NVIDIA (NVDA) — Priced for One More Year
NVIDIA's third quarter beat our July top case and the company raised next year's growth to about 75%. The same week its founder resigned as chief executive, and the stock closed at $155.46, below our fair value for the first time since we began covering it. At that price the market pays for the year the company has guided and bought supply for, and for almost nothing after it. We value the stock at $170.43, from $162 in August.
In July we called NVIDIA "priced for the Clock not breaking": at $202.81 the stock needed about $1 trillion of revenue by fiscal 2031, and the risk sat outside the company, in its customers' capital budgets. In August, at $227.98, the price sat above even our top case. Since then the operating numbers have run ahead of every case we wrote. Third-quarter revenue was $115.8 billion, up 103% on a year earlier; fiscal 2027 is tracking about $424 billion, against the $400 billion our July top case assumed; and on the call the company raised its fiscal 2028 growth outlook to about 75%. The stock went the other way. It closed at a record $239.24 on 6 October, fell 9% in the 5 November chip rout, and fell again in the week Jensen Huang resigned as president, chief executive and director, on 11 November. It closed at $155.46 on 17 November, the day of the print, down 35% from the high. This note asks why, and what the price now assumes.
What this note answers
- What did NVIDIA report, and how much of the beat is the calendar, the memory bill and the launch timing?
- How large is the commitment stack, and what happens to it if fiscal 2028 falls short?
- What does the chief executive's departure change, and how much of the fall is NVIDIA's own?
- What is NVIDIA worth, and what moved the number from $162?
- What does $155.46 require?
- What would change our view?
Terms used throughout, defined once. Fiscal year: NVIDIA's year ends in late January, so fiscal 2028 is roughly calendar 2027. 14-week quarter: NVIDIA's fiscal 2027 has 53 weeks; the extra week falls in the fourth quarter, which is why we restate that quarter on 13 weeks. Supply and capacity commitments: purchase obligations to suppliers, most of which NVIDIA must pay whether or not it needs the parts (its 10-Q notes some can be canceled, rescheduled or adjusted); this quarter's increase was mostly memory. Credit support: a guarantee that NVIDIA pays a lender or landlord if a partner defaults. Private book: NVIDIA's stakes in private companies, carried at cost adjusted for impairments and observed funding rounds. Neocloud: a company that rents out GPU capacity and is not one of the large cloud providers. NOPAT: operating profit after the tax it would bear. FCFF (free cash flow to the firm): NOPAT less what the business reinvests in plant and working capital. ROIC: NOPAT over the capital the operations use. WACC: the blended return the company's lenders and shareholders require.
The cases on either side of the price, in their own terms
The case that $155.46 is fair or high. It does not deny the print. It says the print is the top of a capital-spending cycle, priced for what follows.
- Growth is slowing into a commitment stack that keeps growing. On 13 weeks the fourth-quarter guide is up about +4% on the third quarter, after 20%. Gross margin steps down about three points, to 71–72%. In the same quarter supply commitments rose by $111 billion to $390 billion, $203 billion of it due in fiscal 2028. Guarantees rose to $118.0 billion, net additions to the private stake book were $15.4 billion, receivables rose $18.1 billion and raw materials quadrupled from January. Each is a bet that fiscal 2028 arrives as guided.
- The demand is financed, in part, by NVIDIA. NVIDIA holds about $118 billion of stakes in customers and partners, guarantees $118.0 billion of their obligations and has $32 billion of contingent neocloud agreements. It recorded a $1.8 billion impairment on the private book this quarter. Its interim chief executive told analysts it would take "a very strong look" at whom it helps finance, because "a lot of things have changed over this last quarter" (one transcription).
- The founder's departure is a real loss. Jensen Huang ran the company for 33 years and set the yearly architecture cadence in public. The interim chief executive is also the chief financial officer. The company says an independent committee of the board is investigating, and the board has formed a committee named for national security and export control. In the same quarter, revenue from customers headquartered in China rose nearly fourfold, to $11.05 billion.
- The discount rate went up. The 10-year Treasury yield was 5.12% on 16 November, against 4.55% on 17 July; July's model used a 4.30% risk-free rate. Moving that input to today's yield at July's beta takes $16.56 a share off July's method.
- Margins are a peak, not a plateau. Frontier labs are leaning on custom chips (ASICs), and memory costs are rising. A 75% gross margin on systems others build cannot hold once the market stops being supply-constrained.
The case that $155.46 is too low. The company is supply-constrained in a market it says is still accelerating. It has bought the components for its guided year, and that year alone carries most of today's price.
- The year is bought. Fiscal 2028's $203 billion of commitments is about the cost of goods of the ~75% the company guides, at the 72–73% gross margin it reaffirmed. The four largest buyers' third-quarter capital spending rose 108% on a year earlier; every forward statement they gave pointed up, Microsoft's on its fiscal year to June 2027, and none cut 2026 (their third-quarter results and calls).
- The product cadence held through the change at the top. Vera Rubin generated "nearly $20 billion" of revenue in its first production quarter, which the company calls its fastest ramp. Rubin Ultra is "on track" for next year and Feynman for 2028.
- The price already assumes the plateau the other side fears. In our base case, $155.46 is what the company is worth if growth stops at the guided year. Every year of growth after fiscal 2028 is value the price does not include.
- Returns on capital are extraordinary. NVIDIA earned 206% on its operating capital over the last twelve months, against a WACC of 13.15%.
- Capital is coming back to shareholders. The quarterly dividend went from $0.01 to $0.25 a share in the second quarter of fiscal 2027. The company returned $28.1 billion in the quarter and plans to use the remaining $227.3 billion authorization by the end of fiscal 2028; at $155.46 that retires about 6.0% of the shares.
This note tests the first case's claims 1 and 2 in full, the rise in the rate in claim 4, and claim 3 as far as the price and the filings allow (§3); it sizes claim 5 through the terminal margin and the cases. Whether the rate stays high as the business matures it cannot test: that is a judgment, argued in §4, with its own tripwire. Of the second case it tests claims 1, 3 and 4 and sizes claim 5; claim 2 is the company's statement. Before the 10-Q it cannot test what the new $9.5 billion guarantee backs, how much of the commitments are memory prepayments, or what the board's investigation will find. Those are tripwires in §7.
The quarter: a beat, then the calendar, the memory bill and the launch
Revenue was $115.8 billion, up 20% on the second quarter and 103% on a year earlier. Data Center revenue was $107.6 billion (+110%). Within it, the company's new split shows Hyperscale at $60.1 billion (+71%) and AI Clouds, Industrial and Enterprise at $47.5 billion (+195%). Edge Computing was $8.2 billion (+42%). The new names map onto the old segments without a gap: Data Center is the same line, at $51.2 billion a year ago under either name, so Edge Computing ($5.8 billion a year ago) is everything else, the old Gaming, Professional Visualization, Automotive and OEM lines. Gross margin was 74.8% (GAAP), operating income $77.5 billion and diluted earnings per share $2.70 (GAAP) and $2.68 (non-GAAP), above consensus estimates for non-GAAP earnings that ranged from about $2.47 to $2.52 (Benzinga, Zacks, Chartmill). GAAP earnings include $1.8 billion of net gains on equity stakes: $2.8 billion on public stakes, less a $1.0 billion net loss on private ones that includes a $1.8 billion impairment. Free cash flow was $43.8 billion, 38% of revenue, against 22% in the second quarter.
Four things shape the guide before it reads as a slowdown.
- The calendar. The fourth quarter has 14 weeks. The company gave no 13-week figure; pro rata, $130.0 billion is $120.7 billion on 13 weeks: +4% on the third quarter, not the +12% the headline implies.
- The memory bill. The guided gross margin of 71–72% is about three points below the third quarter's. The company called the fourth quarter "the bottom" in the Q&A (one transcription) and said in its prepared remarks that it expects "Q4 to mark the bottom of gross margins before improving to the 72% to 73% range in fiscal year 28" (both transcriptions). That is the path it gave in August. In the Q&A it tied the trough to honoring purchase orders it had already accepted before moving to new pricing (one transcription). Separately, the supply commitments rose "primarily" on "the multi-year procurement of memory supply" (CFO commentary).
- The launch. Inventory rose $8.1 billion to $39.7 billion "to support our Rubin product ramp". Raw materials were $15.8 billion, against $3.8 billion in January. Receivables rose $18.1 billion to $81.0 billion, 64 days of revenue against 60. The company gives two reasons: shipments came late in the quarter at the Rubin launch, and it gave "extended payment terms on large, multi-quarter agreements with certain investment-grade customers". The first reverses next quarter. The second is not new: the second-quarter 10-Q already disclosed longer terms, of 90 days to a year, for investment-grade customers. It does not reverse with the launch, so it is a tripwire.
- The cash. The company says the sequential rise in operating cash flow, to $48.3 billion, was also affected by the timing of federal cash tax payments. Over nine months, working capital and other operating balances absorbed $46.8 billion. Free cash flow moved from 22% to 38% of revenue in one quarter, and part of that swing is timing, not profitability. That is why our model carries working capital as its own line and no longer uses a flat cash margin (§4).
Exhibit 1 — Quarterly revenue, Q3 FY26 to Q3 FY27, and the Q4 guide with and without its extra week
How to read it. Bars are quarterly revenue in billions of dollars; grey for the four earlier quarters, black for the quarter just reported. The last bar is the $130.0 billion fourth-quarter guide. Its solid gold part is the pro-rata 13-week figure ($120.7 billion); the dashed outline above is the 14th week. Under each bar is growth on the prior quarter, the last on a 13-week basis. Compare the last two growth figures: on a like-for-like calendar the guide implies growth of about a fifth of the third quarter's.
The commitments: the guided year is bought, and the stack is geared to it
The CFO commentary's commitment table is the most important exhibit in the release. Supply and capacity commitments were $390 billion at 25 October: $69 billion due in the rest of fiscal 2027, $203 billion in fiscal 2028, $92 billion in fiscal 2029, then $10 billion in fiscal 2030. Add cloud-service agreements ($28 billion), data-center leases not yet begun ($74 billion, $63 billion of it from fiscal 2032), equity-investment commitments ($20 billion) and capital spending ($6 billion), and the total is $518 billion. Contingent on top of that are $32 billion of neocloud agreements, which "take effect only if the neoclouds secure financing", and $118.0 billion of guarantees.
The gearing is simple arithmetic. At the reaffirmed 72–73% gross margin, the company's guided fiscal 2028 (about $741 billion of revenue at +75%) needs roughly $200 billion of cost of goods. The fiscal 2028 commitments are $203 billion. So the purchases are sized to the guide. If revenue arrives, the stack is ordinary working capital. If it does not, the stack becomes inventory, then write-downs.
Exhibit 2 — Supply commitments due each year against the cost of goods each case needs
How to read it. For each period, the black bar is supply and capacity commitments due, in billions of dollars, from the CFO commentary table at 25 October. The three bars beside it are the cost of goods each case's revenue requires at its gross margin: light grey for the bear case, grey for the base, gold for the bull. Where the black bar stands above a case's bar, that case buys more than it sells and carries the excess as inventory. The base case roughly uses what is bought; the bear case does not, and its excess peaks in fiscal 2028.
We model that directly. In each case, purchases contracted through fiscal 2029 are set against that case's cost of goods. Any front-loaded excess beyond a normal inventory buffer is written down at the case's loss rate. The base case's excess is about $3 billion and costs almost nothing. The bear case has fiscal 2028 revenue of $483 billion and fiscal 2029 of $362 billion. Its peak excess is $75 billion, written down at 25%: a $15.6 billion after-tax charge. That is the commitment write-down mechanism our July note named, sized.
The guarantees are smaller than their headline. The SB Energy credit support ($105.0 billion) begins building by building when each building is ready, the first in fiscal 2029, and then runs down over 20-year leases to OpenAI. On the call the company described the newly added neocloud wording as how the guarantees always worked: "Is it new? No. Did we change anything? No." (one transcription). The new $9.5 billion line, "power and systems credit support", is not explained in the release; its counterparty is a 10-Q item. The company also still holds, rather than has used, its option on about 3.8 more gigawatts at the site. We charge the stack as expected losses on its exposure path: annual default probabilities of 0.5%, 1.5% and 4% in the bull, base and bear cases, with recoveries on the land, power and shell. The cost is $2.5 billion in the base case and $9.3 billion in the bear. No guarantee payment has been made to date.
The stake book. NVIDIA now holds about $118 billion of stakes. Public stakes total $52.4 billion: $44.2 billion of marketable equity, $4.7 billion of long-term public holdings locked up through December 2027, and a $3.5 billion convertible note that the balance sheet counts among "marketable debt securities". Private stakes are $61.65 billion on the rollforward table, or $65.5 billion on the balance-sheet line; the $3.9 billion difference is not explained until the 10-Q. The book has tripled since January. The 13F for 30 September shows a SpaceX stake (130.5 million shares, $19.7 billion) and an Intel stake of 161.1 million shares, down 25% in the quarter.
The book could be read as operating capital, as vendor financing that buys demand, or as a separate asset. We treat it as a non-operating asset carried below its mark, the same way in every calculation. It is excluded from the capital in ROIC and from the WACC weights. In the equity bridge, public stakes are counted at their mark less 15%, for lock-ups on $25.7 billion of them, the tax on embedded gains and concentration in two names. The private book is counted at half its carrying value. Its marks come from funding rounds NVIDIA often joins, and part of what NVIDIA paid bought demand that is already in our revenue forecast; counting both in full would count it twice. The $20 billion of future equity commitments cost half their face for the same reason. The $1.8 billion impairment is the first observed instance of the write-down mechanism, at about 3% of the book. Valued at carrying value, as in August, the private book adds +$1.27 to fair value; valued at zero, −$1.27. Either way the stake book moves fair value by about a dollar a share. It matters more as a sign of how the demand is being financed than as a number.
Hugging Face ($11.9 billion plus up to $1.0 billion of retention awards, expected to close in the first half of 2027 subject to regulatory approval) is about a fifth of cash and debt securities and 0.3% of market value. We treat it as zero net present value: the price paid equals the value received. The 8-K commits NVIDIA to "keep Hugging Face's platform open" and to "support other silicon vendors".
The chief executive and China: what we can observe, and what we price
What the filings say. On 11 November the board accepted Jensen Huang's resignation as president, chief executive and director, effective immediately (8-K, filed 12 November). Colette Kress, chief financial officer since 2013, is interim chief executive and remains CFO. Harvey C. Jones, a director since 1993, is the board's first chair. The search covers internal and external candidates. The filing gives no reason. On the call, the chair said the resignation followed "disclosure of personal behavior inconsistent with the company's code of conduct, standards, and cultural norms" and that the board "does not intend to reinstate Jensen as CEO" (both transcriptions). He said the board had retained Heidrick & Struggles for the search and had formed a National Security and Export Control Oversight Committee, and that it would not take questions on the circumstances or "the ongoing investigation". The earnings release names "the ongoing investigation by an independent committee" of the board; it does not state the scope. We do not characterize the personal matters, and nothing about them enters the model.
What management added to its risk factors. The release's cautionary statement changed this quarter. Against August's, it adds two risk factors: the company's "ability to successfully transition NVIDIA's leadership", and "the scope, timing and outcome of investigations, litigation and government and regulatory actions and proceedings". Among the matters it calls forward-looking it adds the leadership transition and chief-executive search, "litigation, government and regulatory inquiries", the investigation, and the quarter's own 10-Q. In each of the previous nine quarters, NVIDIA filed its 10-Q or 10-K the same day as the release. This time the CFO commentary says only "in the coming weeks". The deadline is 4 December, and a tripwire watches it (§7).
What matters for value. Four things:
- Key-person and execution risk. A chief financial officer is running the company through the Rubin ramp. The observable test is the roadmap the company committed to on this call: Rubin Ultra next year, Feynman in 2028.
- Whether the investigation reaches export-control, sanctions or disclosure exposure. Revenue from customers headquartered in China, including Hong Kong, was $11.05 billion in the quarter (9.5% of revenue), against $2.97 billion a year earlier. Separately, the company shipped $2.4 billion of licensed Hopper data-center products to China in the quarter; it has stopped shipping them, and its guide excludes them. The release does not say how much of the headquarters figure those shipments make up. If all of it, the rest is about $8.65 billion. Headquarters is not destination, and the supplemental table does not say where the products went; the 10-Q geography note is where to read it. The company also said "new export controls were released by the U.S. Commerce Department this week". We have not found the rule text and do not describe it.
- Separation terms. No 8-K yet. They may come in an 8-K or the proxy.
- Ownership and trading. We count and do not characterize. No Form 4 has been filed by any NVIDIA insider since 11 November. Mr. Huang's only Form 4 since our August note (filed 18 September) reports shares withheld for tax and a gift, with 70.1 million shares held directly afterward.
How much of the fall is NVIDIA's own. From the 4 November close to the anchor, NVIDIA fell 22.3%. The sector ETF SMH fell 8.8%, while AMD rose 3% and Broadcom (AVGO) 5%. Split the window and the picture is clear. The 5 November rout was a sector event: NVIDIA fell 9.3%, SMH 11.2%. From the 6 November close, before the first reports, to the anchor, NVIDIA fell 16.7% while SMH was −0.8%, AMD +12% and AVGO +16%. Against SMH scaled by NVIDIA's own sensitivity to it, the company-specific move is about 16%, or about $30 a share. On the 8-K day itself the stock opened at $151.01, traded down to $148.80 and closed at $166.19, off 4.3% on 748 million shares.
Exhibit 3 — NVIDIA against its sector and two rivals, 4 to 17 November (4 November close = 100)
How to read it. Each line is a closing price indexed to 100 at the 4 November close: black for NVIDIA, gold for the SMH semiconductor ETF, grey dashed for AMD, light grey for Broadcom (AVGO). The shaded band is 6 to 12 November, from the last close before the first reports to the close of the day the 8-K was filed. The labels at the right are the change to 17 November. Read the gap between the black and gold lines after the band opens: the sector held its level while NVIDIA fell, and both rivals rose.
That residual is the closest thing to a market price for the key-person event, but it cannot separate the chief executive's departure from the China question attached to it, or from money moving into AMD and Broadcom, both of which rose while NVIDIA fell; and because SMH holds NVIDIA itself, it slightly understates the company-specific move. Our own charge is the ten points of weight the bull case loses to the base (§4), worth $7.29 a share: about a quarter of the residual, with no base rate to anchor it. We put the charge there and only there. The base case takes the company's +75% without a cut: the ten points the bull loses are our charge for the change at the top, and cutting the base's growth as well would count the same risk twice. The discount rate stays out, because key-person risk is specific to the company and can be diversified away; putting it in both the rate and the weights would count it twice.
What it is worth
Method. We value the firm's free cash flow in three cases, weighted 30/45/25 (bear/base/bull), discounted from 17 November at mid-year, the terminal value on the same timing. Every piece of the July and August model was open to change; we kept what we could defend and replaced what we could not. We use free cash flow to the firm rather than to equity because debt is under 1% of capital, so the two converge and the stakes and debt stay out of the cash flows; a dividend model would value a payout policy rather than the cash the business can pay, and a sum of the parts would split one platform with one cost base. In order of effect:
- Revenue by stage, not by hand. July set five revenue levels per case. We build revenue from what drives each stage. The table below states each stage's length and evidence. Fiscal 2028 starts from the company's guide and the supply it has bought. Fiscal 2029 is the last year the commitments cover. A four-year fade then brings growth down to the economy's rate, one GPU life after the 2026–27 installation wave. The base takes fiscal 2028 at the company's +75%. The supply it has bought for fiscal 2028 is about the cost of goods of that guide at the reaffirmed gross margin. That cannot tell +70% from +75%: the two need about $198 billion and $204 billion of cost of goods. We take the guide because no evidence supports cutting it, not because the purchases prove it; +70% is worth $2.37 a share less. The base also needs fiscal 2028 to re-accelerate: first-quarter revenue near $142 billion, the first of four quarters each growing about 18% from the 13-week fourth quarter, after +4%. The 13-week fourth quarter annualized, about $483 billion, is the bear case's fiscal 2028. The bull is +80%, the bear +15%.
- Cash flow line by line. July and August used a flat free-cash-flow margin. We model operating margin, tax, capital spending and working capital (16.5% of revenue in the base, the current ratio). The terminal value reinvests what its growth requires at the return on new capital it earns. That return does not fall at a step: over the fade years (fiscal 2030–33) each year closes a larger share of the gap between today's capital needs and what the terminal requires, so in the base the return on new capital falls from about 79% in fiscal 2030 to 30% in fiscal 2033, the year the cash flows join the terminal value. Each case fades to its own terminal return: bull 40%, base 30%, bear 15%. Competition that takes margin also makes each new dollar of revenue cost more capital to win: customer financing, supply prepayments, stakes.
- Margins. Operating margin is 66% in fiscal 2028 in the base case (72–73% gross margin less operating costs of about 6.5% of revenue). It fades to a terminal 50%: between NVIDIA's pre-AI 26–37% (fiscal 2020–22) and today's 67%, for a franchise that keeps its software and systems lead but faces custom chips and AMD at scale. The bull terminal is 55%, the bear 40%.
- Terminal growth. 4.0%: the Federal Reserve's longer-run projections of 16 September 2026, 2.0% real growth plus 2.0% inflation. The CBO cross-check is about 3.8%.
- Stock-based pay is expensed. GAAP operating income carries it, and since the first quarter of fiscal 2027 so do the company's non-GAAP figures. We do not dilute the share count for it as well. The share count is the third-quarter diluted average, 24,186 million.
| Stage | Years | What drives it | What sets the length | Growth: bear / base / bull |
|---|---|---|---|---|
| Committed build-out | 2 | Hyperscaler, frontier-lab, neocloud and sovereign AI build-outs already ordered and supplied: the company guides FY28 revenue ~75% higher (Q3 call) and has bought the supply for it | FY28-FY29: the purchase commitments that make the revenue possible run $203bn in FY28 and $92bn in FY29, then fall to $10bn in FY30 (CFO commentary table, 25 Oct 2026). Two years is as far as bought supply reaches; the bear breaks inside it | FY28: +15% / +75% / +80% · FY29: -25% / +15% / +25% |
| Maturation fade | 4 | Growth from the installed base's replacement cycle and the annual architecture cadence (Rubin Ultra 2027, Feynman 2028, company roadmap on the call) rather than first-time build-out; share erodes slowly to ASICs and AMD | FY30-FY33: one GPU economic life (five to six years) after the FY26-FY28 install wave is replaced from about FY31; the fade runs four years so that by FY33 every case grows with the economy. Bull and base fade straight from FY29 growth to 4%; the bear is a plateau, not a fade | FY30: +0% / +12.2% / +19.8% · FY31: +2% / +9.5% / +14.5% · FY32: +3% / +6.8% / +9.2% · FY33: +4% / +4.0% / +4.0% |
| Terminal | perpetuity | A mature compute franchise growing with nominal GDP, earning a terminal operating margin of 50% (base; bull 55%, bear 40%) and 30% on new capital (bull 40%, bear 15%); the return on new capital falls to each case's terminal level over the fade years (each year FY30-33 closes a larger share of the gap, so FY33 reinvests g / RONIC), not at a step in FY34 | Perpetuity from FY34 at 4.0%: Fed SEP 16 Sep 2026 longer-run real GDP 2.0% plus PCE inflation 2.0%; CBO cross-check ~3.8% | 4.0% in every case |
How to read it. One row per stage of growth, in order. Each row gives the stage's length in years, what drives growth in it, and the dated observable that sets its length. Growth by case for each year is in the model's Stages sheet. The test of a stage table is whether each length follows from its driver; the price's own answer is in §5.
Cost of equity. 13.2% in fiscal 2028–29, converging in a straight line to 11.6% by fiscal 2033: the 10-year Treasury (5.12% on 16 November) plus a beta times a 4.5% equity risk premium. The measured betas disagree: own 60-month raw 2.23, shrunk toward peers 2.08, Blume-adjusted 1.83, 36-month 1.88, and a peer median of 1.61 across AMD, Broadcom, Micron, Qualcomm and TSMC. The 60-month window opens in November 2021, when gaming was NVIDIA's largest segment and the stock was about to lose more than 60% in the 2022 bust; that measures a different business. The 36-month window measures the AI-platform business alone; shrunk toward the peer median for its larger error, it gives about 1.8, which the Blume figure confirms. Hence 13.2%.
For the terminal we use 1.44, the average of TSMC (1.41) and Broadcom (1.46), the two peers that look most like what NVIDIA becomes in maturity: a dominant franchise with pricing power and a moat in software or process. That is the business the bull and base describe. The peer median, 1.61, is Qualcomm's own beta; it sits higher because Qualcomm carries handset cycles, and Micron, a commodity memory maker, and AMD, a challenger taking share, sit higher still. The bear describes that other kind of business: customers lose access to capital, revenue falls a quarter in fiscal 2029, and the terminal keeps 40% of revenue as operating profit and earns 15% on new capital. So the bear converges to the peer median instead, 12.4%. No case goes lower: no peer below the TSMC/Broadcom pair fits a mature franchise, and the 1.2 of the stable-growth convention is ruled out because mature chip franchises stay cyclical.
This choice carries the verdict, and no filing can test it: at the peer median in every case the value is $159.88, just above the price; at a flat 13.2% in every year it is $149.72. A tripwire on NVIDIA's own measured beta governs any move (§7). The WACC is 13.15%; debt is under 1% of capital, so it sits within a tenth of a point of the cost of equity. July used 11.26% in every year; the change is its own line in the bridge.
| Bear (30%) | Base (45%) | Bull (25%) | |
|---|---|---|---|
| Revenue growth, fiscal 2028 / 2029 | +15% / −25% | +75% / +15% | +80% / +25% |
| Operating margin, fiscal 2028 → terminal | 60.0% → 40% | 66.0% → 50% | 66.5% → 55% |
| Return on new capital in the terminal | 15% | 30% | 40% |
| Working capital, share of revenue | 18.0% | 16.5% | 15.0% |
| Revenue, fiscal 2028 / 2031 / 2033 | $483B / $369B / $396B | $741B / $1048B / $1164B | $770B / $1319B / $1499B |
| Terminal value, share of operating value | 46% | 63% | 67% |
| Guarantees / neocloud / commitment overhang (charges) | $9.3B / $7.2B / $15.6B | $2.5B / $2.4B / $0.1B | $0.7B / $0.0B / $0.0B |
| Value a share | $56.43 | $193.23 | $266.16 |
How to read it. One column per case, with its weight in the heading. The rows above the line are the inputs we set: growth in fiscal 2028 and 2029, operating margin in fiscal 2028 and in the terminal, the return on new capital in the terminal, and working capital. The rows below are what follows: revenue in fiscal 2028, 2031 and 2033, the terminal value's share of operating value, and the charges the bridge carries for guarantees, neocloud agreements and the commitment overhang, in billions of dollars. The last row is value a share. Compare each column's inputs with the evidence in §1–§3.
The weighted value is $170.43. The base contributes $86.96, the bull $66.54 and the bear $16.93. The bear is lower than July's ($104) because it now carries a demand break, a commitment write-down and guarantee losses. The bull is higher than August's ($210) because fiscal 2027 already exceeds what August's bull assumed.
Weights, argued on their own. Bear 30%, base 45%, bull 25%, against August's 30/35/35 on differently built cases. The bear stays at 30%. What it describes, customers losing access to capital, is a risk the company itself names in its outlook ("our customers' access to capital", one transcription); it is visible in neocloud credit and in the company's own pullback on financing; and a CFO now runs the company. In the Q&A the company said it allocates supply first to customers whose land, power and capital are in place (one transcription). The bull falls to 25%: it requires beating a guide that interim management has just raised, and keeping a 55% margin for good against custom chips, without the person who set the strategy. Those ten points go to the base, the guide-anchored path that the bought supply makes most likely. This change is worth $7.29. At 40/40/20 the value is $153.10; at 25/45/30, $180.91.
Exhibit 4 — Fair value from August ($162.14) to this note, one change at a time
How to read it. Fair value in dollars a share, on an axis that starts at $100. The first black bar is August's value as its engine computes it; each bar after it adds one change, gold if it raises the value and grey if it lowers it, with the size above it; the last black bar is this note's value. The dashed gold line is the $155.46 close. The first three changes are the quarter's data run through August's method; the rest are method, rate and weights, as in the table below. A different order would change the split, not the total.
| Change | Kind | $ a share | Fair value after |
|---|---|---|---|
| Data: FY27 revenue at the Q4 guide (9M $293.7bn + $130bn ±3%), FY28 at August's growth on that base | data | +0.86 | 163.00 |
| Data: net cash and stakes at 25 Oct ($17.7bn financial + $117.9bn of stakes at book), 24.19bn diluted shares | data | +1.46 | 164.46 |
| Data: risk-free rate 4.30% (July's input; the 10-year was 4.55% on 17 Jul) -> 5.12% (10-year, 16 Nov) at July's beta and ERP (11.3% -> 12.1%) | data | -16.56 | 147.90 |
| Method: operating margin, tax, capex and working capital in place of a flat FCF margin; return on new capital fading over FY30-33 to the terminal's (same revenue, rate, weights) | method | -3.75 | 144.15 |
| Method: revenue built by stage (FY28 off the company's guide and bought supply, FY29 the last committed year, 4-year fade to 4%) | method | +33.58 | 177.73 |
| Method: stakes below their mark (public -15%, private 50%, commitments at half), guarantee, neocloud and commitment-overhang expected costs | method | -2.61 | 175.12 |
| Cost of equity: 12.1% flat -> 13.2% in FY28-29 (beta 1.80), converging to 11.6% (beta 1.44; bear 12.4%, beta 1.61) by FY33; WACC on the same rates | rate | +2.17 | 177.29 |
| Method: China data-center reopening as its own leg (10% probability; out of every case, as out of the guide) | method | +0.42 | 177.72 |
| Weights: bear/base/bull 30/35/35 -> 30/45/25 | weights | -7.29 | 170.43 |
| Fair value 162.14 → 170.43 | +8.29 | 170.43 |
How to read it. Each row applies one more change to the running value, in the order shown, starting from August's $162.14 as its engine computes it (published as $162). Kind says whether the change is data, method, rate or weights. The third column is its effect in dollars a share, the fourth the value after it. The first three rows are the quarter's data run through August's method; a different order would change the split, not the total.
What moved the number from $162. August's engine reproduces to $162.14. The quarter's data alone, run through August's method, give $147.90. A higher fiscal 2027 and more stakes per share add a little; moving the risk-free rate from July's 4.30% input to today's 5.12% 10-year yield takes off $16.56. The method changes then add +$27.22 net, before the cost of equity and the weights. Building revenue by stage adds the most, +$33.58, because August held fiscal 2029–31 at levels that fiscal 2027 has nearly reached. Modeling working capital and reinvestment explicitly costs $3.75. Carrying the stakes below their mark and charging the contingent items costs $2.61. The cost of equity adds +$2.17 net: a higher explicit-year rate, more than offset by the convergence to 11.6% (12.4% in the bear). The China leg adds +$0.42 and the weights take off $7.29.
| Change (one at a time) | Fair value | vs this note |
|---|---|---|
| Discount rate | ||
| Explicit-year cost of equity 12.4% (peer-median beta 1.61) | $174.42 | +$4.00 |
| Explicit-year cost of equity 14.5% (Vasicek read, beta 2.08) | $164.58 | −$5.85 |
| Terminal beta 1.2 bull and base, bear keeps 1.61 (stable read, 10.5%) | $189.56 | +$19.14 |
| Terminal beta 1.6 (peer median, 12.4%) | $159.88 | −$10.55 |
| Flat 13.2% in every year (no convergence) | $149.72 | −$20.71 |
| Terminal growth 3.5% | $164.27 | −$6.15 |
| Terminal value at end-year timing | $164.79 | −$5.63 |
| Growth and margins | ||
| Base fiscal 2028 at the bull's +80% | $172.80 | +$2.37 |
| Base fiscal 2028 at +70% (the figure before this quarter's raise) | $168.05 | −$2.37 |
| Base fiscal 2028 at August's +50% haircut | $158.53 | −$11.89 |
| Fade 2 years (base) | $164.93 | −$5.50 |
| Fade 6 years (base) | $174.70 | +$4.27 |
| Base terminal operating margin 45% | $163.73 | −$6.69 |
| Base terminal operating margin 55% | $177.12 | +$6.69 |
| Terminal return on new capital 20% in every case | $158.26 | −$12.17 |
| Terminal return on new capital 40% in every case | $175.23 | +$4.81 |
| Balance-sheet calls and the model's own rules | ||
| Private stake book at carrying value (August's convention) | $171.70 | +$1.27 |
| Private stake book at zero (pure vendor financing) | $169.15 | −$1.27 |
| Working capital at 20% of revenue in every case | $169.19 | −$1.23 |
| Rule: guarantee default probabilities doubled in every case | $170.26 | −$0.17 |
| Rule: commitment write-down rate doubled in every case | $170.23 | −$0.20 |
| Rule: bear overhang written off in full (loss rate 100%) | $169.85 | −$0.58 |
| Rule: neocloud agreements lost in full in the bear case | $170.22 | −$0.21 |
| Export-control penalty $5bn (investigation reaches the company) | $170.22 | −$0.21 |
| Weights and China | ||
| Weights 40/40/20 | $153.10 | −$17.33 |
| Weights 25/45/30 (bear/base/bull) | $180.91 | +$10.49 |
| August weights 30/35/35 | $177.72 | +$7.29 |
| China leg probability 0% | $170.01 | −$0.41 |
| China leg probability 30% | $171.25 | +$0.82 |
| China-headquartered revenue lost from fiscal 2028 (−7.5% of revenue, every case) | $158.29 | −$12.13 |
How to read it. Each row changes one input and leaves everything else as in the note. The value is the weighted fair value in dollars a share, to compare with $170.43 and the $155.46 close. The rows are grouped: discount rate, then growth and margins, then the balance-sheet calls and the model's own charging rules, then weights and China. The rule rows barely move the value: the contingent stack matters as a sign of how demand is financed more than as a charge. The rate rows move the value most, which is why the rate is argued above rather than taken from one read.
Relative cross-check, at zero weight. On trailing GAAP earnings including the third quarter ($9.31 a share), NVIDIA trades at 16.7 times. The brokerage feed's trailing multiples for the peers, taken before any updates for prints after 17 November, are TSMC 24.0, Broadcom 42.6 and Qualcomm 13.3; Micron is at 9.7 and AMD at 110. On our base case's fiscal 2028 operating profit after tax, NVIDIA is at about 9.3 times. We give the comparison no weight. Each peer's multiple prices its own place in its own cycle (Micron's at a memory peak, AMD's on earnings depressed by its own build-out), so a median says little about a company whose next year is already contracted. What it does show is that the market prices NVIDIA's guided year at a multiple usually paid for earnings that are about to fall.
China, as its own leg. Every case excludes China data-center compute, as the company's guide does. A reopening is a separate leg: $30 billion of revenue from fiscal 2029, at each case's margin less 15 points for a licensed-sales levy, with a 10% probability, worth +$0.42 a share. Losing NVIDIA's other China-headquartered revenue, about $8.65 billion a quarter or roughly 7.5% of revenue, from fiscal 2028 in every case would take fair value to $158.29. That is the size of the export-control question if the investigation or the new rules reach the customers rather than one product line.
Return on capital. On operating capital, NVIDIA earned 206% over the last twelve months against its 13.15% WACC, and has been above 125% for twelve quarters: a strength regime. The business designs systems that others build, so each dollar of operating capital carries about two dollars of after-tax operating profit a year. What is new is where the capital goes. Counted as capital, the private stake book lowers the return to 152%. Over the window, new capital earned 230%. Our terminal value assumes 30% on new capital: a sevenfold fall, reached over fiscal 2030–33 rather than in one step, and the competition the terminal value bets on.
Exhibit 5 — Return on operating capital against WACC, trailing twelve months, and the return on new capital
How to read it. Each bar is trailing-twelve-month operating profit after tax over average operating capital (equity plus debt, less cash, debt securities and every equity stake), for the twelve quarters to October 2026. The black line is the 13.15% WACC, the test the return must clear. The dashed grey line just above it is the 13.2% cost of equity, shown for reference. At the right edge, the black dot is the return earned on the capital added over the window (230%); the gold triangle is the 30% the model assumes in its terminal value. Read the bars against the black line, then the gap between dot and triangle. The chart shows a business earning many times its cost of capital, which the valuation assumes will fade.
What $155.46 requires
In July, on that note's engine, $202.81 required about $1 trillion of fiscal 2031 revenue. On the same engine, $155.46 requires $755 billion, a 28% compound rate from fiscal 2026. Fiscal 2027 alone is tracking about $424 billion.
On this note's model, each of these is enough on its own to make the base case worth the price, with everything else held:
- Growth stops after the guided year. Fiscal 2028 at about +70%, a little under the company's guide, then 4% a year. (Our base's +75% followed by 4% gives $159.92.) The plateau the first case fears is, in our base case, what the price already pays for.
- Or fiscal 2028 falls well short of the guide. +39% growth with later stages unchanged, against the company's ~75%.
- Or a higher discount rate in every year. 14.1%, with no step-down as the business matures.
- Or lower terminal margins. A terminal operating margin of 37%, close to NVIDIA's pre-AI range.
- Or a heavier bear case. A 39% weight on it, with the base and bull keeping their 45:25 ratio.
- Or a lighter bull case. A 13% weight on it, with the bear and base keeping their 30:45 ratio.
Exhibit 6 — Revenue by case, fiscal 2026 to 2033, and the path the price pays for
How to read it. Each line is annual revenue in trillions of dollars. Black is the base case, gold the bull, grey the bear. The dashed gold line is the path that makes the base case worth exactly $155.46: fiscal 2028 at about +70%, then 4% a year. The vertical line marks fiscal 2028, the year the company has guided. Read the gap between the dashed line and the black line after fiscal 2028: that is the growth our value counts and the price does not.
So the market is pricing either a growth path that ends after fiscal 2028 or a discount rate that never falls. The evidence leans against the first: the largest buyers' forward statements all point up, the company has bought fiscal 2029's supply, and the architecture cadence is dated. On the call an analyst asked about exactly this, a 2028 slowdown after the change in government; the interim chief executive answered with current demand and customers' requests for more supply, not with a 2028 figure (one transcription). It cannot rule out the second. A stock that just lost its founder, carries an open board investigation of unstated scope and sits atop a financed demand cycle can stay priced at a high rate for a long time. That is why our value is only 9.6% above the price, and why the tripwires that move it are dated.
The buyback, as a per-share lever. The remaining $227.3 billion authorization, used at $155.46, retires about 6.0% of the shares and is worth $171.39 a share to holders who stay, if our value is right. The same money at the August reaction close ($227.98) would have been worth $167.95. A buyback below value adds per-share value; one above value subtracts it.
The receipt: July's and August's tests, scored
Threshold stated here: a provision or impairment of $5bn or more in a quarter (about 2% of trailing operating income, $239bn) is material. Q3: a $1.8bn impairment inside a $1.0bn net loss on the private book (3% of the $61.65bn book; prior-quarter impairment history is a 10-Q item), and $680M of ordinary inventory provisions (Ex-99.3). Below the line. The silence rule governs the rest: supply commitments rose $111bn to $390bn ($203bn due in FY28) and net additions to the private book were $15.4bn in the quarter, so more is at risk, not less. Any provision against purchase obligations is a 10-Q note: pending. The impairment is the first observed instance of the mechanism the trigger names, at a size the model already carries (base overhang $0.1bn; bear $15.6bn after tax).
Graded on the four buyers' Q3 reports on 11-09: aggregate capex +108% y/y on the cash-flow basis, every 2027 statement pointing up. An NVIDIA print cannot move this gauge; the company's own read is its guide: Q4 $130bn including a 14th week (about $120.7bn on 13 weeks, +4% q/q) and FY28 growth raised to ~75%. Next read at the hyperscalers' Q4 reports (due 2027-02-05).
No new third-party accelerator-share estimate was read for this note. Per the silence rule the share is carried at July's last measured level, ~80% of AI-accelerator revenue (range 75-88% across IDC, Bloomberg Intelligence and TrendForce, mid-2026, as compiled for the July note). The base case's terminal margin (50%) is where share loss lives in the model.
Maximum gross guarantee exposure $118.0bn (from $108.5bn) against the $150bn line. The +$9.5bn is a new row, 'Power and systems credit support guarantees', which the CFO commentary does not explain beyond 'land, power, shell, and hardware financing arrangements'; the counterparty and term are 10-Q items (pending). The 3.8 GW option is described in the present tense ('We also have the option'): not exercised. No payments made to date.
The Q4 guide is 71.0-72.0%; its low end sits on the trigger's 71% line. The trigger reads on the realized Q4 gross margin (due 2027-02-28), not the guide. The company called Q4 'the bottom' and held FY28 at 72-73% (call).
The supplemental (Ex-99.3) carries the table the brief expected only in the 10-Q: three direct customers at 14%, 11% and 11% of Q3 revenue, 36% in all, on the same direct-customer basis as Q2 (one customer at 16%); nine months 13/12/12/11%. Below 50%. Two direct customers hold 18% and 17% of receivables.
No rating action on OpenAI or the guaranteed affiliate found, and no 8-K. Silence leaves it armed.
'To date, we have not made any payments related to our guarantees' (CFO commentary). No 8-K on a counterparty default. The guarantee-note detail on counterparties is a 10-Q item.
Non-marketable securities $65.5bn on the balance-sheet caption and $61.65bn on the rollforward (Ex-99.3), against $55bn: the book kept growing (+$15.4bn net additions in the quarter, $46.4bn in nine months). The trigger's own reading: leverage, not exit. The $13.3bn of nine-month reclassifications out of the book are stakes that listed, not sales. The 10-Q Note 6 is the trigger's named source: confirm there.
Not tested in this note.
Not tested in this note.
NVIDIA's 13F-HR for 30 Sep 2026 (acc. 0001045810-26-000084, filed 16 Nov) lists 161,082,488 Intel shares against 214,776,632 at the prior filing: -53,694,144 shares (-25.0%). The trigger's own reading: the only unlocked big position sold into strength reads as trade, not alignment. The 13F carries eight positions worth $48.3bn; the largest is now SpaceX (130,472,925 shares, $19.7bn).
Reads on filed text; the Q3 10-Q is not filed.
Not tested in this note.
July's and August's live claims, re-tested.
- "The price embeds about $1 trillion of fiscal 2031 revenue." At $155.46, July's engine gives $755 billion.
- "The swing factor sits outside the company." It held for revenue, which still follows the buyers' capital budgets. It failed for the stock this quarter: the largest move since August was NVIDIA's own (§3).
- "About 80% share." Carried at the last measured level; no new estimate.
- The Cisco comparison. Concentration fell: the top three direct customers took 36% of third-quarter revenue, against two customers at 36% in fiscal 2026.
- August's $108.5 billion of guarantees and $279 billion of supply commitments are now $118.0 billion and $390 billion.
- "Fair value $162 against $228, above even the bull." The price is now below the fair value.
Tripwires to the next update
Each tripwire states its reading rule and what it does to the value, and each is checked daily or watched on its filing or news feed.
- Q1 FY28 revenue guide (late Feb 2027) against the base case's FY28 path. The base case's FY28 ($741bn, the company's +75%) needs Q1 FY28 near $142bn: four quarters growing 18% each from the 13-week Q4 run rate ($120.7bn). Guide midpoint at or above $150bn: base FY28 +75% -> the bull's +80% ($172.8). $135-150bn: no change. The band is a no-change zone; its two ends follow one rule, the FY28 pace the guide implies on the same quarterly path ($135bn ~ +66%, $150bn ~ +84%), capped at the bull's +80% because the base cannot pass the bull without rebuilding both. Below $135bn: base FY28 +65% ($165.67); below $120bn (under the 13-week Q4 run rate): also bear 30% -> 40%, base 45% -> 40%, bull 25% -> 20% ($153.1 on today's case values). Reading: NVIDIA Q4 FY27 release / CFO commentary: Q1 FY28 revenue outlook midpoint.
- Permanent CEO named, and from where (internal / external). Named by 30 Jun 2027 with the roadmap dates reaffirmed (Rubin Ultra 2027, Feynman 2028): bull 25% -> 30%, base 45% -> 40% ($174.07). Not named by 30 Jun 2027, or the interim arrangement (CFO as CEO) extended without a date: bear 30% -> 35%, base 45% -> 40% ($163.59). Either rule acts alone. Reading: NVIDIA 8-K Item 5.02 / press release; board statement.
- Outcome of the board's independent investigation, as the company states it. Company states no export-control, sanctions or disclosure finding: no change (nothing is priced for it beyond the weights). A penalty or settlement without business restrictions: subtract it per share ($5bn = $0.21). Any license restriction, denial order or consent agreement that limits sales to China-headquartered customers: remove the non-Hopper China-headquartered revenue from FY28 in every case ($158.29). Reading: NVIDIA 8-K / 10-Q / 10-K legal-proceedings note; BIS or DOJ settlement announcements naming NVIDIA.
- Q3 FY27 10-Q filed by the 4 December 2026 deadline (40 days after the 25 October quarter end). NVIDIA filed its 10-Q or 10-K on the release day in each of the previous nine quarters (Aug 2024 - Aug 2026); this quarter the release says only 'in the coming weeks' and lists the 10-Q among its forward-looking statements. Filed by 4 December with the release's figures unchanged and no new material weakness: no change, and the pending 10-Q items (commitment, guarantee, purchase-provision, China-geography and private-book notes; Item 1A risk factors against the Q2 10-Q) are read before the note is final. A Form 12b-25, no 10-Q by the deadline, a change to the quarter's reported figures, or a material weakness: re-underwrite the cases before relying on any figure from the quarter. Neither branch moves an input by itself. Reading: EDGAR: NVIDIA 10-Q for the quarter ended 25 Oct 2026, or a Form 12b-25 (NT 10-Q, due one business day after the deadline).
- Vera Rubin Ultra ships in calendar 2027 (company: 'next year'; Feynman 2028). Ships in the second half of 2027: no change. Ships in the first half of 2027, ahead of the company's own timing: the base case's fade lengthens from four years to five ($172.97). Slips into 2028: the base case's fade shortens to three years ($167.75) and bull 25% -> 20%, base 45% -> 50% ($166.78 on today's case values); a slip and an unnamed CEO together do not stack beyond the two rules. Reading: NVIDIA release / CFO commentary: Rubin Ultra production shipments.
- A provision against purchase commitments or a stake impairment of $5bn or more in a quarter. $5bn or more in any quarter: bear 30% -> 35%, base 45% -> 40% ($163.59 on today's case values): a charge that size is the mechanism running, and the bear case already carries the overhang loss. Below $5bn: no change. Mirror: no provision through the FY27 10-K and FY28 supply commitments drawn down as scheduled (the 10-K table shows the FY28 column used, not rolled forward): bear 30% -> 25%, base 45% -> 50% ($177.27). Reading: NVIDIA 10-Q / 10-K: purchase-obligation provisions (inventory and commitment note), non-marketable impairments (rollforward).
- Receivables days (DSO) on the extended-terms build. DSO 60-70: no change. Above 70 at the Q4 FY27 or Q1 FY28 print: working capital 16.5% -> 20% of revenue in every case ($169.19). Back below 60 (the launch timing reversing and the extended terms small): 16.5% -> 15% ($170.82). Reading: NVIDIA CFO commentary: accounts receivable and DSO.
- China data-center compute: licensed shipments resume, or China-headquartered revenue is restricted. Licensed data-center compute shipments to China resume (company statement): China leg probability 10% -> 30% ($171.25). A rule or order that stops sales to China-headquartered customers outside data-center compute: non-Hopper China-headquartered revenue removed from FY28 ($158.29). Reading: NVIDIA CFO commentary / 10-Q geography note; BIS rules.
- NVIDIA's own 36-month beta, re-measured at each update through FY29 (capm.py). The verdict rests on the bull and base terminal beta (1.44, the TSMC/Broadcom mean; the bear already uses the 1.61 peer median). 36-month beta 1.5-2.0: no change. Above 2.0: terminal beta to the peer median 1.61 in every case ($159.88). Below 1.5, which would contradict the cyclicality argument: bull and base terminal beta to the 1.2 stable-growth read, the bear keeping the peer median its own business sets ($189.56). Re-read at each update through FY29; the due date is the first read. Reading: NVIDIA monthly returns against the S&P 500 over the latest 36 months, at the Q4 FY27 update.
- Settled close above $230, over the value of a 50% bull weight. The mirror of the next rule. At today's case values a 50% bull weight (bear and base in their 30:45 ratio) is worth $202.34. A settled close above $230 (about 48% above the anchor) sits over even that: re-underwrite the cases before the next print. The price itself is not an input. Reading: NVDA settled close (Schwab closePrice / Robinhood close.price).
- Settled close below $120, under the value of a 50% bear weight. At today's case values a 50% bear weight (base and bull in their 45:25 ratio) is worth $137.86. A settled close below $120 (about 23% below the anchor) sits under even that: re-underwrite the cases before the next print. The price itself is not an input. Reading: NVDA settled close (Schwab closePrice / Robinhood close.price).
What would change our view
Toward a lower value: a first-quarter guide below $135 billion (base fiscal 2028 to +65%, $165.67); a provision or impairment of $5 billion or more in a quarter (bear weight to 35%, $163.59); Rubin Ultra slipping into 2028; no permanent chief executive by 30 June 2027; or a finding that limits sales to China-headquartered customers ($158.29). Toward a higher value: a first-quarter guide at or above $150 billion (base fiscal 2028 to the bull's +80%, $172.80); a permanent chief executive named by 30 June 2027 with the roadmap dates reaffirmed (bull weight to 30%, $174.07); or licensed data-center sales to China resuming (China leg to 30%, $171.25). Rubin Ultra shipping in the first half of 2027, receivables days back below 60, or the fiscal 2028 commitments used as scheduled with no provision each move the value up by the same rules. On the rate: if NVIDIA's own 36-month beta, re-measured at each update through fiscal 2029, is above 2.0, the bull and base terminal beta moves to the peer median the bear already uses ($159.88); below 1.5, which would contradict the cyclicality argument, the bull and base terminal beta moves to 1.2 and the bear keeps the peer median its own business sets ($189.56). A settled close below $120 or above $230 triggers a re-underwriting of the cases. Each rule has a mirror and moves a single input, with one exception: a first-quarter guide below $120 billion, under the 13-week fourth-quarter run rate, also moves the weights (bear to 40%, $153.10), because at that level the bear's mechanism is visible in the guide itself; no upside tier matches it. No two rules stack beyond their own terms.
The call
We value NVIDIA at $170.43 a share against $155.46: 9.6% above the price, the first time in our coverage that the price has sat below our number. The quarter beat the case we called most optimistic in July. The stock fell on rates, on a sector sell-off, and on something specific to NVIDIA that took about $30 a share off it beyond the sector in the weeks its founder left and its board opened an investigation. In our base case, $155.46 pays for one more year of supernormal growth, the year the company has guided and bought supply for, and nothing after it. We think the evidence supports more: dated product cadence, buyers guiding up, supply bought through fiscal 2029. But the margin is thin. It rests on a discount rate that falls as the business matures, and on demand that does not break while NVIDIA is helping to finance it. July said the risk sat outside the company. This quarter part of it moved inside.
Sources & Method
Filings. NVIDIA Form 8-K, 17 November 2026 (acc. 0001045810-26-000088): Exhibit 99.1 earnings release, Exhibit 99.2 CFO commentary (platform revenue, commitments and guarantee tables, outlook), Exhibit 99.3 supplemental financial information (income statement, balance sheet, cash flows, securities, non-marketable rollforward, inventories, geography and customer concentration). Form 8-K, 12 November 2026 (acc. 0001045810-26-000082; Item 5.02, chief executive transition) and Form 8-K, 3 September 2026 (acc. 0001045810-26-000078; Hugging Face). Form 13F-HR for 30 September 2026 (acc. 0001045810-26-000084). Forms 4 and 144 since 27 August 2026 from EDGAR. Quarterly history from SEC XBRL company facts (10-Qs and 10-K through Q2 FY27). The Q3 FY27 10-Q was not filed at the time of writing; the company says "in the coming weeks".
The call. 17 November 2026, 2:00 p.m. PT. The company's transcript was not posted at the time of writing. Quotes are checked against two independent transcriptions (singjupost.com in full; Equibles and Investing.com, verbatim through the prepared remarks). Quotes that only one transcription carries are marked as such. All are archived with their sources.
Market data. Settled closes and daily and monthly bars from the brokerage feed (Robinhood close price; split-adjusted); 10-Year Treasury (DGS10) from FRED; trailing P/E ratios for the peer cross-check from the same feed. Hyperscaler capital spending and outlooks from the four companies' third-quarter results and calls.
Prior notes. NVIDIA — Priced for the Clock Not Breaking (July 2026) and its engine; What $213 Has to Believe and NVIDIA's $108 Billion Footnote (August 2026) and the August rerun; The Fifth Seat; The Visible Hand; Neocloud Credit 2026.
Method. Free cash flow to the firm by case, weighted 30/45/25, mid-year from 17 November 2026 (terminal included), explicit through fiscal 2033 and a Gordon terminal from fiscal 2034 at 4.0% growth with reinvestment of g over the return on new capital. Cost of equity 13.2% in fiscal 2028–29, converging to 11.6% by fiscal 2033 (12.4% in the bear); the return on new capital fading over fiscal 2030–33 to each case's terminal return; WACC built at market weights. Equity bridge: financial cash (cash and debt securities, less the $3.5 billion convertible) less debt; public stakes at their mark less 15%; the private book at 50% of its rollforward carrying value; equity commitments at half their face; expected costs of guarantees, neocloud agreements and the commitment overhang by case; China leg at its probability. Return on capital from operating income after tax over equity plus debt less cash, debt securities and every stake. The model and every input are in the downloadable workbook. The price decomposition regresses NVIDIA's daily returns on SMH's from July to 3 November to get NVIDIA's sensitivity to the sector (0.66).
Limitations. The 10-Q is not filed. The commitment, guarantee, purchase-provision, China-geography and private-book notes may change the inputs above, and we will read them before treating this note as final. The fiscal 2028 growth, the fade, the terminal margin and return on new capital, the stake discounts, the guarantee loss rates and the cost of equity are judgments; each is shown with its effect. The decomposition of the price fall cannot separate the key-person event from the China question or from rotation into competitors.
Disclosures
Information only. TON618 Capital. This report is for information purposes only. Nothing here is an offer to sell or a solicitation of an offer to buy any security, fund interest, or digital asset, and nothing here is personalized investment advice or a recommendation regarding any instrument.
Publisher's exclusion. All research is published solely as general, impersonal information of regular circulation. It is not tailored to the objectives or circumstances of any individual and is not issued in connection with compensation from any client. The Fund has no clients and distributes all research free of charge. On that basis it publishes in reliance on the publisher's exclusion from the definition of "investment adviser" under the Investment Advisers Act of 1940 (§202(a)(11)(D); cf. Lowe v. SEC, 472 U.S. 181 (1985)).
Registration & conflicts. TON618 Capital is not registered as an investment adviser or broker-dealer in any capacity. The Fund is a Bitcoin fund and may hold or transact in the securities or digital assets it discusses; material conflicts are disclosed where they exist. Ownership: the Fund holds no position, long or short, in NVDA or in any other security named in this report as of the report date, and has no economic interest in their price. Compensation: the Fund received no compensation from any party in connection with this report and charges nothing for it.
Use of AI. Artificial intelligence is used in the creation of this research. All methodology and data integrity are reviewed and approved before publication by TON618 Capital's Chief Investment Officer, Keyth Beck; errors may nonetheless occur, and readers should verify independently.
CFA. This report was prepared to align with CFA Institute analytical standards (methodology only). CFA® and Chartered Financial Analyst® are registered trademarks owned by CFA Institute. That reference describes the analytical framework applied; it does not imply the report was prepared, reviewed, or authored by a CFA charterholder, and the report is not issued, reviewed, endorsed, certified, or approved by — nor affiliated with — CFA Institute.
Risk & feedback. Past performance is not indicative of future results. Digital assets and equities are volatile and may result in total loss of capital. Corrections and feedback are welcome — please direct them to CIO Keyth Beck at keyth@ton618capital.com. Version 1.0.