CALVERELL PERSPECTIVES

The November 5 Selloff: Earnings Have Held Up. The Cost of Capital Has Not Gone Away.

· · 18 min read

The November 5 collapse in US technology stocks exposed a widening gap between corporate earnings expectations and the prices investors are willing to pay for them. The Nasdaq Composite fell 5.52%, the S&P 500 declined 4.36%, and the Philadelphia Semiconductor Index lost 12.25%. The VIX rose from 16.61 to 28.18. Nvidia lost 9.3% and AMD fell 17.3%.

Conceptual illustration of clear olive-edged optical glass, a silicon wafer and warm ivory limestone.

A historic semiconductor decline, a partial rebound and the growing divide between earnings forecasts and market valuations

US Equity Strategy | Information through the close of 6 November 2026

A market repricing ahead of an earnings downturn

The immediate catalyst was the political uncertainty following the US midterm elections. Investors began reassessing the prospects for data-centre construction, electricity access and regulatory restrictions on AI infrastructure.

But the election result was already known on November 4, when major indices advanced. The subsequent collapse is difficult to explain through politics alone.

By November 3, according to Citadel Securities, forward earnings estimates for the S&P 500 semiconductor constituents had risen approximately 30% since June 30, while forward valuation multiples had fallen 35%. The group's forward P/E stood at 13.8 times, its lowest level since July 2022.

These figures describe a defined S&P 500 semiconductor basket, not the Philadelphia Semiconductor Index. Nevertheless, the broader message is important. Valuations had already deteriorated substantially before the November 5 collapse, without a corresponding decline in published earnings forecasts.

The November 6 economic data added a second constraint.

US employment growth substantially exceeded expectations, reinforcing the possibility of further Federal Reserve tightening. Yet the underlying bond-market evidence is more nuanced than the headline reaction suggests.

Ten-year Treasury real yields actually declined during Thursday's equity collapse, then rose modestly following Friday's employment report. The selloff cannot therefore be explained as a mechanical response to higher real interest rates on the day.

The distinction is central to our assessment.

November 5 was primarily a repricing of equity and financing risk against an already restrictive real-rate environment. November 6 demonstrated that the market may have to absorb that repricing without the immediate benefit of monetary easing.

For long-term investors, the question is whether the deterioration in valuations reflects legitimate doubts about the durability of AI earnings, or whether financial-market pressure has moved faster than the underlying economics of the industry.

What the bond market actually said

US Treasury data provide a useful decomposition of the two trading sessions.

US 10-year Treasury yields and breakeven inflation
US 10-year Treasury measure Nov. 4 Nov. 5 Nov. 6
Nominal yield 4.90% 4.84% 4.86%
TIPS real yield 2.64% 2.60% 2.63%
Breakeven inflation 2.26% 2.24% 2.23%

Daily Treasury par yields. Breakeven inflation is calculated as the difference between nominal and inflation-indexed par yields of the same maturity. Values are rounded to basis points and do not represent a separate, directly traded inflation forecast.

On November 5, the nominal 10-year Treasury yield declined six basis points. Four basis points came from lower real yields, with the remaining two reflecting lower breakeven inflation.

Ordinarily, falling real yields provide some support to the present value of long-duration corporate earnings. Semiconductor equities nevertheless suffered one of their worst sessions on record.

This is evidence that changes in earnings durability, risk premiums, liquidity or positioning outweighed the modest improvement in the risk-free discount rate.

The following day, the pattern reversed.

The 10-year real yield rose three basis points to 2.63%, while nominal yields increased two basis points. Breakeven inflation declined another basis point.

The movement was consistent with the stronger employment report reinforcing expectations of tighter monetary policy, rather than a sudden increase in market-implied inflation compensation.

There is also a longer-term constraint. The 10-year real yield stood approximately 80 basis points above its level a year earlier. Although real yields had fallen from their October highs, the inflation-adjusted cost of capital remained elevated.

This matters for AI infrastructure in two ways.

First, higher required returns reduce the present value of future corporate cash flows. The sensitivity is particularly important for businesses whose valuations depend on profits expected many years ahead.

Second, higher real financing costs can change the economics of actual investment decisions. A data-centre project expected to become profitable under one cost-of-capital assumption may become marginal if financing remains expensive or commercial deployment is delayed.

The two effects should not be conflated. The Treasury real yield is only one component of a company's cost of capital. Equity risk premiums, credit spreads, debt structure and project-specific risk remain important.

The November 5 market reaction suggests those additional components were unusually influential.

Strong employment, mixed inflation signals

The October employment report released November 6 showed nonfarm payroll growth of 165,000, against expectations of approximately 65,000. The unemployment rate fell from 4.2% to 4.1%, while revisions added a combined 84,000 jobs to the preceding two months.

Markets increased their assessment of the likelihood of a December Federal Reserve rate increase. Reuters reported that the probability reached approximately 80% following the release, although estimates differed by observation time and data provider.

The report weakened the case for an imminent shift toward easier financial conditions.

Other releases were less threatening to the inflation outlook. The University of Michigan's preliminary November survey showed one-year inflation expectations declining from 4.7% to 4.3%, while five-year expectations edged down from 3.5% to 3.4%.

Third-quarter nonfarm productivity increased at a 3.6% annualised rate, accompanied by a 0.7% decline in unit labour costs.

These observations suggest some relief in underlying inflation pressures, although they do not establish that inflation is under control.

The immediate implication for technology valuations is that strong earnings may be necessary but insufficient for a sustained recovery. Investors also need evidence that the real cost of capital has stopped rising, or that improving earnings can offset it.

Friday's rebound recovered less than a third of Thursday's semiconductor losses

The November 6 rally deserves a more careful interpretation than the headline percentage suggests.

The Philadelphia Semiconductor Index rebounded 4.21% to 9,390.25 after closing at 9,011.14 the previous day.

That followed a decline from 10,268.65 on November 4.

Philadelphia Semiconductor Index closing levels
Philadelphia Semiconductor Index Closing level
November 4 10,268.65
November 5 9,011.14
November 6 9,390.25

Thursday's decline amounted to 1,257.51 index points. Friday's recovery added 379.11 points.

The rebound therefore recovered only 30.1% of the preceding day's point loss. The index remained approximately 8.6% below its November 4 close.

The difference between a 12.25% fall and a 4.21% rebound is economically important. Percentage returns are calculated from different starting prices. A subsequent gain of 4.21% does not offset a preceding loss of 12.25%.

Nor was the recovery uniform.

Nvidia rose 2.78% on Friday, recovering approximately 27% of its November 5 dollar-price decline. AMD rebounded 11.04%, recovering approximately 53% of its previous day's loss.

AMD experienced a substantially larger technical rebound, but it had also suffered a much more severe initial decline.

The contrast illustrates the risks of interpreting percentage gains in isolation. A large rebound following an extreme decline may reflect short covering, position rebuilding and temporarily improved liquidity without establishing that the company's fundamental outlook has changed.

Nomura had estimated that leveraged ETFs might face approximately $47 billion of rebalancing-related selling demand near Thursday's close. The estimate was not a confirmed measure of completed transactions, and it cannot be used to assign a precise share of the index decline to forced selling.

Nevertheless, the combination of exceptional volatility, concentrated exposures and substantial potential rebalancing requirements provides a credible mechanism for price movements to overshoot changes in underlying earnings expectations.

Friday's recovery is consistent with some of that pressure easing.

It does not prove that the broader deleveraging process is complete. The semiconductor index remained deeply below its preceding close, and the evidence from credit markets and corporate capital spending had not materially changed.

The more useful observation is that the market recovered part of an extreme liquidity shock while leaving most of the previous day's losses intact.

Earnings estimates remain resilient, but the comparison must be made carefully

Publicly available analyst estimates continue to show upward revisions for Nvidia and AMD.

For Nvidia, FY2028 EPS consensus increased from $15.68 approximately 30 days earlier to $15.91 in the latest publicly displayed estimate snapshot.

For AMD, calendar 2027 EPS consensus increased from $15.72 to $16.78.

These are different corporate reporting periods, so the resulting valuation multiples cannot be treated as directly comparable next-twelve-month measures.

A rigorous cross-company NTM comparison would require dated quarterly consensus estimates for the same four-quarter forward window, the same earnings definition and consistently timestamped share prices.

The necessary fully archived quarterly series is not available in the public data reviewed for this report.

Rather than construct a synthetic NTM series and present it as actual analyst consensus, we use a narrower comparison: changes in each company's own valuation against the same forecast earnings period.

The price observations are October 8 and November 6, approximately one month apart. Earnings estimates use the closest publicly available current and 30-day-prior snapshots.

Within-company earnings and valuation observations
Measure Nvidia AMD
Earnings period FY2028 CY2027
EPS estimate, approximately 30 days earlier $15.68 $15.72
Latest displayed EPS estimate $15.91 $16.78
EPS estimate revision +1.5% +6.7%
October 8 closing price $230.48 $620.68
November 6 closing price $186.64 $449.04
Earlier price / earlier EPS 14.7× 39.5×
Latest price / latest EPS 11.7× 26.8×
Indicative multiple contraction -20.2% -32.2%

The table is a within-company valuation comparison. It is not a harmonised NTM peer-valuation table. The earnings observations use rolling historical snapshots rather than an institutionally archived point-in-time forecast database; accordingly, the multiple changes are indicative. Earnings periods differ across companies, and the historical estimate snapshot is not guaranteed to coincide exactly with the October 8 market close.

Within those limits, the result is informative.

Nvidia's selected fiscal-year earnings estimate improved modestly while its share price declined approximately 19% over the observation period. Its indicative multiple contracted by about one-fifth.

AMD experienced a larger earnings upgrade but a much sharper share-price decline, producing a multiple contraction of approximately one-third.

The calculations reinforce the conclusion from Citadel Securities' separate semiconductor-basket study: valuation compression has substantially outpaced the deterioration, if any, in the publicly observed earnings estimates.

They do not demonstrate that either company is undervalued.

Nvidia's FY2028 estimate depends on continued earnings expansion through a future fiscal year. AMD's calendar 2027 estimate incorporates expectations for rapid growth in its accelerator and data-centre business.

A lower multiple is attractive only if the earnings denominator is sufficiently reliable.

It is also possible for the market to reduce its valuation of profits expected beyond 2028 without materially changing forecasts for 2027 or 2028. Long-term competitive conditions, hardware pricing and the profitability of AI applications may have changed in investors' assessments before those changes become visible in near-term consensus.

This is the central unresolved question in the semiconductor selloff.

Capital expenditure has not collapsed, but its cost has become more visible

The latest disclosures from major cloud platforms still indicate substantial investment.

Alphabet maintained 2026 capital expenditure guidance of $195 billion to $205 billion and anticipated significant further spending in 2027. Amazon maintained approximately $220 billion of cash capital expenditure for 2026. Meta raised the lower bound of its 2026 guidance to $135 billion.

Microsoft expected approximately $170 billion of calendar-year investment while acknowledging delays in the delivery of some third-party capacity.

These observations contradict the claim that the largest cloud providers have already collectively abandoned their AI expansion programmes.

They also do not settle the question of future returns.

The economic problem is the relationship between incremental cash generation and the investment required to produce it.

AI infrastructure requires large upfront commitments to computing equipment, electricity, networking and physical facilities. Revenues may arrive over several years, while equipment becomes less competitive as new generations are introduced.

A project can remain commercially useful without delivering an adequate financial return to its owners.

Higher real yields increase the required return on new capital. Wider credit spreads can make the problem substantially worse for independent computing operators.

CoreWeave's five-year credit default swap spread reached approximately 1,267.9 basis points on November 5, extending credit concerns already visible in October.

Its credit position should not be extrapolated directly to cash-generative companies such as Microsoft or Alphabet.

The more important transmission mechanism is whether expensive financing causes independent operators to defer projects, and whether such deferrals eventually affect semiconductor orders.

There is currently stronger evidence of pressure on financing conditions than of a coordinated reduction in hyperscaler capital expenditure.

For that reason, we continue to distinguish a valuation-led correction from a confirmed sector-wide earnings downturn.

A recovery thesis that can be tested

Our base case remains a volatile, selective recovery rather than an immediate return to the previous broad AI-led rally.

The case rests on resilient published earnings estimates, substantial ongoing cloud investment and evidence that market positioning amplified Thursday's decline.

The case is conditional. It should change if earnings revisions, capital expenditure plans, credit spreads or real yields begin moving persistently against it.

For an investment committee, monitoring rules need to be observable and consistently applied. They also need to distinguish ordinary market noise from changes large enough to warrant reassessing the thesis.

We propose using rolling historical distributions rather than untested fixed thresholds.

For each market series, calculate the change over 20 trading days and compare it with the distribution of equivalent changes during the preceding 252 trading days. A movement exceeding two historical standard deviations in the adverse direction would trigger a formal review, subject to confirmation in the underlying data.

These are proposed analytical rules, not empirically calibrated forecasts. Their reliability and false-positive rates must be assessed against a longer historical sample before being used as automated trading signals.

Investment committee monitoring framework

Investment committee monitoring framework
Variable Measurement Review trigger
US 10-year real yield Daily Treasury TIPS yield; rolling 20-session change Increase exceeding two standard deviations of historical 20-session changes
Semiconductor NTM EPS Consistent, fixed-methodology consensus series for SOX constituents Negative revision exceeding two historical standard deviations, accompanied by negative revisions across a majority of index earnings weight
Hyperscaler 2027 capex Sum of comparable, company-reported spending guidance, adjusted for reporting differences Aggregate guidance reduction confirmed by company disclosures and affecting more than one major provider
AI infrastructure credit Five-year CDS changes and underlying cash financing conditions Abnormal spread widening confirmed by weaker debt issuance terms, refinancing difficulty or project deferrals
Market repair SOX price recovery, breadth, volatility and relative performance Price rebound without corresponding stabilisation in earnings and credit indicators

The historical-volatility approach has limitations. One year of daily observations may be insufficient for structural changes, and overlapping 20-day windows are not statistically independent.

The rules therefore identify conditions requiring renewed analysis. They do not establish that every threshold breach signals a durable regime change.

Corporate guidance requires a different treatment. Changes in company forecasts occur irregularly, and a strict statistical threshold may be misleading when disclosure conventions vary.

For capital expenditure, the decision rule should focus on confirmed aggregate changes after normalising the treatment of cash expenditure, finance leases and fiscal-year reporting.

Credit signals should similarly be confirmed against actual financing transactions and investment decisions. A wider CDS spread alone does not prove that a company's projects have become uneconomic.

How the thesis would change

Three developments would strengthen the recovery case.

First, Nvidia's November 17 earnings release would need to support the current trajectory of revenue and profitability without evidence of significant customer procurement delays.

Second, major cloud providers would need to maintain their investment plans, allowing the market to separate isolated financing problems from aggregate demand.

Third, real yields and credit spreads would need to stop imposing additional valuation pressure. A recovery in earnings expectations can support equity prices, but persistent increases in required returns would limit the scope for multiple expansion.

The adverse case would become more persuasive if semiconductor earnings revisions turned broadly negative, project deferrals began affecting actual orders and real borrowing costs remained elevated.

Under those conditions, the market would be moving from uncertainty about future returns to observable deterioration in corporate cash-flow expectations.

This distinction determines whether the current decline provides a valuation opportunity or merely reflects lower sustainable earnings power.

The investment judgement

The most important feature of the November 5 selloff is the timing of the repricing.

Across the semiconductor sector, share prices and valuation multiples have already deteriorated considerably. Yet the published earnings estimates available through early November have remained relatively resilient.

Thursday's collapse occurred despite a four-basis-point decline in the 10-year real Treasury yield. That is strong evidence against a purely interest-rate-driven explanation of the day's losses.

Friday brought a partial recovery, but the Philadelphia Semiconductor Index regained only 30.1% of the previous session's point loss. Meanwhile, the employment report lifted real yields and reinforced expectations of further monetary tightening.

The result is a more demanding recovery environment than the earlier decline in valuations might suggest.

Nvidia's November 17 earnings release will test whether the expected profitability of the AI investment cycle remains intact. The forthcoming inflation data and December Federal Reserve decision will help determine whether the real-rate environment can support a sustained recovery in valuation multiples.

Our current assessment is that the market has repriced the risk of an AI earnings slowdown before such a slowdown has become broadly visible in the earnings data.

That creates the possibility of mispricing, particularly among financially resilient companies. It also means the downside cannot be evaluated from low P/E ratios alone.

The November 5 decline was not caused by rising real yields. The next phase of the market will depend on whether earnings can remain strong enough to withstand real yields that are already high.

OCEAN CAPITAL

Market Intelligence

7 November 2026

Research methodology: US Treasury nominal and real yields are daily par-curve observations. Breakeven inflation is nominal yield less the corresponding real yield. Semiconductor price changes are calculated from historical closes. Publicly displayed fiscal-year EPS estimates are used only for within-company comparisons; they are not represented as harmonised NTM consensus. The monitoring framework specifies proposed review rules whose historical calibration remains to be completed.

Disclaimer

This publication is provided for general informational and research purposes only. It does not constitute investment advice, a recommendation, an offer, or a solicitation to buy or sell any security, fund, asset, or financial instrument. The views expressed reflect the assessment of Ocean Capital Research as of the date of publication and may change without notice. All information is drawn from sources believed to be reliable, but no representation or warranty is made as to its accuracy, completeness, or timeliness. Past performance is not indicative of future results. Readers should conduct their own due diligence and consult qualified advisers before making investment decisions.

Originally published at Ocean Capital Insights.