Superseded — archived edition of August 6, 2026. This memo is kept as a dated record of what was published on August 6, 2026 and is not maintained. The current edition, with a later valuation date and later public information, is at nvidia/memo.html. Figures below reflect only what was known at the time.
Equity research · internal memorandum · 2026-08-06

NVIDIA Corporation (NVDA)

Six-year segment-driver model with a vintage depreciation schedule. Base year TTM to 2026-04-26, all statement inputs from SEC XBRL.

NO POSITIONConviction 1/312-month horizonEdge: none

Spot
$218.99
as of 2026-08-06
Expected value
$224.91
+2.7% vs spot
Risk / reward
0.04 : 1
long framing
Breakeven p(bull)
23%
vs 25% assumed

Thesis

  1. What we thinkExtraordinary economics: 64% EBIT margin, 2.6% capex intensity, $119bn of free cash flow. The business is not the question.
  2. Why it is not pricedIt is priced. Our base case is 10% below spot, our revenue path is inside the analyst range every year, and we hold no variant view.
  3. What makes it workNothing currently observable resolves the terminal-margin question, which is where the entire disagreement sits.
Edge type: none. We have no edge here and the honest output is no position. Our revenue path sits inside the analyst range in every covered year, expected value is 3% above spot -- inside the noise -- and the scenario range spans $67 to $437. No defensible position size survives that spread.

The variant — where we differ from consensus

Fiscal yearOur revenueStreet avgStreet low–highΔPercentileAnalystsOur EBIT mgnStreet EBIT mgn
2027$381bn$393bn364–403-3.2%44th4064.0%48.1%
2028$552bn$563bn454–622-2.0%58th4061.9%48.1%
2029$699bn$685bn597–786+2.1%54th2759.2%48.1%
2030$805bn$774bn647–886+4.0%66th1356.4%48.1%
2031$863bn$1,005bn839–1,150-14.2%7th2253.6%48.1%

Why the margins differ: the depreciation schedule

We forecast EBITDA margin — a cash margin driven by mix, pricing and opex — and derive EBIT by subtracting depreciation built from the capex programme, vintage by vintage. Forecasting EBIT margin directly, which both the published consensus feed and most sell-side models do, hides the depreciation assumption inside a single number.

The consensus feed carries D&A at a flat 3.1% of revenue in every year and EBIT margin flat to one decimal place. NVIDIA Corporation spent 2.6% of revenue on capex in the last twelve months and depreciated 1.3% — a ratio of 2.0×. Those cannot both persist.
YearRevenueGrowthEBITDA mgnCapexCapex %revD&AD&A %revEBIT mgnFCFF
2027E$381bn76.4%64.5%$7bn1.8%$2bn0.5%64.0%$181bn
2028E$552bn44.9%62.5%$14bn2.5%$3bn0.6%61.9%$256bn
2029E$699bn26.7%60.0%$22bn3.2%$6bn0.8%59.2%$314bn
2030E$805bn15.2%57.5%$28bn3.5%$9bn1.1%56.4%$348bn
2031E$863bn7.1%55.0%$30bn3.5%$12bn1.4%53.6%$362bn
2032E$889bn3.1%53.5%$31bn3.5%$16bn1.8%51.7%$366bn

How year 1 is anchored

Forecast year 2027 is part history. $1.8bn of capex through Q1 is a filed fact, not an estimate. The remaining 3 quarters are held at the last observed quarterly rate of $1.8bn (2026-04-26), giving $7.0bn for the full year. Freezing the exit rate is deliberately conservative: quarterly capex has risen in each of the last seven quarters, so this assumes the ramp stops today. Anchoring on a trailing-twelve-month ratio instead — the conventional choice — would put the full year below the half-year already reported.

Price-to-value bridge

Contributions are exact Shapley values over all 27 = 128 driver coalitions, so they sum to the total without residual (residual $-0.0000). One-at-a-time sensitivity would not, because DCF driver interactions are large.

Step$/share
Street-calibrated value$231.94
revenue-48.69
ebitda margin+15.48
da pct revenue-1.34
Our base case$197.39
Market price$218.99

The gap from our base case to the market price is not decomposed into drivers, because it is not a forecast disagreement — it is the discount rate and terminal assumptions the market is applying. That is stated below rather than disguised as precision.

What the market is paying for

At $218.99, holding everything else at our base case, the market is paying for either

Revenue CAGR
29.3%
vs our 26.6% base
Terminal EBIT margin
58.2%
vs our 51.7% terminal
Discount rate (WACC)
9.07%
vs our 9.75% build

Scenarios and expected value

CaseValue/sharevs spotProbabilityWACCTerminal gTV % of EVAnchor
Bear$67.39-69.2%25%11.00%2.0%53%historical base rate
Base$197.39-9.9%50%9.75%3.0%71%consensus dispersion
Bull$437.47+99.8%25%9.25%3.5%80%judgment
Expected value$224.91+2.7%100%
Street-calibrated (their revenue, EBIT and D&A; our capex, our engine)$231.94+5.9%
Street published price target (median 300, range 218–500)$319.48+45.9%
Bear · 25%

A digestion year of the kind semiconductor cycles have produced repeatedly, plus custom-ASIC share gain on incremental sockets and gross-margin compression from HBM and advanced packaging cost. A cycle, not a thesis break.

Base · 50%

Our revenue path sits within the analyst range in every covered year, so this is genuinely the central expectation rather than a variant. The disagreement with the Street is on terminal margin, not near-term demand.

Bull · 25%

Rubin holds pricing, networking attach keeps climbing, and accelerated computing takes a larger share of a still-growing infrastructure budget. The width of this case against the bear case is the reason we carry no position.

Position sizing

Computed through a constraint cascade, not chosen. Basis: illustrative $1bn book.

TermValueNote
Payoff ratio b0.04reward ÷ risk
p(win)75%probability mass favouring the position
Kelly f-565.1%full Kelly fraction
Quarter-Kelly0.00%0.25 × Kelly
Liquidity cap4180.00%20% of ADV over 5 days
Risk-budget cap2.17%1.5% of NAV at risk to the adverse case
Concentration cap5.00%single-name hard limit
Position size0.00% of NAVbinding constraint: quarter-Kelly

Kill criteria

#If this happensObservableThresholdByAction
k1Equity stakes in customers exceed 40% of revenue
Currently 28.6%, up from ~3% a year ago. The cleanest tell on demand quality.
Equity securities (marketable + non-marketable) / TTM revenue> 40%2027-02-28review
k2Gross margin below 68% for two consecutive quarters
Would confirm the cost and mix compression the base case models.
GrossProfit / Revenues, quarterly< 68% x22027-05-31review
k3Inventory grows faster than guided forward revenue
Inventory is $25.8bn and rising; it leads demand inflections both ways.
InventoryNet vs next-quarter revenue guidancetwo consecutive quarters2027-05-31review

Catalyst path

DateEventWhat we learnTests
2026-08-26 (tentative)Q2 FY2027 resultsRubin ramp pricing, networking attach, China contributionk2
2026-11-18 (tentative)Q3 FY2027 resultsInventory and purchase commitments against the forward guidek3
2027-02-25 (tentative)Q4 FY2027 results and Form 10-KCustomer concentration and the full equity-stake disclosurek1

What would make us wrong

The case for owning it anyway: at 37.8x economic earnings for a business growing revenue 85% at 64% operating margins the multiple is not obviously wrong, and our terminal-margin fade to 53.5% may be too aggressive for a company with this much architectural lock-in. If terminal margin holds at 60% the base case clears spot comfortably.

The case for shorting it: $72.6bn of equity stakes in customers who use the proceeds to buy accelerators means some share of reported demand is self-funded, and that share is rising fast. We cannot size it from public disclosure, which is precisely why we will not take the other side either.

Pre-mortem

Twelve months out we have missed a double. What happened: Rubin shipped into a market still short of compute, networking attach kept climbing, and the terminal-margin debate never arrived because revenue grew fast enough to make it irrelevant on any horizon that mattered. We stood aside because our range was wide -- and a wide range is not the same thing as an unfavourable one.

Appendix A — quality of earnings

TTM to 2026-04-26$bnComment
Revenue253.5
Operating income (EBIT)162.364.0% margin
Non-operating income27.214.3% of pretax income
  of which equity-securities gains0.0non-cash marks on investment stakes
Reported net income159.6includes the marks above
NOPAT (EBIT after tax)136.7at 15.7% effective rate
Reported diluted EPS
$6.54
as filed
Economic EPS
$5.61
NOPAT basis
Reported P/E
33.5x
what screens show
Economic P/E
39.1x
what you actually pay

Appendix B — segment build

Every revenue line is the product of named quantities, not a growth rate. Each driver can be disagreed with individually and checked against disclosure.

Segment / driver2027E2028E2029E2030E2031E2032E
Data Center ($bn)355523668772826851
  units+52.0%+30.0%+18.0%+10.0%+5.0%+2.0%
  asp+16.0%+10.0%+6.0%+3.0%+1.0%+0.0%
  attach networking+4.0%+3.0%+2.0%+2.0%+1.0%+1.0%
Gaming ($bn)181921222223
  growth+12.0%+8.0%+6.0%+5.0%+4.0%+3.0%
Professional Visualization ($bn)445566
  growth+20.0%+15.0%+12.0%+10.0%+8.0%+6.0%
Automotive ($bn)345678
  growth+35.0%+30.0%+25.0%+20.0%+15.0%+12.0%
OEM & other ($bn)111111
  growth+5.0%+5.0%+5.0%+5.0%+5.0%+5.0%
Total revenue ($bn)381552699805863889

Appendix C — equity bridge and balance sheet

Item$bnNote
Enterprise value (base case)4,702
+ Cash & marketable securities52.5
+ Equity investments (stakes)72.628.6% of revenue — $2.98/share. Omitted from the published report's bridge.
− Debt8.5
− Operating leases4.3
Equity value4,815÷ 24,391m diluted shares
Value per share$197.39

Appendix D — sources and method

Known limitations