Meta to Sell up to $95B in Shares to Fund AI Initiatives

Meta announced on Monday (26) a plan to sell up to $95 billion in shares to finance the construction of data centers for artificial intelligence. The first phase, amounting to about $35 billion, combines $22.5 billion in Class A common stock, traded through forward sale contracts, and approximately $12.5 billion in depositary shares backed by mandatory convertible preferred stock, according to the prospectus registered with the SEC and reproduced by Benzinga, Investing.com, and other outlets.

The remaining $60 billion would come from a gradual market sale program, known as at-the-market, utilizing a new Class C of non-voting shares. Meta stated that the timeline, size, and terms of this program have yet to be defined, and it will only commence once the new class begins trading.

Suppliers on the Other Side of the Counter

The most striking detail lies in the list of interested parties. Nvidia, AMD, and Broadcom have expressed non-binding interest in purchasing up to $5 billion in Class A shares collectively, according to the offering documents. These companies supply Meta with the accelerators and custom chips that the capital raised will fund.

This approach is not new. Nvidia had previously signed a letter of intent in September 2025 to invest in OpenAI, another major buyer of its processors. Critics label the arrangement as circular financing: the supplier injects capital into the customer, who then returns the money in orders. Nonetheless, there are counterarguments. The $5 billion is about one-seventh of the $35 billion in the first phase, and the expression of interest does not obligate anyone to buy.

A Profitable Company Choosing to Dilute

Meta projects capital investments between $135 billion and $145 billion in 2026, a range echoed by Benzinga and other outlets based on guidance from the company. Until now, the group has financed its expenses through cash generation from advertising and, for larger projects, with off-balance-sheet structures. The Hyperion campus in Louisiana, designed for 5 gigawatts, was established in 2025 as a joint venture with funds from Blue Owl Capital, which retained 80% of the partnership, according to Meta's announcement at that time.

Issuing shares is the most expensive source of capital for a profitable company since it divides future profits among more shares. This choice indicates something about the scale of the plan: Meta concluded that internal cash, debt, and infrastructure partnerships are insufficient for the pace of construction it intends to maintain. Those interpreting this move as a sign of tightening are misdiagnosing the situation. A company with advertising margins financing capex does not compare to a lab burning venture capital. The risk lies in the return on these assets, which has yet to appear as separate AI revenue on the balance sheets.

The non-voting Class C solves another problem. By selling shares without voting rights, Mark Zuckerberg expands the capital base without diluting his control maintained through Class B shares with multiple votes. For governance funds, it raises the same objection raised against Google's dual structure a decade ago.

Where the Money Will Go

The majority of the $95 billion will leave the United States and return to the Asian semiconductor supply chain. In Taiwan, TSMC manufactures processors for Nvidia and AMD and the custom chips designed with Broadcom. In South Korea, SK Hynix and Samsung supply the high-bandwidth memory that accompanies each accelerator. For these manufacturers, a capital raise of this magnitude serves as an order book with funding already secured.

In Europe, the impact arrives through energy and licensing. Meta operates data centers in Denmark and Ireland. In Ireland, the network operator EirGrid has been restricting new data center connections in the Dublin area since 2022, and the capital raised in the U.S. cannot purchase megawatts that the European grid cannot deliver. Therefore, Meta's AI expansion is likely to focus on areas with available power, with Hyperion in Louisiana being the most clear example.

For the CIO purchasing cloud AI infrastructure, Meta's offering has an indirect and tangible effect. Meta does not sell capacity to third parties but competes for the same batches of GPUs, memory, and electrical transformers that AWS, Microsoft, and Google resell to corporate clients. When a single buyer raises $95 billion in equity to build for itself, the delivery queue shortens for them and lengthens for the rest of the market.