Oracle revenue rose 30%, IaaS revenue 121% and RPO reached $664bn.
Demand and commercialisation remain real.中文|EN
Monitor status awaiting verificationLast checked 21 Sept, 08:31Research through 2026-09-187 pending events
Price pressure repaired quickly. Persistent rejection is not confirmed.
Oracle had fallen 8.23% by T+3 after earnings, but the loss narrowed to 1.54% by T+5. By 18 September, the five tracked price proxies were again above EMA20 and EMA50, with no synchronised operating or financing deterioration. A systemic break is still not confirmed.
Oracle fell -8.23% by T+3, but the loss narrowed to -1.54% by T+5.
T+30 is pending. Without named consensus or synchronised business deterioration, this is not proof of persistent rejection.Oracle spent $28.5bn on CapEx, reported negative $5.4bn FCF, received $11.4bn of financing-like customer prepayments and sold $20bn of equity.
Real growth and overinvestment risk can coexist.What would overturn this view?
Free cash flow recovers, incremental AI profit keeps growing faster than depreciation, lease and financing costs, orders convert normally into revenue, and project cancellations do not spread.
One moving-average break or one weak stock cannot establish a systemic AI-cycle break.DemandExpandingVERIFIED
CommercialisationReal, unevenPARTIAL
Capital returnIncompleteCONCERN
Bubble stateNot declaredNOT_DECLARED
PRICE CLOCK / THROUGH 2026-09-18
Where would an AI bubble break first in price?
SMH is the default view. It represents the semiconductor cycle, not the whole AI industry. Price weakness triggers review; it cannot declare a bubble break on its own.
Adjusted closeEMA20EMA50SMA200Research event
Failed-breakout path hypothesis
A failed breakout, return to the prior range, high-volume support break, failed retest and lower high may describe a top-forming path.
UNVERIFIED · State-transition eligibility:PROHIBITEDBIS: AI investment-race mechanism
Individually rational pre-investment may create industry-wide overbuilding, circular financing and specialised-asset liquidation risk.
VERIFIED · State-transition eligibility:PROHIBITEDHistorical cycle evidence
Price can fall before the business breaks
A recoverable correction and a structural break initially look similar. The distinction appears when revenue, margins, orders and financing either hold or begin to deteriorate.
PRICE PATH
First 100 trading days after the peak
Peak = 100. This compares mechanisms; it does not predict that AI must repeat either year.
1997 recoverable correction2000 structural break
Funding stress does not spread and the market can repair key levels within months.
Multiple independent signs emerge across revenue, margins, orders, inventory or financing.
Remain unclassified. One price event cannot establish a systemic bubble break.
^IXIC
1997 Nasdaq correction
Price indexed to the approved highest adjusted close. Moving-average events remain one price-evidence lineage.
Price / peak
1,745.85
1997-12-24
Approximately 0.2 years
0.4 years after the peak
Price damage was sharp but temporary: the prior peak was recovered without a lasting collapse in Cisco's revenue or gross margin.
A valuation and liquidity correction occurred while operating demand continued to absorb capacity.
A recoverable 1997-style correction cannot be assumed unless earnings, orders and financing remain intact.
Do not guess a top date. Test whether risk is escalating.
Test failed breakouts, support breaks, failed retests, lower highs, breadth and persistent volatility.
Test credit spreads, funding costs, margin pressure and contractual rigidity.
Test orders, guidance, margins, free cash flow, utilisation and project cancellations.
Five decision questions
01Why can a real technology produce a bubble?
Real demand, competitive investment and financing commitments can create overcapacity at the same time.
Why can a real technology produce a bubble?
Real demand, competitive investment and financing commitments can create overcapacity at the same time.
- Current evidence
- The reviewed sample shows real AI infrastructure demand and commercialisation, while CapEx, leases and purchase commitments are rising faster and returns remain uneven.
- What raises concern
- Investment continues to accelerate while incremental revenue, profit and free cash flow fail to cover depreciation, leases, energy and the cost of capital.
- What weakens the thesis
- Hyperscalers keep funding investment from operating cash flow and AI-related profit grows faster than the full capital-cost burden.
- Next observation
- CapEx, depreciation, free cash flow and project pacing in the next hyperscaler updates.
02When does price begin to weaken?
Mid-week price pressure repaired quickly and has not become persistent weakness.
When does price begin to weaken?
Mid-week price pressure repaired quickly and has not become persistent weakness.
- Current evidence
- Oracle had fallen 8.23% by T+3 after earnings, but the loss narrowed to 1.54% by T+5. By 18 September, LITE, COHR, SMH, ANET and QQQ were all back above EMA20 and EMA50, while operating, order and financing evidence had not weakened in sync.
- What raises concern
- A failed breakout becomes a support break, the retest fails, a lower high forms and weakness spreads across suppliers and end customers.
- What weakens the thesis
- Price repairs the trend while earnings expectations, orders and cash flow continue to improve.
- Next observation
- Whether Oracle's T+5 repair holds through T+30 and whether suppliers again break their medium-term trends together.
03What would break market belief?
Not one red candle, but simultaneous failure in orders, returns on capital and financing conditions.
What would break market belief?
Not one red candle, but simultaneous failure in orders, returns on capital and financing conditions.
- Current evidence
- The reviewed sample has not shown broad project cancellation, synchronised operating deterioration or financing contagion.
- What raises concern
- Hyperscaler CapEx cuts, project delays, order cancellations, impairments and stress at shared funding nodes appear together.
- What weakens the thesis
- Investment converts into usage, revenue and cash returns while reliance on external financing declines.
- Next observation
- Project cancellations, uncommenced leases, purchase commitments, debt financing and customer prepayments.
04Why are there so many false bottoms?
Price and liquidity can rebound before the business and financing system have cleared.
Why are there so many false bottoms?
Price and liquidity can rebound before the business and financing system have cleared.
- Current evidence
- Historical cases repeatedly reclaimed short moving averages before the final bottom. All price indicators remain one evidence lineage.
- What raises concern
- Rebound volume fades, old support becomes resistance, fundamentals are revised lower and price makes a new low.
- What weakens the thesis
- Price forms a higher low while orders, margins, free cash flow and financing conditions stop deteriorating.
- Next observation
- T+5 and T+30 after tactical rebounds, plus subsequent earnings revisions.
05What counts as a real recovery?
Price, business conditions, financing and returns on capital must repair separately.
What counts as a real recovery?
Price, business conditions, financing and returns on capital must repair separately.
- Current evidence
- A golden cross or one strong earnings report proves only one layer. Neither establishes fundamental recovery or a return to the old peak.
- What raises concern
- Price recovers only through multiple expansion while revenue, utilisation and cash returns do not improve.
- What weakens the thesis
- A weaker earnings, financing or demand update can overturn the recovery thesis; one moving average cannot establish it.
- Next observation
- EMA50 and SMA200 repair, stabilising earnings expectations, free cash flow and returns on capital.
How a bubble forms
Not investing may lose the customer; everyone investing can create excess capacity
When demand is strong and supply is tight, pre-booking chips, optics, data centres and power can be rational for each company. The risk appears when peers make the same decision and supply arrives together.
- Current support
- LITE, COHR, ANET and hyperscalers still show expanding construction.
- Falsifier
- Usage, revenue and incremental profit keep absorbing new capacity without lower utilisation.
Leases, procurement and power contracts lock today's optimism into future cost
Revenue can disappoint while long-term leases, equipment orders and power commitments remain difficult to cancel. The key comparison is total capital cost versus incremental operating profit and free cash flow.
- Current support
- Capital spending and long commitments are rising; Alphabet, Amazon and Oracle deserve closer cash-conversion review.
- Falsifier
- Free cash flow recovers and AI profit grows consistently faster than depreciation, leases and financing cost.
Not every part of the chain is burning cash for growth
Microsoft's cash coverage, Meta's advertising gains and ANET's margins and cash flow show that real commercialisation can support high investment.
- Current view
- This looks more like a return-on-capital verification period than a demand-free speculative fiction.
- Boundary
- BIS supplies a mechanism; company filings supply current evidence. Neither provides an exact market threshold or top date.
Three observation clocks
Price pressure repaired; persistence is unconfirmed
By 18 September, LITE, COHR, SMH, ANET and QQQ were all back above EMA20 and EMA50. Oracle's T+3 loss narrowed sharply by T+5, while current operating evidence remained intact.
Demand expands; returns remain under review
Optics, networking and cloud commercialisation remain supported, but cash conversion, incremental returns and utilisation are incomplete.
Rigidity is rising; macro confirmation remains incomplete
Data-centre debt, leases and long-term guarantees are increasing. The VIX research proxy fell to 14.81 on 18 September, but credit spreads could not be refreshed, so financing stress is not confirmed.
PRICE CONFIRMATIONNext 2–8 weeks
BUSINESS CONFIRMATION2026 Q3–Q4
SYSTEM TEST2026 Q4–2027 H1
These are conditional validation windows, not a precise top forecast.External signals
Physical demand remains firm and funding is orderly, while local leverage and structural buffers warrant attention
Physical demandSupportive
Dollar fundingNeutral, improving
Korea leverageLocal risk high
Bubble stateUnchanged
Korea HBM
Taiwan$<3.0B
Malaysia$~1.3B
Demand persists while packaging routes shift
Korea-origin HBM flows remain elevated while more volume routes through Malaysia. This supports continued accelerator buildout, but does not directly establish COHR orders, optical demand or end returns on capital.
Method, sources and limits
- SemiAnalysis ChipBookMethodology and monthly Korea HBM physical-flow context
- User-supplied Korea HBM exports screenshotEvent lead only; image not republished
57/ 100
LooseNeutralTight
Funding remains orderly; the structural buffer is thinner
Dollar liquidity is neutral and improving, with an 11.7% reserve buffer. A separate fragility dashboard shows 0/3 funding legs active, while its Treasury basis-crowding proxy stands at 90.
Method, sources and limits
- Market Fragility DashboardContextual fragility and funding-resonance read
- Dollar LiquidityDollar-liquidity stance and structural-buffer context
55.5%Korea leverage thermometer
40%45%
Korea can amplify a selloff without proving a global demand break
The source classifies 55.5% as critical, while its explicit margin-financing ratio is 33.92%. This is a Korea-specific deleveraging risk, not evidence of a systemic AI break.
Method, sources and limits
- KIM Premium public APIKorea retail leverage proxy
Flip 7645SPX 7650.5
Net GEX +$31.56B
Currently dampening, but close to the amplification regime
SPX was in positive gamma but only 5.5 points above the estimated flip. A break below could turn dealer hedging from dampening to amplifying; this is not a long-horizon bubble indicator.
Method, sources and limits
- LevelBrief GEXDelayed SPX gamma structure
A top-forming path
- 01Breakout fails
- 02Price returns to the prior range
- 03Key support breaks on high volume
- 04Retest cannot reclaim support
- 05A lower high forms
- 06Weakness spreads across companies
False bottoms and recovery
First establish that the market is no longer making repeated lows
RSI, EMA20 or one violent rebound show only temporary price relief. A higher low, slower drawdown and broader participation matter more.
Counterevidence: fading rebound volume, old support becoming resistance and a new observed low.Mechanical selling must stop reinforcing itself
If margin finance, leveraged-ETF rebalancing and credit spreads keep worsening, the first rebound may be short covering rather than recovery.
Counterevidence: funding pressure expands and flows remain dominated by forced risk reduction.Reality stops worsening and expectations finish resetting
Sustainable recovery requires orders, revenue, margins or free cash flow to stabilise while the market stops demanding implausible growth.
Next: earnings revisions, order conversion, free cash flow, utilisation and funding cost.Cross-confirmation
High-sensitivity upstream optical signal
Earnings and guidance accelerated; T+30 remained 15.02% above the pre-release closeOptical peer
Demand confirmed; capacity and cash conversion remain concernsNetwork equipment layer
Revenue, margins and cash generation confirm network demandCustom silicon & data center connectivity
Demand and cash support expansion; margins, commitments and shared financing need scrutinyFY2027 Q2 · Reviewed 2026-08-31End-demand and CapEx anchor
Cloud demand grows; capital-return visibility is unevenMRVL: demand grows; returns need verificationView evidence
Designs chips for cloud and network customers, using foundry and assembly partners; spans custom compute, electro-optics, switching and storage.
Quarter ended 2026-08-01 · Results released 2026-08-27 · Reviewed 2026-08-31
- Revenue ($m)
- 2739.3
- Data center YoY (%)
- 46
- Operating cash flow ($m)
- 605.5
- Equipment purchases ($m)
- 126.7
Silicon and connectivity add demand confirmation; data center revenue is not pure AI revenue.
Revenue guidance rises, but adjusted margin guidance is below Q2; growth does not guarantee better unit profitability.
Q2 GAAP / adjusted gross margin: 53.1% / 58.9%. Q3 revenue guidance ($m): 3150 ±5%; adjusted margin guidance: 57.5–58.5%.
Operating cash flow less equipment purchases is positive; factoring, working capital and equity costs still matter.
605.5 − 126.7 = 478.8 $m. A cash-conversion proxy, not fully adjusted FCF; excludes licence, acquisition and equity costs. Q2 receivables sold ($m): 111.1.
Supplier commitments, customer concentration and the shared NVIDIA financing node limit independence; this is not a new independent end-demand source.
Multi-year supplier commitments ($m): 8518.9; NVIDIA preferred investment ($m): 2000. Direct customer / distributor shares: 16% / 44%. Not end-customer concentration. Customer warrants cover up to (million shares): 59, subject to vesting, not immediate dilution.
A second-half custom-business acceleration is management's outlook, not realized revenue.
Foundry cancellations require incurred costs and may add fees; some customer orders can be cancelled or deferred at short notice.
Next checks: Next-quarter revenue and margin versus guidance; Custom production ramp converting to cash; Supplier commitments versus customer cancellations or delays; Changes in factoring dependence and customer equity incentives.
MRVL-only supplement. Named consensus and post-earnings price reaction are not assessed. Other issuer and price dates are unchanged; the bubble state remains not declared.
Optical and networking demand is confirmed. Returns on capital, commitments and financing quality are not yet sufficient to confirm a systemic break.
Case evidence
Cisco / JDSU → LITE
Price paths, moving averages, drawdowns, fundamental lead-lag and a constrained historical mapping.
LITE expectation drift
Initial snapshot, final pre-event freeze, T+1, T+5 and T+30 are now recorded.