Published Tuesday, October 27, 2026, the week of the Fed’s October meeting and the Microsoft, Alphabet and Amazon reports.
Our thesis: this week the AI trade stopped being only a story about demand and became a story about financing. When the biggest buyers of AI infrastructure start raising outside money to pay for it, the price of that money matters as much as the next chip order. And the price of money is exactly what the Federal Reserve decides on Wednesday. We think the AI build-out has quietly become a cost-of-capital trade, with the most risk sitting with whoever needs outside funding to keep spending.
The week the bill showed up
On Monday, October 26, Meta announced a financing plan of roughly $95 billion: about $35 billion of underwritten offerings (around $22.5 billion of Class A common stock and around $12.5 billion of mandatory convertible preferred shares) plus a planned at-the-market program of about $60 billion that it does not expect to start before the first quarter of 2027. Meta said it expects $135 to $145 billion of capital spending in 2026, and that it will fund its investments through a mix of operating cash flow, investment-grade debt, partnerships and equity (Meta press release).
In its third quarter Meta reported operating cash flow of $32.79 billion against capital spending, including finance-lease payments, of $34.91 billion, so free cash flow was negative $2.12 billion, even as revenue grew 26% to $64.60 billion (ETF.net, citing Meta’s filing). Meta is not desperate: it reports about $82 billion of cash and marketable securities. But a leading buyer is now telling the market that cash flow alone will not cover the plan.
One detail stood out to us. Meta said Nvidia, AMD and Broadcom, which are suppliers to the AI build-out, indicated interest in buying up to $5.0 billion of the new common stock, though those indications are not binding. When sellers help fund the buyer, demand and financing start to blur.
The same day, Anthropic filed its S-1 to list on Nasdaq under the ticker “ANT” (Anthropic). According to CNBC’s reading of the filing, revenue was $16.29 billion in the first half of 2026, up from $1.16 billion a year earlier, while the reported net loss was $80.7 billion, about $72 billion of it from non-cash accounting charges tied to its rising valuation. CNBC also reported that Anthropic’s noncancelable purchase commitments grew from $54.6 billion at the end of 2025 to $265.6 billion by June 30, 2026 (CNBC).
Why the Fed is part of the AI story now
The Fed’s target range is 3.75% to 4.00% after a quarter-point increase on September 16, and the committee announces its decision at 2 p.m. ET on Wednesday, October 28 (ETF.net). In a Reuters poll taken October 16-20, 94 of 100 economists expected a hold this week, and all 100 expected a hike in December (Reuters via Investing.com). The 10-year Treasury yield ended Friday, October 23, at 5% (ETF.net).
Higher rates make debt more expensive, especially for smaller data-center builders, and raise the bar for long-dated projects: a data center that pays off over many years is worth less today when safe bonds pay more. CNBC noted that so-called neoclouds such as CoreWeave and Nebius had lost over 30% of their value in October as credit risk rose across the data-center space (CNBC).
History: booms end when funding gets harder
The telecom build-out. In the late 1990s, carriers funded fiber networks with stock and heavy borrowing, and equipment makers such as Lucent and Nortel lent money to customers so they could keep buying gear. When capital markets closed after 2000, the spending stopped faster than demand did. Global Crossing filed for bankruptcy in January 2002 and WorldCom in July 2002. The lesson we take: the build-out ended when funding dried up, not when the internet stopped growing.
The 1994 bond shock. In February 1994 the Fed began raising rates after a long pause and kept going through the year, faster than bond investors expected. Bond prices fell hard and leveraged investors were hurt. The lesson: the price of money can reprice assets before it changes the real economy.
Shale oil in the 2010s. U.S. shale producers grew output quickly, funded heavily by debt and new shares. When oil prices collapsed in 2014-2016, many producers went bankrupt and investors began demanding “capital discipline”: spend less, return cash. The lesson: when investors stop funding growth at any price, spending plans change.
Illustrative example, not real data.
Imagine a company that wants to spend 100 dollars this year and earns 90 dollars of operating cash. It must raise 10 dollars from outside. If its spending rises to 120 while cash stays at 90, the outside gap triples to 30. Small changes in spending can make a company far more dependent on markets and rates. The numbers are made up to show how the gap grows.
What is different this time
To be fair, today’s biggest spenders are not the junk-rated carriers of 1999. Meta has Aa3/AA- ratings and chose equity over piling on debt, the more conservative choice for its balance sheet, if not for existing shareholders. Anthropic’s revenue growth is real. Reuters estimates hyperscaler capital spending at around $800 billion this year (via ETF.net), and much of that is still funded from very large operating profits. These differences lower the odds of a 2001-style collapse. They do not make the cost of capital irrelevant.
What we watch this week
- The Fed statement and press conference on Wednesday. Not the hold itself, which is expected, but whether December sounds firm.
- How Microsoft, Alphabet and Amazon say they will pay. Microsoft and Alphabet report after Wednesday’s close and Amazon on Thursday. Listen for spending plans next to free cash flow, debt and any equity talk, and for AI revenue keeping pace.
- Credit for smaller builders. Pulled listings and higher borrowing costs at data-center companies are the early warning, as they were for telecom.
How we see the future
Good case. The Fed sounds patient, yields ease, and this week’s reports show AI revenue rising fast enough that the largest buyers can fund most spending from cash. New equity is absorbed smoothly.
Likely case (our view). Rates stay high and a December hike stays likely. The largest companies keep spending but explain financing more carefully, and the market keeps separating self-funded spenders from those that need outside money. Stress shows up first in smaller, debt-funded builders.
Bad case. Inflation and oil keep yields climbing, equity buyers grow tired of new issuance, and debt-funded data-center projects struggle to refinance. Spending plans get cut at the same time, hitting suppliers, much as funding stress ended the telecom boom, even while AI use keeps growing.
What would change our mind. We would worry less if this week’s hyperscalers show free cash flow comfortably covering capital spending, if yields fall meaningfully, and if new AI equity offerings are absorbed without strain. We would worry more if other large buyers follow Meta into equity markets, if more AI or data-center listings are pulled, or if suppliers increasingly finance their own customers.
Our take
The question for the AI build-out used to be “will customers want it?” This week added a second one: “who pays, and at what price?” Companies that fund their ambitions from their own cash are in a very different position from those that need the market’s help, and the Fed now sits in the middle of the AI story. For a longer view of the cycle itself, see The AI build-out is a spending cycle, and cycles turn. This is our opinion for education, not a recommendation about any stock.
Sources
- Meta Platforms press release, October 26, 2026: equity offerings and financing plan
- ETF.net, “Week ahead: Fed rate decision, Microsoft, Alphabet and Amazon earnings” (Oct 25-26, 2026)
- Anthropic, “Anthropic files S-1 with the SEC for proposed initial public offering” (Oct 26, 2026)
- CNBC, Anthropic IPO live coverage of the S-1 filing (Oct 26, 2026)
- Reuters poll via Investing.com, “Fed to skip October but raise rates again in December” (Oct 20, 2026)