A preliminary U.S.–Iran agreement (MOU declared Sunday, signing slated for Friday in Geneva, Strait of Hormuz to reopen "toll free") flipped the risk switch. The result was a textbook risk-on melt-up:
That's the story on every screen. Here's the one buried underneath it.
For three years, the AI capex supercycle was funded the boring way — out of hyperscaler free cash flow. That was self-limiting: you can only spend what you earn. In the last few sessions, four data points confirmed the model has fundamentally changed.
Nvidia is launching its first bond deal since June 2021 (when it raised just $5B) — a seven-tranche issue, maturities from 2 to 30 years, with the long bond guided around +90bps over Treasuries, led by JPMorgan, Morgan Stanley and Goldman Sachs. Proceeds: "general corporate purposes," including refinancing. The consequence: this is a company with roughly $100B of cash and free cash flow. Unlike Oracle, Google or Meta, it has no hole to plug. When the strongest balance sheet in technology chooses term debt over its own cash, it is telling you (a) it wants to lock long-duration capital before rates or risk appetite turn, and (b) the entire complex now treats leverage as the default tool. When the bulletproof player borrows, the cycle is maturing — not beginning.
On June 9, Apollo led an initial $35B capital solution for Broadcom's new AI XPV Platform, alongside Blackstone and a syndicate of global banks — structured to enable more than 20 gigawatts of compute through 2028, with the first tranche funding Anthropic's 1GW+ buildout starting mid-2026. Apollo's Jamshid Ehsani called it "the largest private financing ever executed," and described AI compute as "one of the most compelling new asset classes in finance, characterized by contracted cash flows, mission-critical utility." Read that again: a private-credit giant is now underwriting the chips themselves as a fixed-income asset class. Days earlier, KKR, Nvidia, Vistra and the Kuwait Investment Authority committed $10B to a separate vehicle (Helix Digital Infrastructure).
The Bank for International Settlements flagged in March that hyperscaler bond issuance topped $100B in 2025 (mostly >5-year maturities), that CDS spreads have risen — especially for lower-rated issuers — and that a growing share of the buildout is funded off balance sheet: special-purpose vehicles and JVs capitalized by private credit, with debt serviced by lease and capacity-offtake cash flows. The BIS calls these obligations "economically akin to debt but largely residing outside corporate balance sheets," and warns they create new shock-transmission channels between hyperscalers, private-credit funds, insurers, and the banks that backstop them.
JPMorgan estimates roughly $260B of equity issuance in 2026 (SpaceX alone pulled $85.7B after its greenshoe), on top of an AI-driven debt wave. The bull case — and it's a fair one — is that ~$1.5T of expected buybacks plus a record ~$900B of H1 M&A can absorb it; IPO and lockup supply is barely ~1% of a $65T+ market. I largely agree the equity supply gets digested. My concern isn't absorption. It's that the marginal dollar funding AI is now a credit dollar, and credit cycles don't end with indigestion — they end with a gate.
The first crack in a debt-funded buildout never shows up in Nvidia's stock price. It shows up in spreads and redemption gates. It already is:
And the quality of the contracted cash flows is not bulletproof: Anthropic, valued at $965B, is reportedly asking Google — its own rival — to guarantee its 1GW of lease payments before it IPOs. "Contracted" only matters if the counterparty pays.
The cleanest way to be long the AI capex supercycle is to be the lender, not the bettor. The lenders get paid first (contracted cash flows, investment-grade structuring, fat origination spreads) and they get paid whether or not any individual AI bet pays off — right up until the credit cycle turns, which the BDC gates suggest is a 2026–2027 risk, not a today risk.
Apollo arranged the largest private financing ever and sits at the center of the new AI-credit asset class. AUM is $1.03T with fee-generating AUM up 40% YoY, yet the stock trades at just ~17x forward earnings — a growth-of-credit franchise priced like a value stock. At $137.67 it sits mid-range of its $99.56–$157.28 52-week band.
| Entry | Target | Invalidation | Timeframe |
|---|---|---|---|
| $130–138 (add $120–125) | $165 (stretch $180) | weekly close < $118 | 6–12 months |
The higher-quality, deeper-moat sibling (ROE 36%, 3.86% yield, $1.3T+ AUM) and co-anchor of the same deals — but priced for it at ~50x forward, with the BCRED gate as a live overhang. I prefer APO on valuation; own BX on weakness.
| Entry | Target | Invalidation | Timeframe |
|---|---|---|---|
| $115–122 | $150 | close < $108 | 6–12 months |
I flagged AVGO as a long at $360–385 this morning; it's already $393.75. Broadcom is the literal silicon at the center of the Apollo $35B XPV platform — meaning its AI backlog is now committed-capital-funded, which de-risks the very thing the market feared after Q2. Record $30B quarterly AI bookings, Q3 AI revenue guided to ~$16B (+200%).
| Entry | Target | Invalidation | Timeframe |
|---|---|---|---|
| $360–385 | $475 (stretch $500+) | weekly close < $350 | 6–12 months (catalyst: Q3 print early Sep) |
Memory demand is a physical shortage (HBM sold out), not financed demand — the purest fundamental beneficiary. But the stock is parabolic: $1,071, up ~224% YTD, now a ~$1.1T company off a $103 low a year ago. Don't chase the vertical. Earnings ~June 24.
| Entry (on pullback) | Target | Invalidation | Timeframe |
|---|---|---|---|
| $900–950 | $1,250 | daily close < $850 | 1–3 months |
Oracle (ORCL, $194, +5.5% today on a $638B orders headline): the poster child of the wrong side of this theme — free cash flow around −$24.7B TTM, ~$50B FY26 capex, $261B of off-balance-sheet leases, net debt ~$123B (Baa2), and an $18B bond now facing litigation alleging its offering documents were misleading. Trim strength into $195–205; only a long-term value buyer at $160–175. The leveraged neoclouds (CoreWeave) and dilutive server names (SMCI) sit in the same bucket.
The market spent today pricing the end of a war. The more durable story is the beginning of a credit cycle. AI's center of gravity has shifted from the income statement to the balance sheet — and increasingly, to balance sheets you can't even see. That's bullish for the volume and duration of the buildout, which is why I want to own the franchises that get paid to finance it (APO, BX) and the silicon whose backlog is now funded for it (AVGO). It's bearish for the over-levered borrowers (ORCL, neoclouds) and it puts the private-credit BDC complex (OWL, BCRED, FSK) on the watch list as the early-warning system for the whole trade. Own the toll booth. Monitor the gate.