Broadcom’s $370B AI Bet: Why the Scary Number Is Misleading
Broadcom’s $370 Billion AI Financing Deal: Scary Headline, But Here’s the Real Story
TL;DR: Broadcom’s stock dropped because a bank warned about a massive $370 billion financing program. But that number is a theoretical maximum for future deals, not actual debt. The real commitment right now is capped at $29 billion—and Broadcom is making record profits to cover it.
What Happened to Broadcom’s Stock?
Last week, Broadcom (NASDAQ: AVGO) shares fell 5.9%, closing around $393—that’s 21% below their 52-week high. The drop wasn’t about bad earnings. Instead, Bank of America downgraded Broadcom’s bonds (not its stock) to "market weight"—Wall Street speak for "neutral."
Why? A new financing platform Broadcom built with two giant investment firms: Apollo Global Management and Blackstone.
Important: This wasn’t a stock downgrade. It was a bond downgrade—meaning the bank got nervous about Broadcom’s debt obligations, not its business performance.
The Big Scary Number: $370 Billion
The headline that spooked everyone: Bank of America analysts estimate this platform could reach $370 billion in financing by mid-2029, with Broadcom guaranteeing much of it.
That sounds terrifying. $370 billion is more than the GDP of many countries!
But here’s the catch: That’s not what Broadcom owes today. It’s not even what they’ve signed up for. It’s a theoretical ceiling—a "what if everything goes perfectly and we max out the platform" number.
What the $370 Billion Actually Means
Let’s break it down like you’re five:
| What People Think | What It Actually Is |
|---|---|
| Broadcom borrowed $370B | Zero dollars borrowed |
| Broadcom owes $370B | Zero dollars owed |
| Broadcom signed $370B in guarantees | Only $29B signed so far |
| $370B is on the balance sheet | It’s a future projection, not current debt |
Key Distinction: The $370 billion is what happens if every possible future deal gets done at maximum scale by 2029. It’s like saying "If I buy a lottery ticket every day for 5 years, I could spend $10,000"—but you haven’t spent a dime yet.
The Real Numbers: What Broadcom Has Actually Signed Up For
According to Broadcom’s own quarterly filing (10-Q), here’s the actual commitment as of June 2026:
The First Deal: Capped at $29 Billion
- June 8, 2026: Broadcom arranged for an investor partner (Apollo) to buy AI server racks built on Broadcom’s custom chips
- AI labs (like Anthropic) lease those racks for 5-year terms
- Broadcom guarantees the lease payments—this is the "backstop"
- Maximum exposure: $29 billion (grows as racks deploy, shrinks as customers pay)
- If customer defaults: Broadcom takes the racks back or sells them to recover value
Bank of America’s "Worst Case" Models
| Scenario | Estimated Broadcom Loss |
|---|---|
| Every customer defaults at once (full platform) | ~$42 billion |
| 25% default rate | ~$10.5 billion |
| First deal only, total wipeout | $29 billion (filing cap) |
Reality Check: These models assume Broadcom recovers some value by reselling the server racks. The $29B filing cap assumes zero recovery—absolute worst case.
How the AI Financing Platform Works (Simple Explanation)
Think of it like car leasing, but for AI supercomputers:
The Cycle (Step by Step)
- Broadcom designs custom AI chips (their specialty)
- Apollo/Blackstone (big investors) buy the server racks containing those chips
- AI companies (Anthropic, OpenAI) lease the racks instead of buying them outright
- Broadcom promises: "If the AI company can’t pay, we’ll cover it"
- Broadcom sells more chips → Investors get lease income → AI labs get computing power without huge upfront cost
Why This Exists
- AI labs need MASSIVE computing power (20+ gigawatts = ~20 nuclear power plants worth!)
- They don’t have hundreds of billions in cash sitting around
- Broadcom wants to sell more chips—so they help finance the demand
- Investors (Apollo/Blackstone) want steady returns from creditworthy tech companies
Named Customers So Far:
- Anthropic — First phase: 1+ gigawatt starting mid-2026
- OpenAI — Also named as a platform customer
Broadcom’s Financial Health: Can They Handle This?
Let’s look at the scorecard:
Q2 Fiscal 2026 Results (Ended May 3, 2026)
- Revenue: $22.2 billion (+48% year-over-year)
- Earnings: $9.3 billion (+88% year-over-year)
- Q3 Guidance: ~$29.4 billion revenue (+84% year-over-year)
Putting $29 Billion in Perspective
| Metric | Value |
|---|---|
| Max loss on first deal | $29 billion |
| Quarterly earnings (current pace) | ~$9.3 billion |
| Months of profit to cover max loss | ~3 months |
| Months of profit at Q3 pace | ~2.5 months |
Bottom Line: Even a total wipeout on the first deal = ~3 months of profits. Painful? Yes. Company-ending? Not even close.
The Growing Risk: Future Deals Could Add Up
Here’s the real concern the article highlights:
Today’s cap: $29 billion (first deal only)
Platform design: 20+ gigawatts (many more deals coming)
If each new deal gets same backstop → committed number keeps climbing
The Structural Risk
- Customers are private AI labs (not public companies with transparent finances)
- Lease payments depend on AI boom continuing (if funding dries up, they can’t pay)
- Broadcom is effectively financing demand for its own chips
- Stock valuation: ~65x earnings, ~25x forward earnings — priced for perfection
The Real Risk Isn’t $370 Billion. It’s that Broadcom’s growth now depends on guarantees it extends to keep the AI ramp going. Each new deal adds to the commitment pile.
Should You Buy Broadcom Stock Now?
The article ends with a pitch for Motley Fool Stock Advisor—which notably did NOT include Broadcom in its current "10 Best Stocks" list.
Questions to Ask Yourself Before Buying
- Am I comfortable with a company using guarantees to fuel growth?
- Do I believe AI labs will keep getting funded for 5+ years?
- Is 65x earnings reasonable if growth slows?
- Can Broadcom manage increasing backstop exposure without hurting credit?
- Are there better AI plays with less financial engineering?
Disclosure: The Motley Fool owns Broadcom and Blackstone stock. Bank of America is an advertising partner. The author has no position.
Summary
| What’s True | What’s Exaggerated |
|---|---|
| Broadcom guarantees lease payments for AI server racks | Broadcom owes $370 billion |
| First deal capped at $29B max loss | $370B is current debt |
| Worst-case modeled loss: $42B (full platform) | Broadcom will actually lose $42B |
| Broadcom earnings growing 80%+ YoY | Business is in trouble |
| Stock dropped on bond downgrade, not earnings miss | Stock crash = fundamental problem |
The $370 billion is a "what if we max out everything" number. The $29 billion is what’s real today. Broadcom prints enough profit to cover the real number several times over—but the trend of adding guarantees is worth watching.
FAQ
Is Broadcom in financial trouble?
No. Revenue up 48%, earnings up 88% last quarter. They guided for 84% revenue growth next quarter. The concern is about future commitments, not current health.
What does "market weight" mean for bonds?
It’s a neutral rating—not "sell." Bank of America is saying: "These bonds are fairly priced given the risk, but we’re not excited." It’s a warning to bondholders, not stockholders.
Why would Broadcom guarantee customer leases?
To sell more chips. If AI labs can’t afford billions upfront for servers, Broadcom loses sales. By backstopping leases, they enable the purchase—and book the chip revenue.
Could Broadcom really lose $29 billion?
Only if: (1) Anthropic stops paying entirely, AND (2) the server racks are worth near zero. Racks with custom AI chips likely have significant resale value, so actual loss would be lower.
Should I sell my Broadcom shares?
That depends on your risk tolerance. If you own it for AI chip dominance, the thesis is intact. If you’re worried about off-balance-sheet leverage growing, that’s a valid concern to monitor quarterly.