Meta announces approximately $95 billion financing plan for AI infrastructure and personal superintelligence

As of the end of September, Meta still had about $82 billion in cash and marketable securities on its books, but under the pressure of computing bills, the company is still preparing to ask the capital markets for a huge sum.

On local time October 26, Meta announced a financing plan totaling about $95 billion to fund AI infrastructure and the expansion of "personal superintelligence."

Meta stated in the announcement that the financing plan is divided into two parts, one of which is a concurrent underwritten public offering of about $35 billion, including about $22.5 billion of Class A common stock and about $12.5 billion of depositary receipts representing preferred stock; in addition, the company will launch an at-the-market (ATM) offering program of about $60 billion.

A so-called at-the-market offering is one in which a company sells shares in batches according to market conditions; the actual timing, size and terms of the issuance have not yet been determined, and the company expects it to be no earlier than the first quarter of 2027.

Based on Meta's previously projected 2026 capital expenditures of $135 billion to $145 billion, $95 billion exceeds two-thirds of its capital expenditure forecast.

The company said that Nvidia, AMD and Broadcom, as cornerstone investors, have indicated an intention to purchase up to $5 billion of Class A common stock, but these intentions do not constitute binding subscription commitments. The AI infrastructure race continues, and the binding of capital and computing power is becoming deeper and deeper.

As for the reason for the financing, Meta mentioned that it is investing in expanding infrastructure and computing capacity to support the opportunities brought by AI and to build out personal superintelligence.

According to Meta's disclosure, within a few weeks of Muse's launch, it attracted more than 10 million users in the United States and Canada. Shopping is its main use case, with more than 25% of weekly active users participating in shopping activities. Meta CEO Mark Zuckerberg said at a previous earnings call that the average purchase amount per user exceeded $200, including travel spending.

But computing power supply has constrained the rollout of this AI product. Zuckerberg said the company is building computing capacity that can serve more than 1 billion people in the coming years, but Muse is highly compute-intensive, and efficiency and capacity are important limiting factors for further expanding coverage.

But the problem is that the more users there are, the bigger the bill may be. While building computing power, Meta also proposed to increase Muse's computational cost efficiency by more than tenfold over the next year, reducing the unit cost of service by making fuller use of off-peak computing power, more efficient models and inference optimization.

Currently, Meta has data centers in 23 U.S. states, is building four superclusters each with capacity exceeding 1 gigawatt, and is expected to nearly double total infrastructure capacity year over year in 2026.

For Meta, product expansion, revenue growth and cost optimization must all advance at the same time. Zuckerberg said Muse will start monetizing earlier than the company's previous products, due both to operating costs and to the commercial demand users have already shown. Management expects that by expanding the model API, Muse Code and enterprise agent businesses, the new AI business is expected to begin contributing incremental revenue starting in 2027.

From a financial data perspective, Meta is still in a growth range, and the third-quarter report released on October 21 exceeded market expectations: revenue in the third quarter of 2026 was $64.6 billion, up 26% year over year; fourth-quarter revenue is expected to be $73 billion to $76 billion.

But on the other hand, free cash flow turning negative has become a focus of market attention. In the third quarter, Meta's capital expenditures, including principal repayments on finance leases, reached $34.91 billion, exceeding $32.79 billion in cash flow from operating activities, and free cash flow was negative $2.12 billion.

Meta is not the only company to post negative quarterly free cash flow and need to raise huge amounts of money for AI expansion.

In June this year, Google's parent company announced the largest financing in its history: it plans to raise $85 billion to invest in AI infrastructure construction. In its second-quarter 2026 financial report, Google recorded negative free cash flow of $5.9 billion for the first time in its history.

Just recently, AI unicorn Anthropic is also preparing for an IPO. Its prospectus discloses that Anthropic's revenue in the first half of 2026 was $16.29 billion, but its net loss was as high as $80.75 billion. Anthropic expects to commit at least $518 billion in spending on cloud services, computing power and infrastructure over the next decade.

Tech giants are successively turning to the capital markets and becoming more reliant on external financing, reflecting that the funding needs of the entire AI industry are still expanding. Goldman Sachs expects that from 2026 to 2031, total AI-related capital expenditures may reach $7.6 trillion, including computing power, data centers and power infrastructure.

AI giants that successfully raise funds may be able to expand infrastructure on a large scale, but whether they can convert this into revenue and profit in the future and cover computing costs remains unknown to investors. At the close on local time October 26, Meta's U.S. shares fell 1.88% to $630.56 per share, with a current total market value of $1.61 trillion.