Broadcom Is Borrowing Up to $100 Billion to Keep Anthropic's Chips Flowing.

News Breakdown · FiscEdge Academy

Broadcom is in talks with lenders to raise as much as $100 billion in private debt to bankroll a wave of custom AI chips, with Anthropic as the anchor customer, according to Bloomberg and CNBC reporting this week. The structure being discussed splits into a senior tranche of roughly $45 billion to $70 billion and a junior tranche of about $30 billion to $35 billion, funneled through a special-purpose vehicle rather than sitting on Broadcom's own balance sheet.

Blackstone and Apollo Global Management, the two private credit giants already inside this arrangement, are the lead investors again. That's the tell: this isn't a new idea, it's an expansion of the $35 billion capital solution the three parties announced together in June, built to deliver more than 20 gigawatts of AI compute to frontier labs by 2028, with the first 1 gigawatt already earmarked for Anthropic's training and inference buildout.

The number is the least interesting part

A single company borrowing tens of billions isn't, on its own, remarkable in 2026. What matters is where the money sits. S&P Global counts $225 billion in bonds issued by hyperscalers and chip-adjacent companies so far this year, and six major AI players issued roughly $244 billion in the first half alone, more than double last year's pace. Layer in private-credit structures like Broadcom's SPV, largely invisible to public bond markets, and analysts now estimate total hidden AI-related borrowing has climbed past $1.65 trillion.

That's the real story: the AI buildout has quietly shifted from being funded by equity (VC rounds, IPO proceeds, retained earnings) to being funded by debt that doesn't show up on the headline balance sheet of any single company. Broadcom isn't borrowing to build its own chips. It's borrowing through a vehicle so a third party, Anthropic in this case, gets compute capacity without either company's stock reflecting the leverage directly.

Why an SPV, and why now

Special-purpose vehicles exist to isolate risk and keep debt off a parent company's consolidated balance sheet. Using one here means Broadcom's credit rating and Anthropic's own balance sheet both stay comparatively clean while the actual lenders, Blackstone and Apollo's credit funds, take on the exposure in exchange for the yield. It's the same basic mechanic that financed data centers and telecom buildouts in prior cycles, just now sized for AI compute and moving at a pace that has bond investors publicly uneasy: S&P has flagged that yields on this debt are rising relative to risk-free benchmarks, a sign the market is starting to price in real risk rather than treating AI-linked credit as automatically safe.

What this means if you're building or raising right now

You don't need a Broadcom-sized balance sheet for this to matter. Three things are worth internalizing:

• Compute is becoming a financing product, not just a cost line. If your SaaS or AI product depends on inference costs, the capital stack behind your GPU supplier is now a real variable in your own unit economics, because debt loads upstream eventually show up in the price you pay for compute.
• Private credit is becoming the venture debt of AI infrastructure, not just of startups. The instrument financing Anthropic's chips is structurally similar in spirit to the credit facility Anthropic itself raised weeks earlier, which we covered in our breakdown of its $10 billion revolver. Debt is stacking at every layer of the AI supply chain at once, not just at the foundation-model layer.
• Leverage this size only works if revenue keeps compounding. The entire structure assumes Anthropic's usage and revenue keep climbing fast enough to justify 20-plus gigawatts of committed capacity. If growth stalls anywhere in that chain, the SPV structure that currently hides the risk is exactly what transmits it fastest. Stress-test your own growth assumptions with that same scrutiny when you build a financial model, because the underlying math applies at any scale, not just at $100 billion.

If you remember one thing

The AI boom's biggest bottleneck was never just chips, it's the debt now being engineered to pay for them, one layer removed from the companies whose names are on the press releases. When financing structures get this creative to keep compute flowing, that's a market signal about how much slack is left in the system, not just a financing footnote.

---

We teach how to build a durable capital and growth plan in FiscEdge's financial modeling course and startup strategy track. Browse the full blog for more breakdowns like this one. Follow @fiscedge for daily Business & AI analysis.