The Split Verdict
Broadcom's Q3 (reported post-close September 2) was a referendum that returned a split decision. EPS of $3.32 beat consensus of $3.24. Revenue of $29.59 billion beat estimates and grew 86% year-over-year. The AI revenue trajectory was raised again: $58 billion in FY2026, an outline of $115 billion in FY2027, and $230 billion by FY2028 — with non-GAAP EPS guided toward $30+ versus a $25.86 consensus. On the demand side, nothing broke.
And yet the stock initially fell 5% pre-market on September 3, because the Q4 revenue guide of $34.8 billion came in below the $35.03 billion consensus — and because of what the earnings call revealed about how this growth is being financed.
The beat-but-sell pattern, which NVIDIA briefly broke on August 26, has resumed. Two referendums in eight days, one pass (NVDA), one split (AVGO). The pattern is not noise; it is the market repricing the marginal AI dollar as the funding side of the trade deteriorates.
The Disclosure That Matters More Than the Guide
Buried in the CFO's commentary was the most important single data point of the quarter: Broadcom carries residual value guarantees — contingent liabilities — extended to its AI lab customers. In plain terms: if the equipment's economic value disappoints, Broadcom absorbs part of the loss.
This is the textbook definition of vendor financing, and it is the first explicit marker of the break condition we have been monitoring in the telecom-1999 analog. In that framework, the danger sign was never strong capex — it was capex funded by vendor balance sheets, because it means demand is being pulled forward on credit extended by the supplier itself. Lucent's receivables financing for CLECs in 1999 did not cause the telecom bust, but it was the highest-fidelity leading indicator of it: when the vendor starts guaranteeing the customer's economics, the vendor is no longer pricing demand — it is manufacturing it.
To be precise about what this is and is not:
- What it is: a contingent liability, disclosed, quantified in the risk factors, material enough to warrant CFO airtime. It shifts some AI capex from "end-customer balance sheet" to "supplier balance sheet" — the exact migration that made telecom capex look sustainable two quarters longer than it was.
- What it is not (yet): evidence that demand is fake. Broadcom's AI orders are contracted, multi-year, and backed by named counterparties with real revenue. The demand corroboration this week was overwhelming — Dell beat and raised (+9%), HPE beat, Snowflake blew out, TSMC says its ~20 fabs cannot keep up, Solvay is doubling Taiwan chip-chemical output, Vietnam's AI-driven exports hit records, and Korean semiconductor shipments are up 209% YoY.
The paradox is the point. Demand is real. The funding of the marginal dollar is getting worse.
The Funding Side Is Deteriorating on Schedule
Three funding-side developments landed within 24 hours of the AVGO print:
1. The yen carry unwind completed. USD/JPY reversed violently from 160.27 to roughly 157-158 as markets fully priced a BOJ hike for September 18 (~84% odds) — with US Treasury Secretary Bessent publicly urging Tokyo to hike, a rare instance of Washington lobbying an ally's central bank toward tighter policy. The AI trade's cheapest funding leg is now unwinding market-driven, which is faster and less controllable than intervention. This is a global liquidity drain with no policy off-ramp.
2. The three-legged global bear steepening extended. US 10Y ~4.79% (highest since November 2023), 30Y ~5.24%; JGB 10Y at 3.0% (first time since 1996); Bund 10Y 3.38% (2011 highs); OAT above 4.215% (2008 highs); UK gilts at 2008 levels. Meanwhile the diesel crack spread in Europe blew through $100/bbl — a record — as the Iran conflict broadened to Kuwait and refinery bottlenecks turned an oil shock into a refined-product inflation amplifier. Warsh Fed hike odds sit near 68%.
3. The ECB quantified the euro funding leg. Monday's ECB blog put hyperscaler euro-denominated bonds at roughly €40 billion outstanding (~10% of their total), with reverse-Yankee issuance nearly doubling in a year. The AI capex stack is now explicitly borrowing in three currencies, and every leg of that funding curve steepened this week.
Hyperscaler capex running at 1.3-1.5x operating cash flow against a 30-year Treasury above 5% is a different animal than the same ratio at 3% long bonds. The denominator is moving against the trade.
What the Telecom Analog Actually Says
The framework's break conditions, restated: hyperscaler capex exceeding OCF plus sustainable debt capacity → IG spreads on AI issuers widening past ~200bps → capex guides cut → orders cut → startup funding dries → cloud revenue decelerates → reflexive spiral.
The dashboard today: capex/OCF ~1.3-1.5x (danger threshold >1.8x — not yet), IG spreads ~130bps (stress threshold >250bps — not yet), AI startup funding still strong, GPU lead times 6-9 months (no oversupply signal). Four of five monitors are green. The residual value guarantee does not trip any of them — but it is the first entry in the ledger of the condition that eventually does: supplier balance sheets absorbing customer risk.
The honest read: this is a first marker, not a confirmation. Telecom's vendor financing went from footnote to crisis in about six quarters. The correct response is not to forecast the bust — it is to watch the ledger quarterly and demand that each new marker be smaller than the last.
The Watchlist
- ORCL next week — AI capex referendum #3. Watch for any disclosure of back-end financing structures in its cloud backlog.
- IG spreads on AI-issuer names (AVGO, ORCL, MU, hyperscalers) — the single cleanest funding-stress gauge. >200bps = condition advancing.
- AVGO's contingent liability line item next quarter — growing guarantees = vendor financing scaling; shrinking = one-off.
- JGB/USD/JPY — the carry unwind's second leg. A BOJ hike with a compliant yen is benign; a hike with a surging yen (current path) drains liquidity fastest.
- NFP Friday — the Fed-side binary. Hot print + Brent >$95 + 10Y >4.85% arms the bear-steepening cascade; a soft print is the only near-term path to the Goldilocks revival.
- Treasury buyback operations begin September 9 — the first real test of whether official long-end support can offset the private-sector duration strike.
Positioning
The framework's posture is unchanged but its emphasis shifts: the funding-vs-demand paradox is now the central question of the AI trade, and the evidence is split exactly down the middle. Keep the barbell: long the demand corroboration that is verifiable (financials, energy, issuer debt at wide-enough spreads), short the funding fragility (30Y above 5.5% = max-short trigger), and keep the convexity — because if the vendor-financing ledger starts compounding, the unwind will not announce itself in the data. It will announce itself in the spreads.