today that its AI engineering roadmap places an emphasis on making its models more cost efficient. The company plans to achieve that by lowering its LLMs’ hardware footprint and enhancing the underlying inference infrastructure. Meta also intends to make better use of its off-peak capacity, computing power and energy that is underutilized at times when user traffic is low. 

Photo: Meta