For two years, AI capital spending was a virtue the market paid for — the bigger the number, the higher the multiple. In June 2026 the sign flipped. Broadcom reported AI revenue up 143% and still fell ~14% because its next-quarter guide came in ~$1.2B under consensus and it would not raise the full-year target.[1] Oracle grew revenue 17% to a record $67.4B and posted free cash flow of negative $23.7B on capex of $55.7B (+162%).[2] Goldman's Jim Covello put the new mood in one line: at some point, you have got to make money.[3] The capex-to-revenue gap widened to ~46%, exceeding the 32% of the 2001 telecom bust (Allianz), and the BIS compared the buildout to canal mania and the dot-com era.[4][6] Nothing about the technology changed. The question changed — from how fast can you build, to how much did it return.
The tell was Broadcom. On June 4, 2026, it reported AI semiconductor revenue of $10.8B, up 143% year over year — an unambiguous blowout — and the stock fell roughly 14%.[1] The market was no longer paying for growth in AI; it was paying for growth that beat an already-euphoric bar, and Broadcom's next-quarter AI guide (~$16.0B vs ~$17.2B expected) plus an un-raised full-year target missed it.[1] A quarter that would have sent the stock up 15% a year earlier sent it down 14%.
This is the signature of a rotation, not a crash. Across June 2026 the market re-priced the same companies on a new question. Oracle grew to a record $67.4B in revenue and a $638B backlog — and traded down, because its capex hit $55.7B (+162%) and free cash flow fell to negative $23.7B.[2] Investors had stopped reading capex as ambition and started reading it as an unpaid bill.
The reframing had names attached. Goldman's Jim Covello: at some point, you have got to make money.[3] A Forbes analysis put the AI capex-to-revenue divergence at ~46%, past the 32% peak of the 2001 telecom cycle, citing Allianz.[4] Acadian's Owen Lamont noted S&P 500 long-term earnings-growth expectations had reached 20.2%, above the 18.6% peak of the 2000 dot-com bubble.[5] And on June 28 the Bank for International Settlements — the central banks' central bank — compared the ~$1T buildout to canal mania and the dot-com era, warning it could turn a capex boom into a protracted bust.[6]
The honest other side: this is a repricing, not a verdict. JPMorgan called the spend durable and increasingly profitable for now, and the bull case is not empty — Nvidia data-center revenue hit a record $75.2B (+92%) and Microsoft's AI run-rate reached ~$37B, with cloud capacity described as constrained through 2026.[7] The rotation is a change in what the market is willing to pay for, arriving on a few weeks of prices. Whether it hardens into a regime is the open question the capstone holds — this case documents the turn, not its destination.
A record AI quarter met a euphoric bar and a soft guide, and the stock fell ~14%.[1] A year earlier the same numbers were a buy signal. The metric did not change; the question the market asked of it did.
Three weeks in which the market re-priced AI capital spending from virtue to liability.
For two years the market rewarded AI capital spending as proof of ambition — the larger the number, the higher the multiple. Hyperscaler 2026 capex would reach ~$725B, up ~77% from ~$410B in 2025.[7]
The BuildBroadcom reports AI semiconductor revenue of $10.8B, up 143% — and the stock falls ~14%. The next-quarter AI guide (~$16.0B) missed consensus (~$17.2B) and the full-year target was not raised. A blowout was no longer enough.[1]
The TellOn Goldman's podcast, head of global equity research Jim Covello sharpens the ROI question: at some point, you have got to make money. A hot jobs report the same day adds a rate-hike scare, and the Nasdaq has its worst day in about a year.[3]
Oracle posts record FY2026 revenue of $67.4B and a $638B backlog — but capex of $55.7B (+162%) drives free cash flow to negative $23.7B. The market reads the spend as an unpaid bill, not a moat.[2]
The BillThe Bank for International Settlements compares the ~$1T AI buildout to canal mania, the 1840s railways, and the dot-com era, warning it could turn a capex boom into a protracted investment bust. The central banks' central bank stops just short of the word bubble.[6]
SystemicAt some point, you have got to make money. — Jim Covello, Head of Global Equity Research, Goldman Sachs, June 2026
| Dimension | Evidence |
|---|---|
| Revenue (D2) Origin · 90 | The lever is capital-market repricing: the same capex, re-read from ambition to obligation. Broadcom fell ~14% on a record AI quarter; Oracle traded down on record revenue because free cash flow was negative $23.7B.[1][2] D2 is the origin because nothing operational or technological changed in June — what changed is the price the market puts on a dollar of AI spend, and whether it demands a dollar of return with it.The Repricing |
| Operational (D6) L1 · 84 | The repricing exposed an operational fact that was always there: capex is outrunning cash. Oracle capex $55.7B (+162%) drove free cash flow to negative $23.7B; hyperscaler 2026 capex reached ~$725B (+77%).[2][7] D6 amplifies from D2 because the market did not learn a new number — it started weighting a number it had been discounting. Epoch AI dates the aggregate hyperscaler free-cash-flow crossover to around Q3 2026.The Cash Reality |
| Customer (D1) L1 · 82 | The customer here is the investor, and the demand shifted from buildout to return. Covello: at some point, you have got to make money.[3] The Forbes/Allianz framing put the capex-revenue divergence at ~46%, past the 2001 telecom peak.[4] D1 amplifies alongside D6: the demand-side question (where is the return?) is the mirror of the supply-side fact (the cash is going out). |
| Quality (D5) L2 · 76 | The standard of proof rose. A quarter now has to beat a bar set by 20.2% expected long-term earnings growth — above the 2000 peak — to be rewarded.[5] D5 is where the rotation becomes a ratchet: each euphoric print raises the bar the next one must clear, so the same growth rate delivers a smaller and eventually negative reward. Broadcom cleared 143% growth and still missed the bar. |
| Employee (D3) L2 · 70 | The AI-attributed layoffs that the spend was meant to justify became part of the ledger under scrutiny — Oracle cut ~21,000 roles in the same year it posted negative free cash flow.[2] D3 sits here because the workforce reductions were sold as the ROI of the capex; when the market began demanding that ROI, the cuts stopped being evidence of efficiency and started being one more line item that had not yet paid for the build. |
| Regulatory (D4) 64 | D4 is the longest-lag dimension: the systemic/financial-stability layer. On June 28 the BIS compared the buildout to canal mania and the dot-com era and flagged circular financing and non-bank credit exposure as structural fragilities.[6] When the central banks' central bank starts drawing bubble analogies, the concern has migrated from equity desks to the institutions that manage systemic risk — the slowest but highest-stakes dimension to move.Watch — Systemic |
The cascade originates in D2 — Revenue — because the lever is a change in how capital markets price the same spending: capex re-read from ambition to obligation.[1][2] From D2 it amplifies into D6 (the operational reality it exposed — record capex, negative free cash flow) and D1 (the demand-side question, investors asking for returns rather than buildout) together, then D5 (the new standard of proof — a blowout must now beat a euphoric bar) and D3 (the AI-attributed layoffs the spend was supposed to justify). D4 (the systemic/regulatory dimension — the BIS financial-stability warning) is the longest-lag dimension. Cross-references: [UC-063] is the prognostic that called this — the market would stop rewarding AI-driven cuts — now validated at the capex level; [UC-220] is the buildout this reckons with; [UC-247] is the compute-layer moat underneath it. The counter-cascade is deliberate: [UC-044] — if energy-per-token keeps collapsing, the whole bear read is the fragile number. Watch it.
-- UC-251: The Show-Me Rotation: 6D Diagnostic Cascade
-- AI capex re-priced from virtue to liability (cluster: UC-252/253/254/255; counter: UC-044)
FORAGE show_me_rotation
WHERE capex_reread_as_liability = true
AND blowout_met_euphoric_bar = true
AND free_cash_flow_negative = true
ACROSS D2, D6, D1, D5, D3, D4
DEPTH 3
SURFACE show_me_rotation
DIVE INTO capital_repricing
WHEN market_demands_return = true
AND technology_unchanged = true
TRACE virtue_to_liability_cascade
EMIT show_me_rotation_signal
DRIFT show_me_rotation
METHODOLOGY 88
PERFORMANCE 52
FETCH show_me_rotation
THRESHOLD 1000
ON EXECUTE CHIRP high 'In June 2026 Broadcom reported AI revenue up 143% and fell 14%, Oracle posted negative $23.7B free cash flow, and the BIS compared the buildout to canal mania; the market stopped paying for the build and started demanding the return - the technology did not change, the question did'
SURFACE analysis AS jsonRuntime: @stratiqx/cal-runtime · Spec: cal.semanticintent.dev · DOI: 10.5281/zenodo.18905193
Broadcom's +143% AI quarter was a buy signal in 2025 and a sell signal in June 2026. Nothing about the number moved; what moved was what the market was willing to pay for it. A rotation is a change in the question, not the data.[1]
For two years, a bigger capex number meant a bigger moat. Oracle's $55.7B and negative $23.7B free cash flow got read the new way — as a bill that has to be paid back, on a timeline the returns have not yet met.[2]
S&P long-term earnings-growth expectations hit 20.2%, above the 2000 dot-com peak, against ~7% realized historically.[5] When the expected bar clears the highest it has ever been, even a blowout can disappoint. The rotation is the gap between the two closing.
The bull case is not empty — Nvidia data-center revenue set a record, Microsoft's capacity is constrained, JPMorgan calls the spend profitable for now.[7] June documents a change in sentiment on a few weeks of prices. Whether it hardens into a regime is the open question, not a settled one.
Eight sources: Broadcom and Oracle primary filings for the two anchor prints, named analyst and institutional commentary (Goldman's Covello, Allianz via Forbes, Acadian's Lamont, the BIS annual report), and the bull-side balance (JPMorgan, Nvidia/Microsoft). Both the rotation and the case against it are cited.
The spend was never the question. The return always was. Know which one you are being paid for.