Research note · Published 14 November 2026 · Market data to the 13 November close

Update, 21 November 2026 · 11 月 21 日更新。Nvidia beat its own forecast, yet the stock fell 23% in the week as a reported SEC probe turned to accounting, and several verdicts below have changed. Read the update (in Chinese).
英伟达财报超预期,但据报道 SEC 调查涉及会计问题,股价一周跌 23%,下文部分操作建议已调整。阅读中文更新。

Nvidia after Huang: how much of the demand is real?

The stock has lost a third of its value in five weeks on four separate shocks. One of them decides what the company earns next year: the charge that Nvidia has been financing its own customers to keep chips moving. This note separates what the filings prove from what is only reported, tests the “fake demand” claim against data Nvidia does not control, and works out what today’s price already assumes.

Read this first. This note was researched and written by Claude, an AI model made by Anthropic. Anthropic is one of the AI labs discussed below and is in the middle of a share offering, so treat anything said about it as coming from an interested party; the note takes no view on its shares. Nothing here is personal investment advice: it knows nothing about your holdings, time horizon or tax position, and the author is not a licensed adviser. The Wall Street Journal investigation is behind a paywall and was read only through other outlets’ accounts of it; where a claim rests on that, the text says so. Nvidia reports third-quarter results on 17 November, which may overtake several figures here.

The short versionFive conclusions

  1. The financing is real, large and in the filings.Nvidia’s own quarterly report shows $530.5bn of commitments and guarantees at 26 July, against $184bn a quarter earlier. About $190bn of it props up customers. A larger piece, $279bn, is supply Nvidia has promised to buy.
  2. The demand is not a hoax.Cloud revenue, AI-lab revenue and the price of renting a GPU all accelerated this autumn, and Nvidia sets none of them. Compute is scarce. Chips no longer are: the bottleneck has moved to power and to capital.
  3. The growth rate is what gives way.Management said a quarter of next year’s business comes from labs it backs with its balance sheet, and its supply contracts are sized for 70% growth. The executive who reportedly tried to shrink that backing now runs the company. I expect the growth outlook to come down.
  4. The price already assumes a lot of that.At $162.85 the shares trade on 17.5 times this year’s expected earnings. By my arithmetic that discounts roughly a one-in-three chance of an outright fall in revenue next year. My own weighting puts fair value within about 10% of the price: fairly priced for a very wide range of outcomes, not cheap.
  5. The risk sits in one layer of the AI trade, not all of it.The evidence supports cutting concentration in companies that live on borrowed money and vendor support, and it does not support abandoning AI. Chip and infrastructure stocks fall together, so several of them amount to one position.
Share price, 13 Nov close
$162.85
−31.9% from the 6 Oct closing high of $239.24
Market value
$4.0tn
about $1.9tn below the peak
Price to earnings
17.5×
on FY27 consensus EPS of $9.32; 23.2× trailing
Commitments and guarantees
$530.5bn
$184bn one quarter earlier
Days to collect a sale
60
45 one quarter earlier
Profit turned into cash, Q2
40%
$24.1bn operating cash flow on $59.7bn net income

Action list · added 14 NovEvery Nvidia-linked stock: valuation, verdict, action

One line per stock, grouped by verdict. Prices are the 13 November close. “P/E” is the price divided by four times next quarter’s consensus earnings per share. “More attractive below” is the price at which that multiple would leave what I consider a margin of safety; it is not a forecast. Trim means cut to the size cap in section 8. Avoid means no new money, and any existing holding kept only at a size you could lose. This is my judgement from public data, not personal advice.

StockPrice, fall from high, P/EAction and trigger
Add on dips: pays from its own cash flow, or owns what is scarce
Alphabet
GOOGL
$295.79
−28%
23×
Best-placed gainer: its own chips, cloud revenue up 103%. More attractive below about $255 (20×). Watch free cash flow, which has been negative for two quarters.
Microsoft
MSFT
$524.31
−6%
27×
The only big buyer still producing free cash. It has barely fallen, so there is no hurry; more attractive below about $486 (25×).
TSMC
TSM
$398.63
−18%
19×
Makes the chips whoever wins. More attractive below about $350 (17×). Risks: Nvidia cuts that rivals do not backfill, and Taiwan.
Hold: fairly priced, or waiting on data
Nvidia
NVDA
$162.85
−33%
16.5×
Do nothing before 17 November. Add afterwards only if next year’s outlook stays at +40% or better with days-to-collect at 60 or less, or if the price falls below about $135. Trim if the outlook is withdrawn or collection passes 65 days.
Broadcom
AVGO
$346.66
−30%
23×
Gains share in custom chips, but its order book leans on OpenAI. Results 9 December. More attractive below about $300 (20×).
Amazon
AMZN
$266.61
−7%
28×
AWS is speeding up, but free cash flow is −$42bn over twelve months and this is the richest multiple among the buyers. More attractive below about $222 (23×).
Meta
META
$641.99
−18%
19×
Cheapest of the buyers, with no cloud revenue to pay for its capex. Watch how the spending is funded.
Arista
ANET
$197.37
−9%
39×
Its customers pay from cash flow, and it gains if buyers move to open networking. The price already says so; more attractive below about $150 (30×).
Marvell MRVL
Lumentum LITE
$156.74, −52%, 38×
$597.12, −48%, 26×
Custom chips plus optics. Each Lumentum share becomes 4.5 Marvell shares, worth $705 today, so Lumentum is 15% cheaper for the same company. The gap is the price of a deal that closes in 2028.
Dell
DELL
$344.36
−42%
13×
Record AI orders, thin margins, and it ships what Nvidia ships. Results on 24 November decide it.
Vertiv
VRT
$202.04
−47%
22×
Power and cooling are the real bottleneck. The risk is local opposition slowing new sites. Add only if fourth-quarter orders hold up.
Trim: peak earnings, or priced for perfection
Micron
MU
$778.40
−38%
5×
The 5× multiple sits on peak earnings: revenue is 4.8 times last year’s. Memory is the bulk of Nvidia’s $279bn of supply commitments, so it is first in line if Nvidia buys less. Keep only what you could hold through a halving.
AMD
AMD
$534.09
−19%
47×
A share gainer on paper, at the highest multiple here, and it needs a 37% jump in next quarter’s earnings. More attractive below about $345 (30×).
Avoid: inside the financing loop, or earnings the market does not trust
CoreWeave
CRWV
$53.11
−62%
loss-making
Debt of $47bn is 1.6 times its market value. A good quarter, but it lives on Nvidia’s support and open lenders. Speculative size only.
Nebius
NBIS
$115.81
−61%
loss-making
The better small cloud: contracts with Microsoft and Meta, and Nvidia owns 9.3%. Still about 9 times this year’s sales and tied to the same funding window. Small size if at all.
Oracle
ORCL
$125.40
−50%
17×
Looks cheap. Free cash flow was −$23.7bn last fiscal year and the backlog rests on OpenAI.
Super Micro
SMCI
$26.87
−48%
4×
A 4× multiple means the market does not believe the earnings: thin margins, heavy funding needs, reported order cancellations.
Intel
INTC
$59.67
−58%
65×
Not an AI-demand story, and its October outlook was weak. Nvidia’s shareholding does not change its earnings.

Who is most likely to gain from Nvidia’s trouble

  1. Alphabet. It sells the main alternative chip and rents it out. Every doubt about Nvidia strengthens its hand with customers.
  2. Broadcom and Marvell (with Lumentum). They design the custom chips the big clouds switch to, and custom-chip vendors reportedly stand ready to take any TSMC capacity Nvidia gives up.
  3. Microsoft, Amazon and Meta. As buyers they get better prices and terms. If Nvidia stops propping up small clouds, that demand comes back to the large ones.
  4. AMD. The second source for general-purpose AI chips. The gain is real and already in the price.
  5. Arista. It gains if buyers pair custom chips with open networking in place of Nvidia’s full system.
  6. Buyers of compute. AI labs and software firms pay less if Nvidia stops supporting prices. TSMC is roughly neutral: it sells to the winner either way.

One caution. On 13 November AMD and Broadcom fell further than Nvidia did. The market is treating this first as a scare about total AI spending and only second as a shift in share. These companies gain relative to Nvidia; they gain outright only if the big buyers’ budgets hold. The likely losers are the memory makers, the small AI clouds, Oracle and the server assemblers.

1 · What happenedFour shocks, and only some of them change how many chips get bought

Nvidia daily closing price

US dollars, 1 October to 13 November 2026. Numbers mark the events listed below.

  1. 16 Oct · Peak. Closing high of $239.24, a market value near $5.9tn.
  2. 221–30 Oct · Capital. Big-tech results show AI spending running past cash flow: capex at the four largest buyers was about 115% of their operating cash flow, and Alphabet, Amazon and Meta reported negative free cash flow.19
  3. 35 Nov · Politics. Two days after the US midterm elections Nvidia fell 9.3%, its worst day since January 2025, on expectations of tougher AI and data-centre rules. Of 45 ballot measures to restrict data centres, 44 passed, by one consultancy’s count.23
  4. 48–10 Nov · Leadership and supply. Jensen Huang told staff of an extramarital affair; the board opened an independent review of reports that he had been targeted by a foreign intelligence effort, and said it had no reason to think company secrets were compromised. A Taiwanese paper reported, from one unnamed source, that Nvidia may slow orders at TSMC.12
  5. 512 Nov · Resignation. Huang left as chief executive and director after 33 years. Finance chief Colette Kress became interim CEO and Harvey Jones the first board chair. The shares opened 13% lower and closed down 4.3% on about five times normal volume.3,10
  6. 613 Nov · Financing. Accounts of a Wall Street Journal investigation describe Nvidia funding customers to absorb supply it had already ordered.7
Data table: daily closes
DateCloseChange
2026-10-01$230.86
2026-10-02$233.95+1.3%
2026-10-05$238.90+2.1%
2026-10-06$239.24+0.1%
2026-10-07$237.47−0.7%
2026-10-08$230.48−2.9%
2026-10-09$229.28−0.5%
2026-10-12$226.32−1.3%
2026-10-13$228.52+1.0%
2026-10-14$221.30−3.2%
2026-10-15$217.23−1.8%
2026-10-16$214.13−1.4%
2026-10-19$217.09+1.4%
2026-10-20$216.47−0.3%
2026-10-21$221.97+2.5%
2026-10-22$212.21−4.4%
2026-10-23$210.96−0.6%
2026-10-26$216.00+2.4%
2026-10-27$217.94+0.9%
2026-10-28$215.38−1.2%
2026-10-29$213.66−0.8%
2026-10-30$204.64−4.2%
2026-11-02$202.06−1.3%
2026-11-03$206.69+2.3%
2026-11-04$200.13−3.2%
2026-11-05$181.60−9.3%
2026-11-06$186.64+2.8%
2026-11-09$178.67−4.3%
2026-11-10$168.89−5.5%
2026-11-11$173.73+2.9%
2026-11-12$166.19−4.3%
2026-11-13$162.85−2.0%
Source: Robinhood market data.

These are different kinds of risk and they deserve different weights. The capital squeeze, the political backlash and the financing story all bear on the number of chips customers can buy and plug in. The leadership shock bears on who runs the company. Huang had about 60 direct reports and no named successor,11 so a discount for execution is reasonable, but his exit did not cancel an order. The export-control inquiry in Congress is a tail risk more than a forecast item: Nvidia’s outlook already assumes no data-centre chip revenue from China.2,30

2 · Circular financingWhat is proven, what is reported, and what is noise

“Circular financing” means a supplier funds a customer who then spends the money on the supplier’s product. The question for an investor is how much of it there is and who carries the loss if the customer cannot pay. On that, the best source is Nvidia itself.

ClaimStatusBasis
In Nvidia’s own filings and statements
Commitments and guarantees of $530.5bn at 26 July, from $184bn three months before✓ Proven10-Q, as tallied by Hunterbrook and SemiAnalysis1,4,29
A guarantee of up to $105bn on leases for a 4.25-gigawatt Ohio campus built for OpenAI, starting in fiscal 2029 and ending if OpenAI earns a satisfactory credit rating✓ Proven10-Q1
$36bn of promises to buy unused capacity from smaller AI clouds, and $20bn of leases signed to hand on to them✓ Proven10-Q1
$99bn of equity stakes in customers and partners, with $25bn more promised✓ Proven10-Q1
Payment terms stretched to between “90 days up to one year” for some buyers; receivables up $24.6bn in six months to $63.1bn, with five customers owing 70% of it✓ Proven10-Q1
Labs backed by Nvidia’s balance sheet will supply about a quarter of next year’s business✓ ProvenKress, 26 August results call5
Reported by the press, not confirmed by Nvidia
In August Kress pushed to cap customer support, Huang resisted, and the OpenAI guarantee was cut by about half? PlausibleWSJ, via other outlets. It fits the record: Bloomberg reported talks for up to $250bn in July, and the filing shows a $105bn cap6,7
Huang privately called incoming wafer and memory commitments an “avalanche”, while the biggest clouds shifted toward their own chips? PlausibleWSJ, via other outlets. It fits the filing: supply commitments rose from $119bn to $279bn in a quarter, and sales to hyperscalers grew more slowly than sales to everyone else7,27
Nvidia has asked TSMC to slow chip orders? ThinOne unnamed supply-chain source in Taiwan’s Commercial Times; other industry sources say orders are unchanged9
Circulating widely, not supported by evidence I could find
The chip shortage was engineered and the demand is fake✕ UnsupportedAn interpretation added by technology blogs. It conflicts with the independent data in section 38
$200bn of Nvidia chips sit idle in warehouses✕ UnverifiedOne commentator’s estimate from construction balances. The direction has support (Amazon racks waiting for power), the size does not, and waiting for power is not the same as unwanted20

Where the $530.5bn sits

Nvidia’s commitments and guarantees at 26 July 2026, US$ billions. Guarantees are shown at their maximum, not at expected loss.

Supply and capacity it must buy (mostly memory)
$279bn
Lease guarantee, Ohio campus built for OpenAI (maximum)
$105bn
Promises to buy unused capacity from AI clouds
$36bn
Cloud services for its own research
$29bn
Data-centre leases for its own engineering
$25bn
Equity investments promised, not yet paid
$25bn
Leases signed to hand on to AI clouds
$20bn
Capital expenditure
$8bn
Other lease guarantees for AI clouds (maximum)
$3.5bn
Source: Nvidia Form 10-Q for the quarter ended 26 July 2026; grouping by the author. Not included: a preliminary plan, reported after the filing, to put $100bn of equity into OpenAI in ten instalments tied to chip orders, and memorandums with outside lenders for $500bn-plus of third-party financing that Nvidia may partly support.4
My reading

“Circular” describes about a third of the total. More than half is the opposite problem: chips and memory Nvidia must take whether or not anyone wants them. Of the $279bn, $92bn falls due in the two quarters to January. That is $46bn a quarter, against a cost of revenue of $24bn last quarter. Some of it is prepaid factory capacity and not inventory, and 70% growth next year would absorb the rest. Slower growth would not.

That is the “avalanche” in numbers. The supply chain is contracted for the bull case, and financing customers was the tool for making the bull case happen. None of it was hidden; it is in the accounts. It was a bet that buyers’ ability to pay would catch up with supply already ordered.

3 · The demand testThree measures Nvidia does not control

If demand were manufactured, the evidence would show up downstream: clouds unable to sell capacity, AI products without paying users, and falling prices for compute. The opposite happened this autumn.

MeasureLatestEarlierReading
Cloud revenue growth, Q3 202613AWS +39%, Azure +48%, Google Cloud +103%; $110bn in the quarter+20%, +40%, +34% a year beforeAll three sped up
Signed cloud backlog13AWS $578bn, Google Cloud $561bn$496bn and $514bn in JuneCustomers are committing years ahead
AI-lab revenue run-rate14,15Anthropic $81bn (31 Oct), OpenAI about $50bnAbout $9bn (end-2025) and $25bn (February)End users are paying. Run-rates flatter: Anthropic’s booked third-quarter revenue annualises nearer $68bn
GPU rental price17H100 index $3.01 an hour (9 Nov); B200 $9.84, up 32% in a monthH100 at $2.00 in December 2025A three-year-old chip rents for 50% more than last December
CoreWeave, Q318Revenue +168%; backlog $132.5bnBacklog $104bn in JuneThe most levered buyer is still filling its capacity

The other side of the ledger is just as well documented:

My reading

Both sets of facts fit one picture. People and companies are paying for AI at a rate that is growing faster than almost any business on record, and the capacity to serve them is short, which is why rental prices are rising. What is short is powered, financed data-centre capacity. Chips are now arriving faster than sites can be switched on and faster than buyers’ cash flow can pay for them. Nvidia lent its balance sheet so that shipments kept pace with its supply contracts.

So the demand is real and Nvidia’s shipments have run ahead of it. That is how a semiconductor inventory cycle begins, and it does not need a fictional end market. Cisco’s revenue fell 15% in its 2002 fiscal year while internet traffic kept growing.

4 · Earnings qualityWhere the strain already shows

Nvidia data-centre revenue by type of buyer

US$ billions per quarter. The group Nvidia supports financially is the faster-growing one.

Q2 FY26 (Jul 2025)
$24.2bn
$16.9bn
Q1 FY27 (Apr 2026)
$43.0bn
$32.2bn
Q2 FY27 (Jul 2026)
$48.7bn
$40.3bn
Source: Nvidia CFO commentary, Q2 FY27, as reproduced by GPU Insights; prior periods recast by Nvidia.27

5 · The historical parallelLucent: right about the mechanism, wrong about the balance sheet

The comparison everyone reaches for is the telecom-equipment makers of 2000, which lent to start-up carriers so the carriers could buy their gear. It deserves a direct look.

Lucent, 2000Nvidia, July 2026
Customer financing commitments$8.1bnAbout $190bn in guarantees, capacity promises, leases and unpaid equity pledges
As a share of a year’s revenue24%63% of trailing revenue of $303bn
Cash generated in a year$0.3bn operating cash flowAbout $140bn free cash flow, at the first-half rate
Years of cash flow to cover every commitmentAbout 27About 1.4
Who was financedStart-up carriers funded by junk bondsLabs with $50bn–80bn revenue run-rates and far larger spending commitments, plus thinly capitalised AI clouds

Relative to sales, Nvidia has gone further than Lucent did. Relative to cash, it is in a different position. If every guarantee were called at its maximum, which would require OpenAI to default on 20-year leases that do not begin until fiscal 2029, the bill would be about nine months of free cash flow, and Nvidia would be left holding a powered 4.25-gigawatt campus filled with its own systems. This threatens the growth rate and the valuation multiple. It does not threaten solvency.24

6 · ValuationWhat $162.85 assumes

Seventeen times earnings looks cheap for a company that doubled its sales. It is also how cyclical companies look at the top of a cycle, when the “E” is about to fall. The right question is which year-ahead earnings the price is built on. Three cases for the year to January 2028:

Share value in three cases

US dollars per share. Bars span 15 to 20 times earnings; the dot marks the value at the multiple I think fits each case.

Price now $162.85
A · Guidance holdsRevenue about $690bn (+70%), EPS about $14.9$223–297 at 15–20× earnings
B · Financing reined inRevenue about $517bn (+28%), EPS about $10.5$158–210 at 15–20× earnings
C · DigestionRevenue about $304bn (−25%), EPS about $5.2$78–104 at 15–20× earnings
$50$100$150$200$250$300
Author’s estimates, illustrative. Assumes fourth-quarter FY27 revenue of $120bn, 24.6bn shares, and adjusted net margins of 53%, 50% and 42%. Central multiples are 18×, 16× and 18× (a trough multiple; the market paid more than 30× that year’s earnings at the 2022 low, so case C may be too harsh, while a severe bust would cut revenue by more than a quarter).
CaseCentral valuevs priceWeights implied by priceMy weights
A · Guidance holds$268+64%20%25%
B · Financing reined in$168+3%45%50%
C · Digestion$94−43%35%25%
Probability-weighted value$162$174
My reading

The market is pricing about a one-in-three chance of an outright revenue decline and a one-in-five chance that August’s guidance survives. I am a little less pessimistic about the bust, because the independent demand data are strong, and that gets me to $174, or 7% above the price. A gap that small is well inside the error of the inputs. The shares are priced about right for the risk, with outcomes ranging from −43% to +64%.

Two further observations. On 12 November individual investors bought the opening dip at the heaviest rate Vanda Research has recorded,10 which is not what a washed-out stock looks like. And the person now in charge was, by the Journal’s account, the internal critic of the financing. An interim chief with a finance chief’s instincts has every reason to reset expectations early. That makes case B the likeliest and makes 17 November the date it may start to show.

7 · The wider industryWho carries the risk if the financing slows

Fall from 52-week high

Per cent, at the 13 November close. The market has already drawn a line between the layers.

CoreWeave
−61.6%
Oracle
−49.9%
Nvidia
−33.1%
Broadcom
−30.0%
Alphabet
−27.6%
AMD
−18.9%
TSMC
−18.2%
Meta
−17.7%
Amazon
−7.2%
Microsoft
−6.2%
Source: Robinhood market data; 52-week intraday highs.
LayerWhat a financing pullback doesEvidence so far
Capacity built on debt
CoreWeave, other AI clouds, data-centre developers, Oracle
Most exposed. Their growth depends on Nvidia’s support and on lenders staying openCoreWeave owes almost $47bn and some of its bonds yielded over 16%, though its third quarter beat forecasts. Oracle’s free cash flow was −$23.7bn last fiscal year18,31
Nvidia’s suppliers
Memory makers, packaging, TSMC
Exposed if Nvidia trims orders. Memory is most at risk, since it is the bulk of the $279bnTSMC has a cushion: custom-chip customers reportedly would take any capacity Nvidia frees. Nvidia was 19% of TSMC’s 2025 revenue9,28
Rival silicon
Broadcom, AMD, Google’s TPU
Gaining share, but selling into the same capex budgetsAMD trades on 116 times trailing earnings and fell 8.7% on 13 November, more than four times Nvidia’s 2.0%
Hyperscale buyers
Microsoft, Alphabet, Amazon, Meta
A chip glut lowers their costs, and they own the powered sites that are actually scarce. Their risk is their own cash burnMicrosoft alone had positive free cash flow in the third quarter. Microsoft and Amazon sit 6–7% below their highs19
AI labs
OpenAI, Anthropic and others
Cheaper compute helps them; a shut funding window hurts them more than anyoneTheir revenue is the strongest evidence that demand is real, and their commitments are the largest unfunded promises in the chain

A bust for the builders need not be a bust for the technology. If Nvidia stops supporting prices with financing, compute gets cheaper, and that helps everyone who buys it. The fibre glut of 2001 ruined the companies that laid the cable and made the next decade of the internet cheap. In that kind of cycle value moves from the sellers of capacity to its users.

8 · Position sizingShould you hold less AI?

The data do not answer that for any one portfolio. They do support a way of working it out.

  1. Sort holdings by who pays them. Spending funded from operating cash flow is the most durable. Spending funded by debt and vendor support is the least. Spending funded by venture and IPO money sits between. The risk described in this note is concentrated in the second group.
  2. Count correlated positions as one. On 13 November AMD fell 8.7%, CoreWeave 6.2%, Broadcom 4.9%, the semiconductor ETF SMH 3.0% and Nvidia 2.0%, while Meta rose 2.9% and Microsoft 0.3%. Five chip names are one bet.
  3. Size from the bad case.
    Largest sensible weight = loss you could accept on the whole portfolio ÷ fall in the bad case
    With a bad case of −43% and a tolerance of 3% of the portfolio, that is about 7%. For a levered name where −60% is plausible, it is 5%.
  4. Do not treat a low P/E as the safety margin. In a cyclical business the multiple is lowest when earnings are highest.
My reading

The case for trimming is strongest where the business model is the financing loop: borrowed money buying chips on the strength of long contracts with a handful of young labs. It is weakest where spending comes out of cash flow and where the company owns what is scarce, meaning power, sites and paying users. On Nvidia itself, valuation alone does not justify adding before 17 November: the next catalyst is information, and the spread of outcomes is wide. A holder already above the weight that rule 3 gives has a reason to trim that does not depend on any forecast. A holder below it gives up little by waiting three days for the numbers.

9 · What would change this viewA checklist for 17 November and after

IndicatorNowReassuringWarning
Outlook for the year to Jan 2028About +70% (August)Reaffirmed with less customer supportCut below about +40%, or withdrawn
Days to collect a sale60Back toward 50Above 65
Supply commitments$279bnFlat or lowerHigher again
Guarantees plus capacity promises$144.5bnFlat; no new very large guaranteeThe $100bn OpenAI equity plan signed with chip orders attached
Hyperscale vs other growth, quarter on quarter13% vs 25%Hyperscale picks upThe gap widens
H100 rental indexAbout $3.00 an hour, risingHolds above $2.50Below $2.00, the December 2025 level
Hyperscalers’ 2027 capex guidance (January–February)Alphabet: “increase significantly”16MaintainedCut
Funding windowAnthropic IPO due to price 18 Nov; some CoreWeave bonds above 16%Offering clears; yields fallOffering pulled or trades down; yields rise

Consensus for the quarter is revenue of about $109bn and adjusted earnings of $2.47 a share, against guidance of $108bn give or take 2%.25 The headline numbers matter less than the eight lines above.

Method and sourcesHow this was put together

Prices, market values, consensus earnings and 52-week ranges come from Robinhood market data at the 13 November close. Company figures come from Nvidia’s results release and Form 10-Q for the quarter ended 26 July 2026. Every ratio, growth rate and scenario value was computed from those inputs and checked by script. Historical figures for Cisco (fiscal 2002) and Nvidia (fiscal 2023) come from the companies’ annual reports. Where a source is a secondary account of paywalled reporting, the entry says so.

  1. Nvidia, Form 10-Q for the quarter ended 26 July 2026 (SEC).
  2. Nvidia, Financial results for second quarter fiscal 2027, 26 Aug 2026.
  3. Nvidia, NVIDIA announces CEO transition, 12 Nov 2026.
  4. Hunterbrook, The Comprehensive Guide to the Bank of Nvidia, Nov 2026.
  5. AI News, A quarter of Nvidia’s business next year comes from labs it is financing, Aug 2026.
  6. Bloomberg, Nvidia’s $750 Billion in Deals Reignite Circular AI Fears, 27 Jul 2026.
  7. Wccftech, account of the Wall Street Journal investigation, 13 Nov 2026 (secondary).
  8. TechPowerUp, NVIDIA Allegedly Engineered Scarcity, 13 Nov 2026 (secondary).
  9. TrendForce, NVIDIA order cuts reportedly loom, 10 Nov 2026 (relaying Commercial Times).
  10. CNBC, Nvidia stock closes down over 4% after Huang announces resignation and live updates, 12 Nov 2026.
  11. TechCrunch, Nvidia’s Jensen Huang resigns; Decrypt, Huang resigns as Nvidia CEO, 12 Nov 2026.
  12. TechPowerUp, NVIDIA May Cut TSMC Chip Orders, Nov 2026.
  13. Midgard Finance, AWS vs Azure vs Google Cloud; Cleus, Amazon Q3 2026; Tech Insider, Cloud market share Q3 2026.
  14. TechCrunch, OpenAI’s quarterly revenue reportedly crosses $10B, 11 Nov 2026; Value Add VC, run-rate history.
  15. 24/7 Wall St., Anthropic’s $81 billion run rate, 11 Nov 2026 (on the S-1 amendment); Blockonomi, AI revenue statistics 2026.
  16. Yield Theory, Hyperscaler capex 2026: guidance by company, Nov 2026.
  17. gr1d, Rent or buy GPU servers (Silicon Data index); Ornn, GPU cloud price comparison, Nov 2026.
  18. CNBC, CoreWeave beats on revenue and lifts guidance, 10 Nov 2026.
  19. Walnut, AI capex tracker; Technology.org, AI capex now outruns Big Tech cash flow, 9 Nov 2026.
  20. Benzinga, Amazon’s Nvidia chips are sitting in warehouses without power, Oct 2026 (relaying Business Insider).
  21. AI World Today, How custom AI chips are challenging Nvidia; Safeguard, AI chip market share 2026.
  22. New Market Pitch, Can OpenAI actually afford all this compute? (relaying The Information and the Financial Times).
  23. Built In, What the 2026 midterm results mean for the AI industry; Gotrade, Nvidia stock news, 6 Nov.
  24. Tomasz Tunguz, Nvidia and the Lucent vendor-financing comparison (Lucent figures).
  25. Yahoo Finance, Nvidia reports earnings Tuesday; Invezz, earnings preview, Nov 2026.
  26. Yahoo Finance, $27 billion of free cash flow vanished; TIKR, Nvidia wants to earn twice.
  27. GPU Insights, NVIDIA data center revenue: the Q2 FY27 split.
  28. TIKR, Nvidia was 19% of TSMC’s 2025 sales, Nov 2026.
  29. Tiger Brokers, NVIDIA off-balance-sheet commitments soar to $530 billion (relaying SemiAnalysis).
  30. International Business Times, Nvidia scandal draws Congress into China chip probe, Nov 2026.
  31. Let’s Data Science, Big Tech AI capex set to outrun cash flow (relaying Epoch AI).