Meta raises $25B for AI

Bonds are usually the boring part of a tech story. Not here. Meta just sold $25 billion of investment-grade debt on April 30, 2026, and the point was simple: keep feeding an AI buildout that is getting more expensive, more power-hungry, and more urgent by the quarter. The timing matters — the sale came one day after Meta raised its 2026 capital expenditure outlook to $125 billion to $145 billion. That tells you this is not routine treasury housekeeping. It is infrastructure finance for an arms race. (bloomberg.com) ### Why borrow that much now? Because Meta’s spending plan just jumped again. In its latest earnings update, the company raised its 2026 capex range from $115 billion-$135 billion to $125 billion-$145 billion. Meta tied that increase to higher component pricing and additi(bloomberg.com) wants more of both. (cnbc.com) ### What is the money actually for? Mostly compute. That means AI-optimized data centers, servers, networking gear, and the silicon that fills them. Meta has been making that unusually explicit in its infrastructure push this year. It broke ground on a new AI-optimized data center in Tulsa in April, calling it its 28th US(cnbc.com)campus in Lebanon, Indiana, described as a more than $10 billion investment. Those are not side projects — they are the physical footprint of Meta’s model ambitions. (about.fb.com) ### Why use bonds instead of cash? Because even giant cash machines like Meta are trying to spread the load. AI infrastructure now arrives in giant, lumpy chunks — a campus here, a long-term chip commitment there, power contracts on top. Borrowing lets Meta keep flexibility while still mov(about.fb.com)ds in October 2025, which makes this latest sale look less like a one-off and more like a financing pattern. (bloomberg.com) ### Did investors like the deal? Yes, but with more caution than last time. Orders reportedly peaked around $96 billion, which is still huge, but below the roughly $125 billion order book for Meta’s October 2025 sale. Nearly all six tranches priced at wider spreads than t(bloomberg.com)pending. (bloomberg.com) ### Why does this matter beyond Meta? Because the AI race is no longer just about models, products, or talent. It is about balance sheets. The companies at the top are turning credit markets into another competitive weapon. Bloomberg has framed this as part of a broader b(bloomberg.com)at can finance tens of billions quickly and everyone else gets wider. (bloomberg.com) ### What does that mean for smaller labs? They probably do not win by copying Meta. Matching hyperscaler spending dollar for dollar is unrealistic for most startups, research labs, and regional players. The more plausible path is partnerships, rented capacity, (bloomberg.com)oo — Bloomberg reported in March that it could spend up to $27 billion over five years with Nebius for AI infrastructure. Even Meta is mixing ownership with access. (bloomberg.com) ### So what changed this week? The new fact is not just that Meta wants more AI. Everyone knew that. The new fact is that Meta went to the bond market immediately after raising its spending target and pulled in $25 billion in one shot. That makes the scale concrete. AI infrastructure is no longer a future plan on a slide deck — it is being financed, built, and locked in now. (bloomberg.com) The bottom line is simple. Meta is telling markets that AI is worth borrowing for at industrial scale. If that bet works, the winners will not just have better models — they will have more power, more chips, and more room to train.