
Buy, $205.86 target: $40bn of content per gigawatt, priced at 11x FY28E earnings
We initiate at Buy with a $205.86 target: 13.4% upside and 13.9% total return, almost all of it the 12.5% cost of equity. NVIDIA sells the whole AI factory at $40bn of content a gigawatt; at 11.4x FY28E consensus EPS it is 33% below the peer median.
- Content per gigawatt is the engine: $40bn against $25bn a generation ago. With capacity up 30% in FY28, revenue is $695bn and adjusted EPS $16.76, 5.4% above the $15.91 consensus.
- The margin fade is in our numbers, not in the peer multiple. Gross margin runs from 73.7% to 68% by FY32E: $183.86 in our DCF, against $288.86 on 17.0x.
- The financing bear case costs $4.0 a share, not the thesis. $153bn of stakes and commitments sit against $191bn of FY27E free cash flow; the bear is $122.60.
What the market is missing. Consensus stops at FY29; the price implies a 52.9% steady-state gross margin against our 68%, a gap worth $20.8 a share.
Catalysts. Q3 FY27 results on 17 November (guide $108bn, 74% gross margin); Q4 FY27 on late February; Broadcom's AI guide in December.
Valuation and target. Ten-year DCF at 12.4% WACC and 3.0% terminal growth: bear $122.60, base $211.47, bull $316.73 (30/50/20), $205.86 weighted; the 17.0x check gives $288.86.
Risks. Permitting (−$43.3 a share, 35% likely), custom ASICs (−$27.0, 35%), memory costs (−$8.1, 40%). One point on the discount rate makes the call Hold.
| US$bn, Jan year end | FY25A | FY26A | FY27E | FY28E | FY29E |
|---|---|---|---|---|---|
| Revenue | 130.5 | 215.9 | 409.0 | 695.2 | 824.5 |
| Growth | 114.2% | 65.5% | 89.4% | 70.0% | 18.6% |
| EBITDA | 83.3 | 133.2 | 271.2 | 464.4 | 531.0 |
| Margin | 63.8% | 61.7% | 66.3% | 66.8% | 64.4% |
| Net income | 72.9 | 120.1 | 242.8 | 382.5 | 437.8 |
| EPS ($) | 2.94 | 4.90 | 10.03 | 16.27 | 19.38 |
| Growth | 146.4% | 66.7% | 104.7% | 62.2% | 19.1% |
| Free cash flow | 60.9 | 96.7 | 191.5 | 345.1 | 434.1 |
| At $181.60, 5 Nov 26 | FY25A | FY26A | FY27E | FY28E | FY29E |
|---|---|---|---|---|---|
| P/E | 61.8x | 37.1x | 18.1x | 11.2x | 9.4x |
| EV/EBITDA | 51.4x | 32.2x | 15.8x | 9.2x | 8.1x |
| P/B | – | 28.3x | 15.4x | 9.0x | 6.1x |
| FCF yield | 1.4% | 2.2% | 4.3% | 7.8% | 9.9% |
| Dividend yield | 0.0% | 0.0% | 0.4% | 0.6% | 0.6% |
| ROE | – | 76.3% | 84.9% | 80.5% | 65.4% |
| Net debt / EBITDA | net cash | net cash | net cash | net cash | net cash |
| P/E at our $205.86 target | 70.0x | 42.0x | 20.5x | 12.7x | 10.6x |
Source: NVIDIA 10-K filings; Penmark Research model, price $181.60 on 5 November 2026. Shaded columns are estimates; P/B and ROE from FY26A, the first period with a full balance sheet. Fuller ratios in Exhibits 8-9.
Base $211.47 is 16% above the price and bear $122.60 is 32% below: $24 of upside to the weighted target against $59 of downside to the bear
Source: NVIDIA closing prices; Penmark Research model. Scenario values are DCF outputs at 26 Jul 2026 rolled 12 months at the 12.5% cost of equity.
FY28 capacity +40%, FY29 +25% on $1.3tn of hyperscaler capex; Vera CPU and LPX add 7% a year to content; gross margin holds near 71% to FY32E.
Guidance delivered: FY28 capacity +30%, FY29 +10%, flat from FY32E; gross margin fades to 68% as custom silicon scales; working capital 13% of revenue.
Permitting curbs after the midterms and ASIC share gains cut FY29 capacity 10%; memory costs take gross margin to 62%; working capital 20%; 50% stake haircut.
| Income statement | FY26A | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| Revenue | 215.9 | 409.0 | 695.2 | 824.5 | 908.9 |
| Gross profit | 153.5 | 301.2 | 504.1 | 585.4 | 636.2 |
| EBITDA | 133.2 | 271.2 | 464.4 | 531.0 | 568.1 |
| EBIT | 130.4 | 266.3 | 455.4 | 519.4 | 554.4 |
| Net financial items | 11.1 | 24.5 | 2.7 | 8.0 | 14.8 |
| Tax | 21.4 | 48.1 | 75.6 | 89.7 | 96.8 |
| Net income | 120.1 | 242.8 | 382.5 | 437.8 | 472.5 |
| Net margin | 55.6% | 59.4% | 55.0% | 53.1% | 52.0% |
| Cash flow | FY26A | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| EBITDA | 133.2 | 271.2 | 464.4 | 531.0 | 568.1 |
| Stock compensation | 6.4 | 9.0 | 13.9 | 16.5 | 20.0 |
| Tax, working capital, other | -36.9 | -78.5 | -115.9 | -92.8 | -93.8 |
| Operating cash flow | 102.7 | 201.7 | 362.5 | 454.7 | 494.3 |
| Capex | -6.0 | -10.2 | -17.4 | -20.6 | -22.7 |
| Free cash flow | 96.7 | 191.5 | 345.1 | 434.1 | 471.6 |
| Dividends and buybacks | 41.1 | 118.2 | 207.1 | 260.4 | 282.9 |
| Acquisitions and investments | -46.2 | -30.7 | 0.0 | 0.0 | 0.0 |
| Balance sheet | FY26A | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| Cash and securities | 49.7 | 105.2 | 243.2 | 416.8 | 605.5 |
| Equity investments | 35.1 | 93.9 | 93.9 | 93.9 | 93.9 |
| Net working capital | 31.9 | 61.3 | 104.3 | 115.4 | 127.2 |
| PP&E | 10.4 | 17.3 | 25.6 | 34.7 | 43.8 |
| Goodwill & intangibles | 24.1 | 24.1 | 24.1 | 24.1 | 24.1 |
| Total net assets | 165.8 | 319.3 | 508.6 | 702.4 | 911.9 |
| Debt | 8.5 | 33.4 | 33.4 | 33.4 | 33.4 |
| Shareholders' equity | 157.3 | 285.9 | 475.2 | 669.1 | 878.6 |
| Per share and KPIs | FY26A | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| Diluted shares (m) | 24,514 | 24,213 | 23,508 | 22,585 | 21,610 |
| EPS, reported ($) | 4.90 | 10.03 | 16.27 | 19.38 | 21.86 |
| EPS, adjusted ($) | 4.90 | 9.44 | 16.76 | 19.99 | 22.63 |
| Dividend per share ($) | 0.04 | 0.76 | 1.02 | 1.02 | 1.02 |
| Free cash flow per share ($) | 3.94 | 7.91 | 14.68 | 19.22 | 21.82 |
| Book value per share ($) | 6.42 | 11.81 | 20.21 | 29.62 | 40.66 |
| Gross margin | 71.1% | 73.7% | 72.5% | 71.0% | 70.0% |
| Data Center rev. growth | 68.2% | 96.1% | 74.2% | 18.9% | 10.3% |
Source: NVIDIA filings; Penmark Research model, base case, 6 November 2026. FY26A is the last full reported year (FY27 has two reported quarters); FY27E-FY30E are estimates. Cash flow: stock compensation is a non-cash add-back and is charged in operating expenses. Scenario values are DCF outputs at 26 July 2026 rolled 12 months at the 12.5% cost of equity. Statements in full in Exhibit 50.
Returns stay above 50% through FY30E, and the stock trades on 11.4x FY28E consensus earnings against a 17.0x peer median
| Growth, margins | FY26A | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| Growth, y/y | |||||
| Sales | 65.5% | 89.4% | 70.0% | 18.6% | 10.2% |
| EBITDA | 59.9% | 103.6% | 71.2% | 14.3% | 7.0% |
| EBIT | 60.1% | 104.2% | 71.0% | 14.1% | 6.7% |
| EPS, reported | 66.7% | 104.7% | 62.2% | 19.1% | 12.8% |
| DPS | n.m. | n.m. | +34.2% | 0.0% | 0.0% |
| Margins | |||||
| Gross margin | 71.1% | 73.7% | 72.5% | 71.0% | 70.0% |
| EBITDA margin | 61.7% | 66.3% | 66.8% | 64.4% | 62.5% |
| EBIT margin | 60.4% | 65.1% | 65.5% | 63.0% | 61.0% |
| FCF margin | 44.8% | 46.8% | 49.6% | 52.6% | 51.9% |
| EBITDA / net interest | 514x | 258x | 321x | 368x | 393x |
| Capex / sales | 2.8% | 2.5% | 2.5% | 2.5% | 2.5% |
| Ratios, valuation | FY26A | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| Returns and leverage | |||||
| ROE | 76.3% | 84.9% | 80.5% | 65.4% | 53.8% |
| ROA | 72.4% | 76.0% | 75.2% | 62.3% | 51.8% |
| ROCE | 66.9% | 70.9% | 76.1% | 62.9% | 51.7% |
| Net debt / equity | -26% | -25% | -44% | -57% | -65% |
| Net debt / EBITDA | -0.31x | -0.26x | -0.45x | -0.72x | -1.01x |
| Valuation at the current price | |||||
| P/E | 37.1x | 18.1x | 11.2x | 9.4x | 8.3x |
| EV/EBITDA | 32.2x | 15.8x | 9.2x | 8.1x | 7.5x |
| P/B | 28.3x | 15.4x | 9.0x | 6.1x | 4.5x |
| FCF yield | 2.2% | 4.3% | 7.8% | 9.9% | 10.7% |
| Dividend yield | 0.0% | 0.4% | 0.6% | 0.6% | 0.6% |
| P/E at our target | 42.0x | 20.5x | 12.7x | 10.6x | 9.4x |
Source: NVIDIA filings; Penmark Research model, price $181.60 on 5 November 2026. P/B is price over book value per average diluted share. Ratios on period-end capital: ROE on shareholders' equity, ROA on total net assets, ROCE on EBIT at a 15% tax rate over net assets. Net debt excludes equity investments. Capex/sales shown positive. DPS growth is not meaningful in FY26A-FY27E: the quarterly dividend went from $0.01 to $0.25 in Q2 FY27.
| US$, fiscal years | Q3 FY27E | Q4 FY27E | FY27E | FY28E |
|---|---|---|---|---|
| Revenue, bn | ||||
| Penmark estimate | 108.0 | 123.1 | 409.0 | 695.2 |
| Consensus | 109.0 | 124.5 | 411.7 | 692.2 |
| Prior-year period | 57.0 | 68.1 | 215.9 | 409.0 |
| Difference vs consensus | −0.9% | −1.1% | −0.7% | +0.4% |
| Growth, Penmark estimate | +89.5% | +80.7% | +89.4% | +70.0% |
| Gross margin | ||||
| Penmark estimate | 74.0% | 71.5% | 73.7% | 72.5% |
| Guidance or consensus | 73.5-74.5% | 71-72% | — | 72-73% |
| Adjusted EPS, $ | ||||
| Penmark estimate | — | — | 9.44 | 16.76 |
| Consensus | — | — | 9.31 | 15.91 |
| Difference vs consensus | — | — | +1.4% | +5.4% |
| Growth, Penmark estimate | — | — | — | +77.5% |
| NVIDIA | AMD | AVGO | TSM | QCOM | MRVL | Peer median | NVDA vs median | |
|---|---|---|---|---|---|---|---|---|
| Price, $ | 181.60 | 404.40 | 291.53 | 353.10 | 156.75 | 133.55 | — | — |
| Consensus EPS, $ | 15.91 | 15.93 | 19.43 | 20.77 | 10.19 | 6.79 | — | — |
| EPS basis | FY28 | CY27 | FY27 | CY27 | FY27 | FY28 | — | — |
| Forward P/E | 11.4x | 25.4x | 15.0x | 17.0x | 15.4x | 19.7x | 17.0x | −33% |
| EPS growth | 70.9% | 109.1% | 66.6% | — | — | 61.3% | 66.6% | +4.3pp |
| PEG | 0.16 | 0.23 | 0.23 | — | — | 0.32 | 0.23 | −0.07 |
| Beta, 5Y monthly | 2.21 | 2.52 | 1.48 | — | — | — | 2.00 | +0.21 |
| Gross margin | 73.7% | — | 69.1% | — | — | — | 69.1% | +4.6pp |
| Operating margin | 65.1% | — | 53.9% | — | — | — | 53.9% | +11.2pp |
Sources: Penmark Research model and guidance from the Q2 FY27 call (26 August 2026); consensus revenue and EPS from Yahoo Finance analyst estimates, September-October 2026 (45 estimates for Q3 FY27 revenue); quarterly revenue derived from the H2 FY27E model column and the Q3 guide. Peer prices are 5 November 2026 closes; consensus EPS from TIKR, StockTitan and company-reported figures with differing fiscal year-ends, as labelled (CY = calendar year). Peer median excludes NVDA. Margin benchmarks are the quarter to 2 August 2026 (AVGO).
Four dated events and four priced risks: the stock is cheap against its own history and against its peers
- 17 Nov 2026Q3 FY27 results. We model $108.0bn against a $106-110bn guide and $109.0bn consensus; gross margin 74.0%. Vera Rubin at about 20% of Data Center revenue is the first read on the ramp.
- Dec 2026Broadcom's FY27 AI semiconductor guide, the cleanest read on custom ASIC share. Our FY29 capacity call, +10% against +21% implied by consensus, is worth −$15.6 a share against the market view.
- Jan-Feb 2027Hyperscaler 2027 capex: the top five have pointed to $1.3tn against nearly $800bn in 2026. Capex that cuts FY28 capacity growth from 30% to 20% is the bear case.
- late Feb 2027Q4 FY27 and the FY28 outlook. Gross margin troughs at 71-72%; we expect a FY28 revenue guide of about +70% ($695bn) and capital return above 50% of free cash flow.
Source: Company results calendar; Q2 FY27 earnings call, 26 August 2026; Penmark Research model.
- Permitting and AI regulation. Data Center approvals slow after the 3 November midterms, curbing capacity additions from FY28. Likelihood medium-high; −$43.3 a share (35%).
- Custom ASIC competition. Lab-designed accelerators win inference share and take pricing from FY28. Likelihood medium-high; −$27.0 a share (35%).
- Memory costs. HBM and DRAM pricing keeps gross margin below 70% into FY28. Likelihood medium; −$8.1 a share (40%).
- Financing and receivables, the one that would end the Buy. A 50% haircut on private stakes and working capital at 20% of revenue. Likelihood low-medium; −$4.0 a share (25%).
Source: Penmark Research model, bear-case value contributions at 26 July 2026.
Source: NVIDIA closing prices (yfinance) and reported GAAP diluted EPS; trailing EPS interpolated between fiscal-year ends. Weekly, Nov 2022 to Nov 2026. Bands: average plus or minus 1 and 2 standard deviations.
Source: Consensus EPS and 5 Nov 2026 closing prices; Penmark Research. Forward year by company: NVDA FY28, AMD CY27, AVGO FY27, TSM CY27, QCOM FY27, MRVL FY28.
| Indicator | Thesis confirmed if | Thesis broken if | Next reading |
|---|---|---|---|
| Data Center revenue, $bn | About $100bn in Q3, $115bn in Q4 | Below Q2's $89bn in any quarter | 17 Nov 2026 |
| Gross margin | 74% in Q3; above 71% in Q4 | Below 71% before Rubin is at full mix | 17 Nov 2026 |
| Platform content per GW, $bn | Rubin at $40bn; content up about 5% a year | Growth below 3% a year (bear case) | late Feb 2027 |
| Hyperscaler capex, $tn | Top five above $1.2tn for 2027 | Below $1.0tn, or a 10% cut | Jan-Feb 2027 |
| Working capital, % of revenue | Falling to 13% from 18.5% (July 2026) | Above 20% for two quarters | 17 Nov 2026 |
| Custom ASIC share | Broadcom AI near its $115bn FY27 guide | Above $130bn, or a lab moves to ASIC | Dec 2026 |
| China Data Center revenue | None in our numbers; licenses are upside | Export rules widen to other products | late Feb 2027 |
Source: NVIDIA filings and Q2 FY27 earnings call; Broadcom filings; Penmark Research model. Thresholds are the levels at which the base case gives way to the bear.
Data Center revenue is up 2.5x in six quarters; the shares rose 45% to July and are now 12% below that level
NVIDIA's revenue is not the question; the multiple is. Between Q4 FY25 and Q2 FY27 Data Center revenue rose from $35.6bn a quarter to $89.0bn, a 2.5x increase, while the share price rose 45%, from $142.25 to $206.61 at the end of Q2 FY27, and has since fallen to $181.60. The market now pays 11.4x FY28E consensus EPS. We think revenue keeps compounding on content per gigawatt and margin fades more slowly than the price implies.
Source: NVIDIA 10-Q filings (revenue by market) and closing prices at each quarter end; Penmark Research.
Source: NVIDIA Q2 FY27 earnings call, 26 Aug 2026 (revenue opportunity per gigawatt); Penmark Research.
Source: NVIDIA 10-K; Yahoo Finance consensus (54-57 estimates, Sep-Oct 2026); Penmark Research estimates.
Source: NVIDIA Q2 FY27 guidance (72-73% FY28); Penmark Research scenarios.
Sources: NVIDIA quarterly filings and Q2 FY27 results (26 August 2026); Penmark Research model. Charts show reported data to Q2 FY27A and estimates from Q3 FY27E, shaded. Revenue per gigawatt is our estimate of platform content; the company has defined the opportunity at about $40bn for Vera Rubin, $25bn for Blackwell and $18bn for Hopper.
Our base-case value is $7.25 above the price, and three debates account for all of it
The price is close to our base case, so the call rests on a few debates, not on a re-rating. At 11.4x FY28E earnings the market pays less for NVIDIA than for Broadcom at 15.0x, TSM at 17.0x or AMD at 25.4x, on EPS growth above the peer median (70.9% against 66.6%). Three debates explain the whole gap between our $188.85 base-case value and the price: FY29 capacity, long-run margin and customer financing. All three resolve within nine months.
Source: Penmark Research model: value per share of each catalyst, probability-weighted case assumptions.
Source: Consensus EPS and 5 Nov 2026 closing prices; Penmark Research. Forward year by company as in Exhibit 15.
Source: Penmark Research model: each debate solved on the DCF engine with the other drivers at base case.
Our counter-evidence. The bull case is not free: at 30% probability the bear takes the value to $122.60, and permitting and custom silicon together account for $70 of the $79 gap between our base value ($188.85) and bear value ($109.85) at the price date. That is why the call is Buy with low conviction rather than Buy with size.
Source: NVIDIA confirmed date for Q3 FY27; later dates are Penmark assumptions; Penmark Research model.
| Debate | Market assumption | Penmark assumption | Gap $/sh | Chapter | Resolves |
|---|---|---|---|---|---|
| 1. FY29 capacity growth | +21%, $903bn revenue | +10%, $824bn | −15.59 | 3. Content/GW | Feb 2027 |
| 2. Long-run margin | 52.9% gross margin | 68% in FY36E | +20.77 | 4. Margin fades slowly | May 2027 |
| 3. Customer financing | 18.5% of revenue | 15% easing to 13% | +2.07 | 5. Financing is contained | Nov 2026 |
| Total | +7.25 |
Source: Penmark Research model. Value gap is the change in base-case value per share if the market's assumption is right; a positive gap means our value is above the market's.
Sources: Penmark Research model; consensus from Yahoo Finance (September-October 2026); peer prices 5 November 2026. Catalyst values are the change in base-case value per share if the event resolves in NVIDIA's favour; debate values are the difference between our value and the value implied by the market's assumption.
Three debates worth $7.25 a share: we differ from the market on FY29 capacity, long-run margin and working capital
Everything else is agreed. On FY27 and FY28 revenue we sit within 0.7% of consensus and on FY28E EPS 5.4% above it; the disagreement is what happens after FY29. Each debate below has an observable number, a value attached and a date. Two of the three move against us if we are wrong, which is why the probability-weighted value is $205.86 and not the $288.86 the peer multiple implies.
| Debate | Market view | Penmark view | Evidence | What would change our view | Resolves |
|---|---|---|---|---|---|
| FY29 capacity growth | Consensus FY29 revenue of $903bn implies 21% more capacity shipped, against 10% in our model. | Capacity adds slow to 10% as power and permitting bind; revenue $824bn in FY29E. | Top five hyperscalers guided $1.3tn of 2027 capex, but the company says supply, not demand, caps FY28 growth at about 70%, and we assume supply growth slows after that. | FY29 capacity orders above 15% growth at the Q4 FY27 call. | Feb 2027 |
| Long-run margin | The terminal value implies a 52.9% steady-state gross margin. | Gross margin fades to 68% by FY32E and holds; EBIT margin 57%. | Custom ASICs take inference share and memory costs rise, but the full stack (rack, networking, software) earns a gross margin 4.6pp above Broadcom's. | A second consecutive quarter below 71% gross margin. | May 2027 |
| Customer financing | Working capital stays at 18.5% of revenue, the July 2026 level, as lab stakes and vendor financing grow. | Working capital eases to 15% and then 13% as collections improve and financing moves off balance sheet. | DSO is 61 days; $93.9bn of equity stakes and $59.5bn of commitments are large but covered 1.25x by FY27E free cash flow. | DSO above 70 days or a stake writedown above $10bn. | Nov 2026 |
Source: Penmark Research model; consensus from Yahoo Finance analyst estimates, September-October 2026; company disclosures from the Q2 FY27 10-Q and earnings call.
Source: Penmark Research model. Bars compare the market-implied assumption with ours, each scaled to the larger of the pair; values are the metric itself. Value of each debate: −$15.59, +$20.77 and +$2.07 a share.
| Implied by the price | Price | Our model |
|---|---|---|
| Terminal growth | 2.57% | 3.00% |
| WACC | 12.56% | 12.44% |
| Terminal EBIT margin | 54.9% | 56.6% |
| FY36 gross margin | 52.9% | 68.0% |
| FY29 capacity growth | 21.0% | 10.0% |
Source: Penmark Research model, reverse DCF at 26 July 2026.
Of $409bn of FY27E revenue, $266bn reaches operating profit and $191bn becomes free cash flow
Source: NVIDIA Q2 FY27 10-Q and earnings call; Penmark Research estimates for FY27E. Customer split applies the Q2 FY27 hyperscale share of Data Center revenue (54.7%) to FY27E; Rubin share is the model's 17.4% of Data Center revenue; R&D and SG&A split opex on the H1 FY27 mix (83%/17%). Regions are Q2 FY27 reported revenue by billing location. Capital returned is FY27E dividends and repurchases of $118bn, 62% of free cash flow.
Source: NVIDIA filings; Penmark Research model, FY27E. Content per gigawatt is our estimate of platform content (GPU, CPU, rack, networking, software); the company has described a $40bn opportunity per gigawatt for Vera Rubin against $25bn for Blackwell.
Source: Penmark Research model, FY27E. Bars are to scale: the first row is the income statement, the second how free cash flow is used. Net working capital rises $29bn in FY27E, from $32bn to $61bn.
13.8 gigawatts times $27.6bn of content: capacity is the constraint now, and content per gigawatt is what carries growth once it stops
NVIDIA's revenue equation has two terms and they trade places. In FY27E revenue is 13.8GW of shipped capacity at $27.6bn of platform content, plus $29bn of edge computing. Through FY28 both terms rise; from FY32 capacity is flat in our model and every dollar of growth comes from content, which is why the Rubin-to-LPX roadmap matters more than the order book.
Source: Penmark Research model, FY27E: capacity is Data Center revenue divided by blended content per GW (derived; the company does not disclose gigawatts). Content per GW: company-stated opportunity of $18bn (Hopper), $25bn (Blackwell), $40bn (Vera Rubin), blended by Rubin share.
Source: NVIDIA 10-Q Q2 FY27, revenue by customer type; Penmark Research.
Two consequences. First, growth is less dependent on four capex budgets and more on the financing capacity of a long tail of neoclouds — which is why the working-capital debate matters (Exhibit 26). Second, the company says it has invested nearly $50bn in frontier labs and offers credit support to one of them, so some of this demand is partly financed by NVIDIA itself.
What we watch. The hyperscale share of Data Center revenue each quarter. If it falls below 50% while revenue keeps growing, the demand base is broadening and the FY29 capacity case improves; if it falls because hyperscale is pausing, the opposite.
Source: NVIDIA 10-K filings; Penmark Research estimates.
Sources: NVIDIA Q2 FY27 10-Q and earnings call (26 August 2026); Penmark Research model. Capacity shipped is our estimate of AI factory gigawatts delivered, derived from Data Center revenue and platform content; the company does not disclose gigawatts shipped. Content per gigawatt: $18bn Hopper, $25bn Blackwell, about $40bn Vera Rubin, as described by management.
A gigawatt of Vera Rubin leaves $29bn of gross profit and $26bn of operating profit: the pricing power sits in the rack, not the chip
Source: Penmark Research model, FY28E margin structure applied to the $40bn per gigawatt revenue opportunity stated on the Q2 FY27 call.
Why the customer pays. The alternative is integrating compute, networking and software from separate suppliers, which costs more power and more engineering time per unit of training throughput. Management puts the revenue opportunity at about $40bn per gigawatt for Vera Rubin against $25bn for Blackwell — a 60% step-up for the same physical footprint, which is only possible if the buyer values the integration.
Fixed versus variable cost. Almost all of the cost of revenue is variable: foundry wafers, HBM and assembly are bought against committed volumes ($279bn of supply commitments). Research and development, $29bn in FY27E, is the fixed base, and it is 7.1% of revenue, down from 8.6% in FY26A as revenue scales.
Source: NVIDIA 10-Q Q2 FY27, Note 10 (commitments and guarantees) and Note 9; equity investments from the balance sheet at 26 Jul 2026.
Where the cash goes. Capex is small ($10bn in FY27E, 2.5% of revenue) because the company does not own fabs. Capital return is the policy: 60% of free cash flow in H1 FY27 against a stated 50% floor, $118bn in FY27E, funding a 3% annual reduction in the share count. The rest builds net cash to $572bn by FY30E — capital with no obvious use beyond the $93.9bn already committed to equity stakes in customers.
| Per gigawatt, US$bn unless stated | Hopper (FY24A) | Blackwell (FY26A) | Vera Rubin (FY28E) | Blended (FY30E) |
|---|---|---|---|---|
| Platform content per GW | 18.0 | 25.0 | 40.0 | 42.0 |
| Gross margin | 72.7% | 71.1% | 72.5% | 70.0% |
| Gross profit per GW | 13.1 | 17.8 | 29.0 | 29.4 |
| Operating margin | 54.1% | 60.4% | 65.5% | 61.0% |
| Operating profit per GW | 9.7 | 15.1 | 26.2 | 25.6 |
Source: Penmark Research model. Content per GW uses management's stated opportunity of about $18bn for Hopper, $25bn for Blackwell and $40bn for Vera Rubin; margins are reported company-wide gross and operating margins for FY24A and FY26A and model estimates for FY28E and FY30E, applied to that content. Gigawatts are derived from Data Center revenue and content; the company does not disclose them.
Sources: NVIDIA FY26 10-K, Q2 FY27 10-Q; Q2 FY27 earnings call (26 August 2026); Penmark Research model. Per-gigawatt gross and operating profit are our estimates: $40bn of content at a 72.5% FY28E gross margin and a 65% operating margin, before allocations of fixed research and development.
A $2tn cloud backlog and $1.3tn of 2027 capex against 13.8 gigawatts of supply: the constraint is physical, not financial
The market is measured in gigawatts, not dollars. Every gigawatt of AI factory built requires about $60bn of systems, of which NVIDIA supplies about $40bn. The top five cloud providers have pointed to $1.3tn of 2027 capex against nearly $800bn in 2026, and the industry backlog is above $2tn; the limiting factor is power, grid connections and memory supply, which is why the company guides FY28 growth to about 70% when it describes underlying demand growth as about 100%.
Source: NVIDIA Q2 FY27 earnings call, 26 Aug 2026 (about $60bn of spend and $40bn of NVIDIA revenue opportunity per gigawatt with Vera Rubin); Penmark Research. The model does not split the $40bn by layer.
Source: Penmark Research model. Gigawatts are derived from Data Center revenue and platform content per gigawatt; the company does not disclose them. Solid bars are actuals, hatched are estimates (Penmark Research).
The demand drivers we can date. Top-five capex of about $800bn in 2026 and $1.3tn in 2027; a cloud backlog above $2tn; AI labs, which management expects to be roughly a quarter of next year's business; sovereign and enterprise demand inside the 45% of Data Center revenue that is not hyperscale.
The threats. Substitution by custom silicon is the real one and our bear case prices it: it is worth $27.05 a share in our bear case. Regulation and permitting are larger, at $43.28. Cheaper inference — smaller models, better utilization — would cut the gigawatt requirement, and that risk is not in our bear case because we have no evidence for its timing.
| Demand pool | 2026 | 2027E | What sets the pace | Our assumption |
|---|---|---|---|---|
| Top five hyperscalers | ~$800bn capex | $1.3tn capex | Power, grid interconnection, memory supply | Delivered; hyperscale was 55% of Data Center revenue in Q2 FY27 |
| AI clouds and neoclouds | $40.3bn revenue in Q2 FY27 | +138% y/y in Q2 FY27 | Third-party financing and GPU-backed debt | Financing stays available; 42% of FY27E revenue in our model |
| Edge Computing customers | $22.2bn FY26A revenue | Not guided | Gaming, automotive and workstation demand | Edge Computing is 7% of FY27E revenue |
Source: Q2 FY27 earnings call (26 August 2026) for hyperscaler capex, cloud backlog and the AI-cloud revenue split; Penmark Research model for the FY27E revenue shares. Capex figures are company-stated totals for the top five, calendar years.
Our FY27E Data Center revenue is 3.3x Broadcom's guided AI semiconductor revenue, and NVIDIA's operating margin is 11.2pp higher
Source: Penmark Research model; Broadcom FY27 AI semiconductor guidance of $115bn as reported on 2 Oct 2026.
The advantage that lasts. Not the GPU — the integration. A buyer choosing a rack of 72 GPUs, a CPU, NVLink switching, InfiniBand and a software stack that already runs every model is buying time: integration work they do not have to do. Custom silicon suits a buyer with a large, stable workload that justifies the design cost; that is our assessment, not a company disclosure.
Where the advantage is thinning. Inference, where the workload is standardized and the switching cost is the model rather than the hardware. That is the same place where margins are lowest and where our bear case assumes ASIC share gains.
Source: Penmark Research model, FY27E; Broadcom quarter to 2 Aug 2026 (10-Q).
| Criterion | NVIDIA | AMD | Broadcom custom | Marvell custom | Evidence that it lasts |
|---|---|---|---|---|---|
| Peak throughput per rack | Leader | Close | Behind | Behind | Rack-level integration of GPUs, CPU and networking. |
| Software and models supported | Leader | Improving | Narrow | Narrow | CUDA software stack; management says NVIDIA runs every frontier and open model. |
| Cost per unit of training | Leader | Behind | Leader at scale | Behind | Custom silicon is cheaper for a workload that does not change. |
| Time to deploy | Leader | Tie | Behind | Behind | Vera Rubin production shipments began in August 2026. |
| Supply security | Tie | Behind | Ahead | Ahead | NVIDIA holds $279bn of supply and capacity commitments. |
Source: Penmark Research assessment from company filings, Q2 FY27 earnings call and peer disclosures. Rankings are relative, not absolute: a "behind" is behind the leader on that criterion, not a failing.
Five years of operating cash flow funded $96bn of buybacks and dividends; the quarterly dividend has just risen 25-fold
Source: NVIDIA 10-K filings FY22-FY26 (Penmark Research model, Historicals). The remaining $17.8bn is other financing and non-cash items.
Returns are policy, not residual. Management has committed to return 50% or more of free cash flow and returned 60% in H1 FY27 ($26bn in Q2 alone: $20bn of repurchases and $6bn of dividends). The board had $99.3bn of repurchase authorization left at 26 July 2026.
Insider selling is heavy but concentrated. Director Mark Stevens sold about $947m of stock between 31 August and 18 September, none under a 10b5-1 plan. The CFO and general counsel sold $7.6m and $6.8m in September under 10b5-1 plans. No officer or director bought.
Source: NVIDIA quarterly revenue outlook and reported revenue, Q3 FY25 to Q3 FY26 (Signals KPI series).
| Quarter | Guided, $bn | Reported, $bn | Beat | Next-day move |
|---|---|---|---|---|
| Q3 FY25 | 32.5 | 35.1 | +7.9% | −4.2% |
| Q1 FY26 | 43.0 | 44.1 | +2.5% | +3.3% |
| Q2 FY26 | 45.0 | 46.7 | +3.9% | −0.8% |
| Q3 FY26 | 54.0 | 57.0 | +5.6% | −3.2% |
Source: Company revenue outlook (Signals KPI series) and reported revenue; next-session move from closing prices (yfinance). Next-session moves are descriptive, not causal.
| Date | Insider | Role | Transaction | Shares | Value, $m | 10b5-1 |
|---|---|---|---|---|---|---|
| 18 Sep 2026 | Mark Stevens | Director | Sale (largest of several) | 1,356,000 | 297.9 | No |
| 21 Sep 2026 | Timothy Teter | Officer | Sale | 30,460 | 6.8 | Yes |
| 17 Sep 2026 | Colette Kress | Chief financial officer | Sale | 34,918 | 7.6 | Yes |
| 16 Sep 2026 | Teter, Kress | Officers | Tax withholding | — | — | Yes |
| 7 Oct 2026 | Tench Coxe | Director | Gift | 500,000 | — | Yes |
| 17 Sep 2026 | Jensen Huang | Director, officer | Gift | 438,000 | — | Not flagged |
Source: Form 4 filings via Signals, 31 August to 9 October 2026. Values are transaction price times shares where disclosed; gifts and tax settlements carry no value. Stevens sold about 4.2m shares in total over the period.
Results days move the stock 4.5% on average, and earnings have grown far faster than the price: the call is about capacity and margin
NVIDIA trades on its own news. Over 106 weeks the market and the chip sector explain 57% of the weekly variance, so 43% is company-specific — high for a $4.4tn company. Of five candidate drivers tested against four tests, capacity passes three, margin and capex pass two; custom silicon and working capital stay as bear-case risks.
| Driver | What moves the stock | Tests confirmed | Market reaction vs our value | Debate | Next number to watch |
|---|---|---|---|---|---|
| 1. Data Center revenue and capacity guidance | Revenue beat and the FY28 outlook; the guide, not the print | Event-day, return decomposition, Q&A (3/4) | +8.7% (+$18 a share) after Q2 FY27; we value supply relief at +$21.64 | 1 | Q3 FY27 revenue, 17 Nov: we $108.0bn, consensus $109.0bn |
| 2. Gross margin | The margin guide for the next quarter, and memory cost commentary | Event-day, Q&A (2/4) | Stock fell 8.5% and 5.5% after the two Q4 prints; our margin debate is worth +$20.77 | 2 | Q4 FY27 gross margin, Feb 2027: we 71.5%, guide 71-72% |
| 3. Hyperscaler capex | The top-five capex number and any change to it | Event-day, Q&A (2/4) | Top-five capex ($1.3tn in 2027) opened the call that lifted the stock 8.7%; a step-up is worth +$32.46 | 1 | 2027 capex guides, Jan-Feb 2027 |
| 4. Custom ASIC share (dropped) | Broadcom and Marvell AI revenue, and any lab design win | Q&A only (1/4): bear-case item | Worth −$27.05 in the bear case; 29% of analyst questions on the last call | 2 | Broadcom FY27 AI guide, Dec 2026 |
| 5. Working capital and receivables (dropped) | DSO, inventory and the stake portfolio | None (0/4): bear-case item | No reaction in eight quarters; −$3.98 in the bear case only | 3 | DSO in the Q3 10-Q, 17 Nov: 61 days at Q2 |
Source: Penmark Research tests on eight results events (Q3 FY25 to Q2 FY27), six years of daily prices and 106 weeks of weekly returns to 5 November 2026; Q2 FY27 transcript. Per-share values are the model's catalyst and bear-case contributions.
Source: NVIDIA closing prices (yfinance), one-day move on the trading day after each of the last eight results calls (Signals call dates).
| From | Price | EPS | Multiple | Divs | Total |
|---|---|---|---|---|---|
| Jan 2025 | +27.3% | +169.0% | −52.7% | 0.4% | +27.7% |
| Jan 2024 | +197.6% | +563.0% | −55.1% | 1.0% | +198.5% |
| Jan 2022 | +695.1% | +1,954% | −61.3% | 2.7% | +697.8% |
Source: NVIDIA closing prices to 5 Nov 2026 (yfinance, split-adjusted); GAAP diluted EPS at fiscal year-end against $7.91 trailing four quarters to Q2 FY27 (filings). Multiple change = (1 + price return) / (1 + EPS growth) − 1; total return = price return plus dividends.
Source: Regression of NVDA weekly returns on the S&P 500 and the Philadelphia Semiconductor index, 106 weeks to 5 Nov 2026; Penmark Research.
Source: NVIDIA Q2 FY27 earnings call, 26 Aug 2026: share of seven analyst questions by topic (Signals transcript).
Sources: NVIDIA filings and results releases; closing prices from yfinance to 5 November 2026; Q2 FY27 earnings call. Event-day moves are next-session price changes and are descriptive, not causal.
Capacity is worth $21.64 a share and margin $20.77; the market pays for neither, and pays for 21% FY29 growth we do not forecast
The claim. Data Center revenue growth is set by gigawatts shipped, and the company guides capacity, not demand. It beat its own revenue guide in each of the four quarters in Exhibit 46, and the FY28 outlook of about 70% growth, supply-constrained, is the number that matters.
What the market does. Q2 FY27 revenue of $96.2bn and the FY28 outlook lifted the stock 8.7% the next day, the largest reaction in eight quarters; it fell after six of the other seven results. The market also pays for 21% FY29 capacity growth, $79bn of revenue more than we model.
What our model says. Supply relief worth 10 points of FY28 capacity is $21.64 a share; the FY29 gap the market assumes costs us $15.59. Net, the capacity driver is worth $6.05 a share to us against the market's implied value.
Next number: Q3 FY27 revenue on 17 November 2026 — we model $108.0bn, consensus $109.0bn, guide $106-110bn. A print above the $110bn top of the guide would be the first evidence that supply is easing, which is worth $21.64 a share in our bull case.
The claim. The terminal margin is the largest single disagreement. The price implies a 54.9% terminal EBIT margin and a 52.9% steady-state gross margin; we model 56.6% and 68%.
What the market does. The stock fell after both Q4 prints (−8.5% and −5.5%) and rose 8.7% after Q2 FY27 even though that call guided Q4 gross margin to a 71-72% trough, so the market is looking through the trough to the FY28 range. Margins and memory were one of seven analyst questions.
What our model says. Each 3 points of terminal EBIT margin is worth about $4.6 a share; the full gross-margin debate (68% against the 52.9% implied) is worth $20.77. Our FY28E gross margin of 72.5% sits inside the 72-73% guide.
Next number: Q4 FY27 gross margin in late February 2027 — we model 71.5% against a 71-72% guide. A print below 71% would put the 72-73% FY28 guide at risk; our memory-cost bear case (FY28 gross margin 70.5%) is worth −$8.11 a share.
The claim. Capacity shipped follows what the five largest hyperscalers spend. Management put their capex at nearly $800bn in 2026 and $1.3tn in 2027, and cloud backlog at more than $2tn.
What the market does. That outlook came on the call after which the stock rose 8.7%. Analysts spent one of seven questions on whether the capex outlook, raised from $1tn over three years, was constrained by supply.
What our model says. Capex in line with $1.3tn lifts FY29 capacity growth to 25% against our base 10%, worth $32.46 a share with a 35% probability. Consensus FY29 revenue of $903bn already assumes 21%, so the market has most of it.
Next number: 2027 capex guides from the hyperscalers in late January to early February 2027. A raise above $1.3tn supports the bull case; a cut puts our 10% FY29 capacity growth at risk.
| Candidate tested | Tests run | Result | Why dropped |
|---|---|---|---|
| Custom ASIC share | Q&A | 1 of 4 | Two of seven questions on the last call; no measured price reaction. Kept as a bear case worth −$27.05 a share. |
| China export headlines | Q&A | 0 of 4 | The outlook has assumed no China data center compute revenue; no analyst question on the last call. |
| Interest rates | Factor | 0 of 4 | Adding the weekly change in the 10-year yield lifts R-squared by 2 points, from 57% to 59%. |
| Working capital | Event-day, Q&A | 0 of 4 | No visible reaction in eight quarters; ecosystem financing was one of seven questions. Kept as a bear case worth −$3.98 a share. |
Source: Penmark Research tests; closing prices from yfinance to 5 November 2026.
What we would add with more data. Weekly short-interest and options-skew data around results, and a longer transcript sample for the Q&A test. Neither would change the conclusion that capacity and margin, in that order, are what the stock trades on.
Chapter 3. Content per gigawatt carries the growth once capacity stops: $40bn a gigawatt is 60% more than Blackwell
Consensus has FY29 revenue of $903bn, which implies 21% more capacity shipped, and no published view beyond FY29. The price is consistent with that: our reverse DCF needs 21% capacity growth in FY29 and a 52.9% steady-state gross margin to reproduce $181.60.
The market believes capacity keeps compounding and content stops mattering; our model has capacity flat from FY32E and content growing 5% a year.
Evidence, three independent sources1. Company disclosure. Management has put the revenue opportunity per gigawatt at about $40bn for Vera Rubin against $25bn for Blackwell and about $18bn for Hopper, and says Rubin will be its fastest ramp, at about 20% of Data Center revenue in Q3 FY27.
2. Reported revenue. Data Center revenue grew 96% in FY27E to $380bn in our model, against capacity up 78% and content per gigawatt up 10% from $25.1bn to $27.6bn; the Q2 FY27 reported $89.0bn is 18% above Q1. The content step-up is therefore mostly still to come in FY28E.
3. Product cadence. Vera Rubin began shipping in August 2026 with the Vera CPU and the LPX inference rack, which add content without adding GPUs. Management says Grace CPU revenue exceeds $5bn over the last twelve months and expects it to more than double in FY28.
Our base case has content per gigawatt rising from $37bn in FY28E to $40bn in FY29E and 5% a year after, with capacity flat from FY32E. Supply relief is worth $21.64 a share and a capex step-up $32.46. Our FY29 revenue is 8.7% below consensus on 10% capacity growth against 21%.
The bullish version is a mix story: Rubin at $40bn a gigawatt and a 72.5% gross margin produces about $29bn of gross profit per gigawatt, against about $18bn for Blackwell ($25bn at the 71.1% FY26A margin): 1.6x for the same power connection.
Signposts17 Nov 2026: Rubin share of Data Center revenue, guided at about 20% in Q3 FY27. late Feb 2027: the FY28 capacity outlook and any content commentary for LPX. May 2027: first full quarter of Rubin mix and the price increases that go with it. Dec 2026: Broadcom's AI guide, the read on whether content is moving to custom silicon instead.
Rests on driver 1 (Data Center revenue and capacity guidance), confirmed by three of four tests; links to debate 1 (FY29 capacity growth).
Source: Penmark Research model: capacity in NVIDIA-equivalent gigawatts and blended platform content per gigawatt.
Source: Penmark Research model, revenue build: Rubin-family share of Data Center revenue.
Sources: NVIDIA Q2 FY27 call and 10-Q (26 Aug 2026); Penmark Research model. Gigawatts and content per gigawatt are our estimates; the company does not disclose gigawatts.
Chapter 4. The margin fades slowly: 68% gross margin in FY32E against 52.9% implied by the price
The price implies a 54.9% terminal EBIT margin and a 52.9% steady-state gross margin — below the 69.1% Broadcom earns on its AI business. The market is not forecasting a cyclical dip; it is capitalising a structural decline in pricing power from the current 74%.
Consensus is narrower: a 72-73% FY28 gross margin, in line with guidance, and no published view beyond FY29. The disagreement is therefore entirely in the terminal value, which is 43% of our enterprise value.
Evidence, three independent sources1. Guidance. The company guided Q3 FY27 to 74% ±50bp, a Q4 trough of 71-72% on memory costs, and a FY28 range of 72-73% with price increases from Q1. That is a guide to a trough, not to a fade.
2. Reported margin through a doubling of revenue. Gross margin was 75.0% in Q2 FY27, 71.1% in FY26A and 75.0% in FY25A: volume has not compressed it, which is what a commoditising product would show.
3. The peer benchmark. Broadcom's custom AI business earns 69.1% gross margin and 53.9% operating margin (quarter to 2 August 2026). Even the substitute product — the thing that takes share from NVIDIA — earns a margin above the one the market implies for NVIDIA in perpetuity.
Each 3 points of terminal EBIT margin is worth about $4.6 a share. Holding 68% rather than the implied 52.9% is worth $20.77 a share against the market's assumption, the largest of the three debates and the reason our base-case value ($188.85) sits above the price ($181.60).
The sensitivity cuts both ways: one point of terminal margin moves the target by about $1.5 a share, while one point of discount rate moves it by about $21 (Exhibit 57). The margin debate matters, but FY28 capacity and the discount rate matter more.
Signposts17 Nov 2026: Q3 FY27 gross margin, guided 74% ±50bp. late Feb 2027: the Q4 trough — below 71% and we cut FY28E EPS by about 4%. May 2027: the first quarter of FY28 price increases, the test of whether the 72-73% guide holds. Dec 2026: Broadcom's AI margin disclosure, the benchmark for custom silicon economics.
Rests on driver 2 (gross margin), confirmed by two of four tests; links to debate 2 (long-run margin durability).
Source: NVIDIA 10-K and 10-Q filings; guidance of 26 Aug 2026; Broadcom quarter to 2 Aug 2026; Penmark Research estimates.
Source: Penmark Research model sensitivities: probability-weighted value rolled to the 12-month target date.
| Margin path | FY26A | FY27E | FY28E | FY29E | FY30E | FY31E | FY32E |
|---|---|---|---|---|---|---|---|
| Revenue, $bn | 215.9 | 409.0 | 695.2 | 824.5 | 908.9 | 973.0 | 1,022.1 |
| Gross margin | 71.1% | 73.7% | 72.5% | 71.0% | 70.0% | 69.0% | 68.0% |
| Operating expenses, % of revenue | 10.7% | 8.5% | 7.0% | 8.0% | 9.0% | 10.0% | 11.0% |
| EBIT margin | 60.4% | 65.1% | 65.5% | 63.0% | 61.0% | 59.0% | 57.0% |
| Gross margin, market-implied | 52.9% |
Source: Penmark Research model. Operating expenses include stock-based compensation. The market-implied gross margin is the steady-state margin in the terminal value that reproduces the 5 November 2026 share price of $181.60.
Sources: NVIDIA Q2 FY27 earnings call and 10-Q; FY26 10-K; Broadcom 10-Q for the quarter to 2 August 2026; Penmark Research model. Terminal margin sensitivities are on the probability-weighted value rolled to the 12-month target date; the tornado shows each input moved alone, all others at base.
Chapter 5. The financing risk is contained: $154bn of commitments against $191bn of FY27E free cash flow
Working capital stays at 18.5% of trailing revenue, the July 2026 level, and the stake portfolio is worth face value. That combination costs $2.07 a share against our base case — the smallest of the three debates, but the one that would turn a growth story into a credit story.
The concern is circularity: NVIDIA invests in labs, the labs buy GPUs, and the revenue funds the next round. In the bear case we haircut private stakes by 50% and hold working capital at 20% of revenue.
Evidence, three independent sources1. Size against cash generation. Stakes and commitments total $153.5bn against $191.5bn of FY27E free cash flow and $105bn of cash at January 2027. A 50% loss on the entire $153.5bn would be $77bn, 0.4x one year of free cash flow.
2. Off-balance-sheet financing is the stated strategy. The company has partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to raise more than $500bn of third-party capital — the point is that the financing moves to parties who can hold it, not onto NVIDIA's balance sheet.
3. Receivables quality. Days sales outstanding were 61 at Q2 FY27 and inventory was $32bn, both as stated by management. Direct customers were 16% of revenue in Q2 FY27, a concentration we track each quarter.
Working capital at 13% of revenue rather than 18.5% is worth $2.07 a share; the full bear-case combination — 20% working capital and a 50% stake haircut — is already inside the $122.60 bear value and contributes $4.0 of it.
The asymmetry is what matters: the financing risk is the smallest of the three debates in value but the largest in narrative, which is why the stock reacts to it more than the numbers justify.
Signposts17 Nov 2026: DSO and inventory in the Q3 10-Q — above 70 days and the bear case gains weight. Q4 FY27: the stake portfolio mark and any new commitments. Through 2027: whether the $500bn third-party financing is drawn, and on what terms. Ongoing: the Department of Justice review of the Hugging Face transaction, a live regulatory file.
Rests on the dropped working-capital candidate (passed none of the four tests), kept as a bear-case item; links to debate 3 (customer financing and working capital).
Source: Penmark Research scenarios; net working capital at Jul 2026 was 18.5% of trailing revenue (10-Q Q2 FY27).
Source: Penmark Research model, bear case: 50% haircut on $51.2bn non-marketable stakes and 50% of $59.5bn support commitments crystallizing; listed stakes at carrying value.
| Balance-sheet exposure, $bn | Jul 2026 | Bear haircut | Bear loss | Note |
|---|---|---|---|---|
| Marketable equity securities | 42.8 | 0% | 0.0 | Listed stakes, marked to market each quarter |
| Non-marketable equity securities | 51.2 | 50% | 25.6 | Private AI labs and neoclouds |
| Total equity stakes | 93.9 | 27% | 25.6 | Balance-sheet carrying value |
| Customer and supplier support commitments | 59.5 | 50% | 29.8 | Base case assumes 10% crystallizes ($6.0bn) |
| Total exposure | 153.5 | 36% | 55.4 | 0.44x FY28E free cash flow of $345bn |
| Net working capital, % of revenue | 18.5% | 20.0% | — | Base case 13.0% by FY28E |
Source: NVIDIA Q2 FY27 10-Q, balance sheet at 26 July 2026; Penmark Research model for the scenario columns. Marketable securities are marked to the 26 July 2026 balance-sheet value.
Sources: NVIDIA Q2 FY27 10-Q and earnings call (26 August 2026); FY26 10-K; Penmark Research model. Stakes and commitments are the marketable and non-marketable equity investments and the customer and supplier support commitments disclosed at July 2026; free cash flow is operating cash flow less capex.
Revenue grew 8x from FY22 to FY26 and free cash flow was 80-91% of net income in every year
Source: NVIDIA 10-K filings FY22-FY26: free cash flow (operating cash flow less capex) over net income.
Source: NVIDIA 10-K filings; Penmark Research estimates. GAAP diluted EPS; the low FY23 base of $0.17 inflates FY24 growth.
| US$bn unless stated | FY22A | FY23A | FY24A | FY25A | FY26A | FY27E |
|---|---|---|---|---|---|---|
| Revenue | 26.9 | 27.0 | 60.9 | 130.5 | 215.9 | 409.0 |
| Growth | 61.4% | 0.2% | 125.9% | 114.2% | 65.5% | 89.4% |
| Gross margin | 64.9% | 56.9% | 72.7% | 75.0% | 71.1% | 73.7% |
| Operating margin | 37.3% | 15.7% | 54.1% | 62.4% | 60.4% | 65.1% |
| Net income, reported | 9.8 | 4.4 | 29.8 | 72.9 | 120.1 | 242.8 |
| Stock-based compensation | 2.0 | 2.7 | 3.5 | 4.7 | 6.4 | 9.0 |
| Investment gains and other | 0.1 | 0.0 | 0.2 | 1.0 | 9.0 | 23.4 |
| Free cash flow | 8.1 | 3.8 | 27.0 | 60.9 | 96.7 | 191.5 |
| Free cash flow / net income | 0.83x | 0.87x | 0.91x | 0.84x | 0.81x | 0.79x |
| Diluted EPS, reported, $ | 0.39 | 0.17 | 1.19 | 2.94 | 4.90 | 10.03 |
| Diluted EPS, adjusted, $ | — | — | — | — | — | 9.44 |
Source: NVIDIA 10-K filings FY22-FY26; Penmark Research estimates for FY27E. FY27E adjusted EPS excludes stock-based compensation and investment gains; the model does not carry adjusted EPS for earlier years.
| Growth decomposition | FY27E | FY28E | FY29E | FY30E | FY31E |
|---|---|---|---|---|---|
| Capacity shipped, GW | 13.8 | 17.9 | 19.7 | 20.7 | 21.1 |
| Capacity growth | +78% | +30% | +10% | +5% | +2% |
| Blended content per GW, $bn | 27.6 | 37.0 | 40.0 | 42.0 | 44.1 |
| Content growth | +10% | +34% | +8% | +5% | +5% |
| Data Center revenue growth | +96% | +74% | +19% | +10% | +7% |
Source: Penmark Research model. Capacity and content are derived from Data Center revenue; the product of the two growth rates reconciles to the revenue growth shown.
Sources: NVIDIA 10-K filings FY22-FY26 and Q2 FY27 10-Q; Penmark Research model. Growth split between volume, content and price is not disclosed by the company; our revenue build attributes growth to gigawatts shipped and content per gigawatt (Exhibit 54).
Forecasts. Data Center revenue of $662bn in FY28E and $868bn in FY30E, then content only: the revenue build from the driver up
| Revenue build, $bn | FY26A | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| Data Center | 193.7 | 380.0 | 662.0 | 787.2 | 867.9 |
| Growth | 68.2% | 96.1% | 74.2% | 18.9% | 10.3% |
| Edge computing | 22.2 | 28.9 | 33.2 | 37.2 | 41.0 |
| Growth | 45.0% | 30.2% | 15.0% | 12.0% | 10.0% |
| Total revenue | 215.9 | 409.0 | 695.2 | 824.5 | 908.9 |
| Growth | 65.5% | 89.4% | 70.0% | 18.6% | 10.2% |
| Capacity shipped, GW | 7.7 | 13.8 | 17.9 | 19.7 | 20.7 |
| Content per GW, $bn | 25.1 | 27.6 | 37.0 | 40.0 | 42.0 |
| Data Center share of revenue | 89.7% | 92.9% | 95.2% | 95.5% | 95.5% |
Source: NVIDIA FY26 10-K; Penmark Research model. Capacity and content are our estimates; the company does not disclose gigawatts. FY26A content per gigawatt is derived from FY26 Data Center revenue and estimated capacity.
Capacity. FY28E +30%, in line with the company's supply-constrained guide of about 70% revenue growth; FY29E +10% as we assume supply limits persist; +5% in FY30E and flat from FY32E.
Content. $27.6bn per gigawatt in FY27E, $37.0bn in FY28E as Rubin becomes the whole mix, $40bn in FY29E with LPX and the Vera CPU, then 5% a year.
Edge. Gaming, automotive and workstation, growing 30% in FY27E and 10-15% a year after, from a small base.
Where we differ from consensusFY27E revenue is 0.7% below consensus and FY28E 0.4% above. The difference is FY29E: our $824bn is 8.7% below the $903bn consensus, because consensus implies 21% capacity growth and we model 10%.
Costs and marginsGross margin 73.7% in FY27E, 72.5% in FY28E, fading to 68% by FY32E. Operating expenses fall to 7.0% of revenue in FY28E as revenue outgrows R&D, then rise to 11% by FY32E. Operating margin peaks at 65.5% in FY28E and fades to 57% by FY32E.
Source: Penmark Research scenarios: total revenue, capacity shipped x blended content per gigawatt plus Edge Computing, FY28E to FY36E.
| Guidance against our model | Period | Guided | Penmark | Gap to midpoint | Inside range |
|---|---|---|---|---|---|
| Revenue, $bn | Q3 FY27 | 106-110 | 108.0 | 0.0 | Yes |
| Gross margin | Q3 FY27 | 73.5-74.5% | 74.0% | 0.0pp | Yes |
| Gross margin, trough | Q4 FY27 | 71-72% | 71.5% | 0.0pp | Yes |
| Gross margin | FY28 | 72-73% | 72.5% | 0.0pp | Yes |
| Revenue growth | FY28 | ~68-72% | 70.0% | +0.0pp | Yes |
| Operating expense growth | FY27 | low 50s % | 51.5% | 0.0pp | Yes |
| Tax rate | FY27 | 16-18% | 16.5% | −0.5pp | Yes |
| Vera Rubin share of Data Center revenue | Q3 FY27 | ~20% | 20% | 0.0pp | Yes |
Source: Q2 FY27 earnings call, 26 August 2026 (CFO outlook); Penmark Research model. Operating expense growth and tax rate on a GAAP basis.
Sources: NVIDIA FY26 10-K, Q2 FY27 10-Q and earnings call; consensus from Yahoo Finance analyst estimates, September-October 2026; Penmark Research model. Scenario paths are capacity shipped multiplied by blended content per gigawatt, FY28E to FY36E.
Forecasts. Cost of revenue is 26-30% of sales, so operating margin follows gross margin: 65.5% in FY28E, 57% by FY32E
| Cost structure, $bn | FY26A | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| Revenue | 215.9 | 409.0 | 695.2 | 824.5 | 908.9 |
| Cost of revenue | 62.5 | 107.7 | 191.2 | 239.1 | 272.7 |
| % of revenue | 28.9% | 26.3% | 27.5% | 29.0% | 30.0% |
| Gross profit | 153.5 | 301.2 | 504.1 | 585.4 | 636.2 |
| Operating expenses | 23.1 | 35.0 | 48.7 | 66.0 | 81.8 |
| % of revenue | 10.7% | 8.5% | 7.0% | 8.0% | 9.0% |
| of which stock compensation | 6.4 | 9.0 | 13.9 | 16.5 | 20.0 |
| Operating profit | 130.4 | 266.3 | 455.4 | 519.4 | 554.4 |
| Operating margin | 60.4% | 65.1% | 65.5% | 63.0% | 61.0% |
| Net financial items | 11.1 | 24.5 | 2.7 | 8.0 | 14.8 |
| Tax | 21.4 | 48.1 | 75.6 | 89.7 | 96.8 |
| Net income | 120.1 | 242.8 | 382.5 | 437.8 | 472.5 |
Source: NVIDIA FY26 10-K; Penmark Research model. Costs are shown positive. FY27E net financial items include $23.4bn of investment gains, which we exclude from adjusted EPS.
Cost of revenue moves with volume: 26.3% of sales in FY27E, rising to 30% by FY30E as pricing fades. Operating expenses are the semi-fixed base: $35.0bn in FY27E, of which about $29.2bn is research and development and $5.8bn selling and administrative on the H1 FY27 mix (our split). They grow 51.5% in FY27E, in line with the low-50s guide, then slower than revenue in FY28E.
Cash, capex and working capitalCapex is 2.5% of revenue and stays there: the company does not own fabs. Working capital is the swing factor, from 18.5% of revenue in July 2026 to 13% in our base case by FY28E; each 1 point of revenue is $7bn of cash. Free cash flow reaches 50% of revenue in FY28E and peaks at 53% in FY29E.
Balance sheetNet cash of $72bn at January 2027 rises to $572bn by FY30E after $207bn of buybacks and dividends in FY28E. Debt is unchanged at $33bn, giving net debt/EBITDA of −0.7x by FY29E. No equity issuance is assumed.
Why our EPS is above consensusFY27E adjusted EPS of $9.44 is 1.4% above consensus and FY28E $16.76 is 5.4% above. Revenue is not the difference; operating expenses at 7.0% of FY28E revenue are.
| Sensitivity of the 12-month target | Downside | Base | Upside | Range, $ | What it tests |
|---|---|---|---|---|---|
| FY28 capacity ±10 points | $173.0 | $205.86 | $239.0 | 66.0 | The supply constraint, and with it the FY28 outlook |
| Discount rate ±1.25 points | $182.55 | $205.86 | $236.29 | 53.7 | Cost of equity at a 12.5% base |
| Terminal growth ±0.5 point | $202.94 | $205.86 | $209.06 | 6.1 | Whether AI capex has a long tail |
| Terminal margin ±3 points | $201.24 | $205.86 | $210.47 | 9.2 | The margin debate (Chapter 4) |
Source: Penmark Research model, probability-weighted value rolled to the 12-month target date. Each input is moved alone, all others at base; the base column is the same $205.86 as page 1.
| Cash flow and balance sheet, $bn | FY26A | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| Operating cash flow | 102.7 | 201.7 | 362.5 | 454.7 | 494.3 |
| Capex | −6.0 | −10.2 | −17.4 | −20.6 | −22.7 |
| Free cash flow | 96.7 | 191.5 | 345.1 | 434.1 | 471.6 |
| FCF margin | 44.8% | 46.8% | 49.6% | 52.7% | 51.9% |
| Dividends | −1.0 | −18.4 | −23.9 | −23.1 | −22.1 |
| Buybacks and withholding | −40.1 | −99.8 | −183.1 | −237.4 | −260.8 |
| Cash and securities | 49.7 | 105.2 | 243.2 | 416.8 | 605.5 |
| Debt | 8.5 | 33.4 | 33.4 | 33.4 | 33.4 |
| Net cash | 41.2 | 71.8 | 209.8 | 383.5 | 572.1 |
| Net debt / EBITDA | −0.31x | −0.26x | −0.45x | −0.72x | −1.01x |
Source: NVIDIA FY26 10-K; Penmark Research model. Buybacks are set to keep capital return at about 60% of free cash flow, above the company's stated 50% floor.
Sources: NVIDIA FY26 10-K, Q2 FY27 10-Q; consensus from Yahoo Finance, September-October 2026; Penmark Research model.
Valuation. A ten-year DCF at a 12.5% discount rate gives $205.86; the 17.0x peer multiple gives $288.86, and we do not use it
Source: Penmark Research model: DCF values rolled 12 months at the 12.5% cost of equity; P/E cross-check on FY28E EPS at the 17.0x peer median.
NVIDIA is a capital-light, cash-generative compounder whose value sits in the terminal period: 43% of enterprise value is beyond FY36. A multiple cannot price a margin that fades over ten years, so the ten-year unlevered DCF is the primary method, with a forward P/E cross-check at the FY28E EPS against the AI-semiconductor peer median.
The assumptions that drive itDiscount rate 12.44%. Cost of equity from a 4.84% risk-free rate (10-year Treasury, 5 November 2026), a 4.23% equity risk premium and a 1.81 Blume-adjusted beta on a 2.21 raw five-year monthly regression, giving 12.50%; WACC 12.44% on $33bn of debt at 5.0% and a 21% tax rate.
Terminal growth 3.0%. Nominal, with a 25% terminal return on new capital and a 56.6% terminal EBIT margin.
Terminal value 43% of enterprise value. Low for a growth company and the main defense against the charge that the target rests on the fade.
Cross-checkAt 17.0x FY28E EPS of $16.76 the stock is worth $288.86, 57% above the DCF. We do not average them: the multiple capitalizes peak earnings without a fade, and the fade is the debate.
| WACC \ g | 2.00% | 2.50% | 3.00% | 3.50% | 4.00% |
|---|---|---|---|---|---|
| 9.94% | 228.30 | 233.95 | 240.38 | 247.79 | 256.41 |
| 11.19% | 197.03 | 200.56 | 204.51 | 208.94 | 213.97 |
| 12.44% | 173.34 | 175.64 | 178.16 | 180.94 | 184.04 |
| 13.69% | 154.80 | 156.33 | 157.99 | 159.80 | 161.78 |
| 14.94% | 139.89 | 140.93 | 142.05 | 143.26 | 144.56 |
Source: Penmark Research model, probability-weighted value per share at 26 July 2026 in US$; discount rate (rows) against terminal growth (columns). The headline $205.86 is this grid's centre rolled 12 months at the 12.5% cost of equity.
| Case | Probability | Value at 26 Jul 2026 | Value at 5 Nov 2026 | 12-month target | Total return | What has to happen |
|---|---|---|---|---|---|---|
| Bear | 30% | $106.54 | $109.85 | $122.60 | −31.9% | FY29 capacity −10%, FY32 gross margin 62%, working capital 20%, 50% stake haircut |
| Base | 50% | $182.98 | $188.85 | $211.47 | +17.0% | FY28 capacity +30%, FY29 +10%, flat from FY32E, FY32 gross margin 68%, working capital 13% |
| Bull | 20% | $273.53 | $282.42 | $316.73 | +75.0% | FY28 capacity +40%, FY29 +25%, FY32 gross margin 71%, content +7% a year |
| Probability-weighted | 100% | $178.16 | $183.86 | $205.86 | +13.9% | Weighted at the probabilities shown, rolled at the 12.5% cost of equity |
Source: Penmark Research model, Scenarios tab. Values are per diluted share; the probability-weighted value reconciles to the DCF case to the cent.
Sources: Penmark Research model: DCF, Scenarios and Sensitivities tabs. Risk-free rate is the 10-year US Treasury at 5 November 2026; the equity risk premium (4.23%) is the Damodaran estimate and the beta is the Blume-adjusted five-year monthly beta.
Valuation. The price implies 2.57% terminal growth and a 54.9% terminal margin, and one point on the discount rate is the difference between Buy and Hold
Source: Penmark Research model: reverse DCF solving for the input that reproduces the 5 Nov 2026 price of $181.60.
Source: Penmark Research model: probability-weighted value per share at the valuation date, 26 Jul 2026.
Holding everything else at base, $181.60 requires a 2.57% terminal growth rate against our 3.00%, a 54.9% terminal EBIT margin against our 56.6%, and a 12.56% WACC against our 12.44%. The price is not demanding: it is 3.8% below our base-case value at the price date, $188.85.
That is the honest description of this call. The stock is fairly valued on our numbers, and the 13.9% total return comes from rolling the value forward at the cost of equity rather than from a re-rating.
Two methods, and why we do not average themThe DCF gives $183.86 at the price date and $205.86 rolled 12 months. The peer multiple at 17.0x FY28E EPS gives $288.86. The 57% gap comes mostly from the terminal period: a market multiple capitalizes FY28 earnings as if they persist, and our DCF fades gross margin by 5.7 points and stops adding capacity from FY32.
We report both, lead with the DCF because it is the only method that prices the fade, and treat the multiple as what a peer-relative buyer would pay — the upper bound of the range, not a target.
A 1.81 Blume-adjusted beta, a 4.23% equity risk premium and a 4.84% risk-free rate give a cost of equity of 12.50% and a WACC of 12.44% with $33bn of debt at 5.0% and a 21% tax rate. The stock's own weekly volatility is 40.4%, and its raw five-year monthly beta against the S&P 500 is 2.21, Blume-adjusted to 1.81 in the model.
A discount rate one point higher takes the target from $205.86 to $188.79, a Hold. A 10-point miss on FY28 capacity takes it to $173, also a Hold. That is why the rating is Buy with low conviction.
What would make us change the methodIf FY29 orders showed the 21% capacity growth the market assumes, our value would rise by $15.59 a share and the multiple method would deserve more weight. If gross margin printed below 68% before FY30E, the DCF would fall toward the bear and the multiple would stop being a cross-check at all.
| Method | Value | vs price | What it assumes | Weight in our call |
|---|---|---|---|---|
| Ten-year DCF, base case | $188.85 | +4.0% | FY28 capacity +30%, flat from FY32E, FY32 gross margin 68%, 12.44% WACC | Primary: the target is the probability-weighted DCF rolled 12 months |
| Ten-year DCF, probability-weighted, rolled 12 months | $205.86 | +13.4% | 30/50/20 bear-base-bull weighting of the three scenarios | Headline target |
| Peer forward P/E at 17.0x FY28E EPS | $288.86 | +59.1% | FY28E EPS of $16.76 persists; no fade; peers re-rate | Cross-check only: upper bound |
| Price implied by our model | $181.60 | — | 2.57% growth, 54.9% terminal margin, 12.56% WACC | What the market is paying |
Source: Penmark Research model. The peer multiple uses the median forward P/E of AMD, Broadcom, TSM, Qualcomm and Marvell at 5 November 2026 and our FY28E EPS; the two methods are not averaged and the gap is explained above.
Catalysts. Seven dated events in twelve months, and the February 2027 outlook is the one most likely to change the rating
The next twelve months are dense with dated events. Three results prints, the hyperscaler capex guides, Broadcom's AI outlook and the FY28 margin path all land before November 2027. Our estimates sit inside guidance, so the risk is the outlook that comes with the print.
Source: NVIDIA confirmed date for Q3 FY27; later results dates are Penmark assumptions; Penmark Research model.
| Date | Event | Penmark expectation | Consensus or guide | Value at stake | Rating impact |
|---|---|---|---|---|---|
| 17 Nov 2026 | Q3 FY27 results and 10-Q | Revenue $108.0bn, gross margin 74.0%, Rubin about 20% of Data Center revenue, DSO 61 days | Consensus $109.0bn; guide $106-110bn and 73.5-74.5% | Supply relief is worth $21.64 a share | None unless DSO rises well above 61 days |
| Dec 2026 | Broadcom FY27 AI semiconductor guide | Broadcom guides AI semiconductor revenue to $115bn in FY27 (about 30% of our FY27E Data Center revenue) | No published consensus in our sources | −$27.05 in the custom-silicon bear case | Negative if custom silicon gains share |
| late Feb 2027 | Q4 FY27 results and FY28 outlook | Revenue $123.1bn, gross margin 71.5%, FY28 outlook of about +70% ($695bn) | Consensus Q4 revenue $124.5bn; guide trough 71-72% | About $33 a share on a ±10 point change in FY28 capacity | A weaker FY28 outlook would push the return toward Hold |
| Jan-Feb 2027 | Hyperscaler 2027 capex | Top five in line with the $1.3tn already indicated | $1.3tn indicated by the companies | +$32.46 in the bull case | Positive if above $1.3tn |
| May 2027 | Q1 FY28 results | First quarter of FY28 price increases; gross margin at or above 72% | Guide 72-73% for FY28 | $20.77 if the margin holds to FY32E | Higher conviction if the margin holds |
| Aug 2027 | Q2 FY28 results | Vera Rubin at full mix; LPX revenue disclosed | Not guided | Content per gigawatt above $40bn | Positive if LPX adds content |
| Nov 2027 | Q3 FY28 results | First read on FY29 capacity | Not guided | −$15.59 if capacity growth is 21% | Moves toward Hold |
Source: Company results calendar and Q2 FY27 earnings call; consensus from Yahoo Finance, September-October 2026; Penmark Research model for the values at stake.
Risks. Permitting and custom silicon together are worth −$70 a share and would take the stock to $119
Source: Penmark Research model: probability and value per share of each bear catalyst in Exhibit 21.
Permitting and custom silicon breaking together are worth −$70.3 a share, taking the value at the price date from $188.85 to $118.52, 35% below the price. Alone, custom silicon takes it to $161.80 and permitting to $145.57, both below the price and above the $109.85 bear value.
Material risks we do not priceKey person. The founder is chief executive; no succession date attaches to the risk. Regulation. Export controls already exclude China data center compute from the outlook; broader controls covering inference parts would cost share rather than margin. Legal. The Department of Justice review of the Hugging Face transaction is a live file; a blocked deal would cost little, a broader theory of harm would cost more.
ESG where it moves valuePower and water use at gigawatt scale is a permitting risk, not a reputational one: the same approvals that gate capacity additions gate revenue. The company has a single share class; we put no value on governance either way.
| Risk | What happens | Likelihood | Value per share | Confirmed by | Mitigant |
|---|---|---|---|---|---|
| Permitting and AI regulation | Data Center approvals slow after the midterms, curbing capacity additions from FY28 | 35% | −$43.28 | Federal permitting or AI rules; grid-connection delays reported by hyperscalers | Hyperscalers are securing power and sites ahead of need |
| Custom ASIC competition | Lab-designed accelerators win inference share and take pricing from FY28 | 35% | −$27.05 | Broadcom AI revenue above its $115bn FY27 guide; a frontier model trained on ASIC | Custom parts use the same fabs and memory, so they do not add supply |
| Memory cost | HBM and DRAM pricing keeps gross margin below 70% into FY28 | 40% | −$8.11 | Q4 FY27 gross margin below 71% | Long-term supply agreements and price increases from Q1 FY28 |
| Customer financing and receivables | Stakes and vendor financing impair; working capital stays at 20% of revenue | 25% | −$3.98 | DSO above 70 days; a stake writedown above $10bn | $500bn of third-party financing arranged with Apollo, BlackRock and KKR |
| Upside to our base case | Supply relief adds 10 points of FY28 capacity and Rubin holds 74% gross margin | 25% | +$21.64 | FY28 capacity guide above +35% in February 2027 | Not a mitigant: this is the bull case |
Source: Penmark Research model, bear-case value contributions at 26 July 2026; probabilities are our judgement. The sum of the four downside risks is not the bear case: they do not all occur at full value together, and the bear case applies them jointly at 30% probability.
| Risk combination | Value | vs price | Read-across | Comment |
|---|---|---|---|---|
| Base case, no risk crystallizes | $188.85 | +4.0% | Above price | Our value is $7.25 above the price |
| Memory cost alone | $180.74 | −0.5% | Below base | Gross margin at 70.5% rather than 72.5% in FY28E |
| Financing alone | $184.87 | +1.8% | Above price | Working capital at 20% and a 50% stake haircut |
| Custom silicon alone | $161.80 | −10.9% | Below price | FY29 capacity 10% lower as ASIC share is won |
| Permitting alone | $145.57 | −19.8% | Below price | Capacity additions curbed from FY28 |
| Permitting and custom silicon together | $118.52 | −34.7% | Below price | Sum of the two largest bear contributions; the full bear case is $109.85 |
Source: Penmark Research model. Each row subtracts the risk value from the base-case value at the price date and compares it with the $181.60 share price; a value below the price is not a rating change: ratings use 12-month total return to the target (page 29).
Ownership and trading. Short interest is 1.25% of shares and the Street target is 60% above ours: positioning is not the constraint
Source: FINRA bi-monthly short interest, 24 settlements to 15 Oct 2026; shown as a share of basic shares outstanding (approximate; float not used).
Source: MarketBeat (53 ratings, Oct 2026) and TIKR (2 Nov 2026); Penmark Research.
| Shares outstanding | 24,147m |
| Market capitalization | $4,385.1bn |
| Average daily value, 65 days | $25.4bn |
| Short interest, 15 Oct 2026 | 301.6m shares, 1.25% |
| Days to cover | 2.91 |
| Change vs prior settlement | −3.1% |
| Street ratings, buy / hold / sell | 51 / 0 / 1 |
| Street target, mean / low / high | $329 / $218 / $500 |
| FY28E EPS revisions, 30 days | 46 up, 0 down |
| Share class | Single class, one vote |
| Index membership | S&P 500, Nasdaq-100, SOX |
| Lock-ups and overhangs | None |
Source: FINRA short interest to 15 October 2026; MarketBeat and TIKR ratings and targets, October-November 2026; yfinance volume to 5 November 2026; Yahoo Finance estimate revisions. Short interest as a share of shares outstanding uses 24,147m shares.
Short positioning is modest. Short interest is 1.25% of basic shares and 2.91 days of volume, inside its one-year range of 229m to 324m shares. The stock fell 9.3% on 5 November as the chip index fell 12.2%, a sector move rather than a short attack. Of 52 ratings, 51 are buys and one is a sell, and the mean target is 60% above ours.
Why we are below the StreetThe Street's $329 mean target is 19.6x our FY28E non-GAAP EPS of $16.76, above the 17.0x peer-multiple cross-check (Exhibit 77) and not a DCF. Our target is the probability-weighted DCF. The difference is not the revenue forecast, which is within 1% of consensus for FY27 and FY28, but that we fade gross margin to 68% by FY32E.
Liquidity and flowAverage daily traded value was $25.4bn over the 65 sessions to 5 November 2026, ample for institutional positions. The company has a single share class.
| Target | Value | vs price | P/E on FY27E EPS | P/E on FY28E EPS | What it implies |
|---|---|---|---|---|---|
| Street high | $500 | +175% | 52.9x | 29.8x | A peer re-rating on top of the FY28E earnings we already forecast |
| Street mean | $329 | +81% | 34.8x | 19.6x | Above the 17.0x peer median applied to our FY28E EPS, with no fade after |
| Street low | $218 | +20% | 23.1x | 13.0x | The Street low is 6% above our target |
| Penmark target | $205.86 | +13.4% | 21.8x | 12.3x | Probability-weighted DCF rolled 12 months at the cost of equity |
| Price, 5 Nov 2026 | $181.60 | — | 19.2x | 10.8x | 3.8% below our base-case value at the price date |
Source: MarketBeat (53 ratings, October 2026) and TIKR (2 November 2026) for Street targets; Penmark Research model for EPS estimates and our target. Multiples use our non-GAAP EPS of $9.44 in FY27E and $16.76 in FY28E.
Sources: FINRA bi-monthly short interest series to 15 October 2026; Form 4 filings; MarketBeat (53 ratings, October 2026) and TIKR (2 November 2026); Yahoo Finance analyst estimates; Penmark Research. Institutional 13F detail was not used in this report.
Financial model. Revenue of $909bn by FY30E, a 52% net margin and $572bn of net cash, with no equity issuance
| US$bn, Jan YE | FY26A | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| Data Center revenue | 193.7 | 380.0 | 662.0 | 787.2 | 867.9 |
| Edge computing revenue | 22.2 | 28.9 | 33.2 | 37.2 | 41.0 |
| Total revenue | 215.9 | 409.0 | 695.2 | 824.5 | 908.9 |
| Cost of revenue | 62.5 | 107.7 | 191.2 | 239.1 | 272.7 |
| Gross profit | 153.5 | 301.2 | 504.1 | 585.4 | 636.2 |
| Operating expenses | 23.1 | 35.0 | 48.7 | 66.0 | 81.8 |
| Operating income | 130.4 | 266.3 | 455.4 | 519.4 | 554.4 |
| Net financial items | 11.1 | 24.5 | 2.7 | 8.0 | 14.8 |
| Tax | 21.4 | 48.1 | 75.6 | 89.7 | 96.8 |
| Net income | 120.1 | 242.8 | 382.5 | 437.8 | 472.5 |
| Depreciation & amort. | 2.8 | 4.9 | 9.0 | 11.5 | 13.6 |
| Stock-based comp. | 6.4 | 9.0 | 13.9 | 16.5 | 20.0 |
| Operating cash flow | 102.7 | 201.7 | 362.5 | 454.7 | 494.3 |
| Capex | −6.0 | −10.2 | −17.4 | −20.6 | −22.7 |
| Free cash flow | 96.7 | 191.5 | 345.1 | 434.1 | 471.6 |
| Dividends and buybacks | 41.1 | 118.2 | 207.1 | 260.4 | 282.9 |
| US$bn, Jan YE | FY26A | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| Cash and securities | 49.7 | 105.2 | 243.2 | 416.8 | 605.5 |
| Equity investments | 35.1 | 93.9 | 93.9 | 93.9 | 93.9 |
| Net working capital | 31.9 | 61.3 | 104.3 | 115.4 | 127.2 |
| PP&E | 10.4 | 17.3 | 25.6 | 34.7 | 43.8 |
| Goodwill & intangibles | 24.1 | 24.1 | 24.1 | 24.1 | 24.1 |
| Total net assets | 165.8 | 319.3 | 508.6 | 702.4 | 912.0 |
| Debt | 8.5 | 33.4 | 33.4 | 33.4 | 33.4 |
| Shareholders' equity | 157.3 | 285.9 | 475.2 | 669.1 | 878.6 |
| Diluted shares, m | 24,514 | 24,213 | 23,508 | 22,585 | 21,610 |
| EPS, reported, $ | 4.90 | 10.03 | 16.27 | 19.38 | 21.86 |
| EPS, adjusted, $ | 4.90 | 9.44 | 16.76 | 19.99 | 22.63 |
| Dividend per share, $ | 0.04 | 0.76 | 1.02 | 1.02 | 1.02 |
| ROE | 76.3% | 84.9% | 80.5% | 65.4% | 53.8% |
| ROCE | 66.8% | 70.9% | 76.1% | 62.9% | 51.7% |
| Net debt / EBITDA | −0.31x | −0.26x | −0.45x | −0.72x | −1.01x |
| FCF margin | 44.8% | 46.8% | 49.6% | 52.7% | 51.9% |
| Quarterly, US$bn | Q1 FY27A | Q2 FY27A | Q3 FY27E | Q4 FY27E | H1 FY27A | H2 FY27E |
|---|---|---|---|---|---|---|
| Revenue | 81.6 | 96.2 | 108.0 | 123.1 | 177.8 | 231.1 |
| Growth, y/y | +85% | +106% | +89% | +81% | +96% | +85% |
| Data Center revenue | 75.2 | 89.0 | 100.5 | 115.3 | 164.3 | 215.8 |
| Edge computing revenue | 6.4 | 7.2 | 7.5 | 7.8 | 13.6 | 15.3 |
| Gross margin, non-GAAP | — | 75.0% | 74.0% | 71.5% | — | 72.7% |
Source: NVIDIA Q2 FY27 10-Q and results release (26 August 2026); Penmark Research model for Q3-Q4 estimates. Q1 FY27 revenue is derived as H1 less Q2; Data Center revenue for Q2 is reported, for Q1 derived. Q1 FY27 gross margin is not shown.
| KPI detail | Q1 FY27A | Q2 FY27A | Change | FY27E | Source period |
|---|---|---|---|---|---|
| Data Center revenue, $bn | 75.2 | 89.0 | +18% | 380.0 | Q2 FY27 results |
| Hyperscale revenue, $bn | — | 48.7 | +13% q/q (company) | — | Q2 FY27 call |
| AI clouds, industrial and enterprise, $bn | — | 40.3 | +25% q/q (company) | — | Q2 FY27 call |
| Edge computing revenue, $bn | 6.4 | 7.2 | +27% y/y | 28.9 | Q2 FY27 results |
| China including Hong Kong, $bn | — | 7.9 | — | No compute in outlook | Q2 FY27 results |
| Taiwan, $bn | — | 27.0 | 28% of revenue | — | Q2 FY27 results |
| Days sales outstanding | — | 61 | +5 days y/y | 61 | Q2 FY27 call |
| Inventory, $bn | — | 32.0 | — | — | Q2 FY27 call |
| Non-GAAP gross margin | — | 75.0% | — | 73.7% | Q2 FY27 results |
| Non-GAAP operating income, $bn | — | 64.0 | 66.5% margin | 266.3 | Q2 FY27 results |
| Free cash flow, $bn | — | 21.3 | — | 191.5 | Q2 FY27 results |
| Capital returned in Q2, $bn | — | 26.0 | 60% of H1 FCF | 118.2 | Q2 FY27 call |
Source: NVIDIA Q2 FY27 results, 10-Q and earnings call (26 August 2026). Dashes mark figures the company does not disclose on a quarterly basis or that we do not model quarterly. FY27E column is the annual model; it is not the sum of quarterly disclosures.
Source: NVIDIA FY26 10-K, Q2 FY27 10-Q; Penmark Research model. Statements reconcile to the checks in the workbook: the balance sheet balances, ending cash reconciles, net income links through the cash flow and equity, and revenue ties to the build.
Reference. Definitions, the reconciliation from adjusted to reported numbers, the peer set and where every figure comes from
This is an initiation of coverage on NVIDIA Corporation under the products-and-units archetype. The valuation is a ten-year unlevered DCF with mid-period discounting, probability-weighted across three cases run on the same engine, and cross-checked against a forward P/E on FY28E EPS. The balance-sheet statement is shown from FY26A rather than FY25A because the model's opening balance sheet is January 2026.
Reconciliation: adjusted to reported| FY27E, $bn | Value |
|---|---|
| Adjusted net income (excl. stock compensation and investment gains) | 228.6 |
| Less stock-based compensation, after tax | −7.5 |
| Plus investment gains, after tax | +19.6 |
| Other items | +2.1 |
| Reported net income | 242.8 |
| Adjusted EPS, $ | 9.44 |
| Reported diluted EPS, $ | 10.03 |
ROE net income over period-end shareholders' equity. ROA net income over total net assets. ROCE EBIT at a 15% tax rate over total net assets. Net debt debt less cash and marketable securities; equity investments excluded. FCF operating cash flow less capex; stock compensation is a non-cash add-back and is charged in operating expenses. P/E and EV/EBITDA at the 5 November 2026 price of $181.60; EV is market capitalization plus debt less cash and investments.
Gigawatt (GW) a unit of AI factory capacity. Content per gigawatt NVIDIA revenue attributable to one gigawatt of capacity, including compute, networking and software. Hyperscale the largest cloud providers; the company reports their revenue separately. Neocloud a GPU-rental provider funding capacity with third-party debt. ASIC an application-specific accelerator designed by a customer, such as Broadcom's or Marvell's custom silicon. CoWoS the advanced packaging that attaches HBM to a GPU die. DSO days sales outstanding: receivables over revenue per day.
Peer setAMD (closest merchant competitor), Broadcom and Marvell (custom accelerators), TSMC (foundry partner and supplier to every competitor) and Qualcomm (smaller inference competitor). The median forward P/E of 17.0x excludes NVIDIA and uses each company's next fiscal year consensus EPS: AMD CY27, Broadcom FY27, TSM CY27, Qualcomm FY27, Marvell FY28.
Data sources and as-of datesFinancial statements and segment revenue: NVIDIA FY26 Form 10-K, Q2 FY27 Form 10-Q and results release (26 August 2026). Guidance: Q2 FY27 earnings call, 26 August 2026. Operating KPIs: company results releases and the Q2 FY27 call. Share prices: closing prices 5 November 2026. Short interest: FINRA settlement 15 October 2026. Consensus revenue and EPS: Yahoo Finance analyst estimates, September-October 2026 (45 estimates for Q3 FY27 revenue); EBITDA consensus was not available from any source used and is not shown. Peer multiples: TIKR, StockTitan and company-reported figures, October-November 2026. Ratings and targets: MarketBeat (53 ratings, October 2026) and TIKR (2 November 2026). Insider transactions: Form 4 filings to 9 October 2026. Model date: 6 November 2026; price date 5 November 2026.
ChecksThe workbook reports all checks clear: the balance sheet balances in every year, ending cash reconciles to the cash flow, net income links through equity, the revenue build ties to reported revenue, scenario ordering is bear < base < bull, sensitivity-grid centres equal the headline values, and no forecast or valuation cell contains a typed number.
| Exhibits | Section | Source |
|---|---|---|
| 1-7 | The call and the model | NVIDIA 10-K/10-Q; Penmark model; yfinance prices |
| 8-11 | Ratios, consensus, peers | Penmark model; Yahoo Finance consensus; TIKR; StockTitan |
| 12-16 | Catalysts, risks, valuation context, monitoring | Penmark model; Q2 FY27 call; yfinance |
| 17-28 | Thesis in charts and key debates | Penmark model; Yahoo Finance consensus; company disclosures |
| 29-31 | Business model, flows, driver tree | Q2 FY27 10-Q and call; Penmark model |
| 32-37 | Market size, demand pools, unit economics | Q2 FY27 call; company filings; Penmark model |
| 38-43 | Competitive position | NVIDIA and Broadcom filings; TIKR; Penmark assessment |
| 44-47 | Capital allocation and insider activity | FY22-FY26 10-Ks; Form 4 filings; results releases |
| 48-53 | What moves the stock | yfinance prices; 106-week regression; Q2 FY27 transcript |
| 54-61 | Thesis chapters | Q2 FY27 call; Penmark model; consensus |
| 62-65 | Track record and quality of numbers | 10-K filings; Penmark model |
| 66-71 | Forecasts, guidance and sensitivities | FY26 10-K; Q2 FY27 call; Penmark model |
| 72-77 | Valuation | Penmark model: DCF, scenarios, sensitivities; peer multiples |
| 78-82 | Catalysts and risks | Company calendar; consensus; Penmark model |
| 83-86 | Ownership and trading | FINRA; MarketBeat; TIKR; yfinance volume |
| 87-91 | Financial model and reference | FY26 10-K; Q2 FY27 10-Q; Penmark model |
Source: Penmark Research. Full source lines appear beneath each exhibit; consensus figures are as of September-October 2026 and prices as of 5 November 2026.
Analyst certification and important disclosures
Ratings
Ratings reflect the expected total return to our 12-month price target, including dividends, weighted across bull, base and bear scenarios.
| Buy | Expected total return above +10% |
| Hold | Expected total return between −10% and +10% |
| Sell | Expected total return below −10% |
We review ratings at each set of results and whenever the facts change materially. A share price can move outside its band before the rating changes, and targets may not be reached.
Analyst certification
The Penmark Team certifies that (1) the views in this report accurately reflect its views about the subject company and its securities, and (2) no part of its compensation was, is or will be directly or indirectly related to the specific ratings, price targets or views in this report.
Contact
Penmark Research
Questions about our research:
research@penmarkresearch.com
All our research and coverage:
penmarkresearch.com
Basis of preparation
This report uses information publicly available up to its publication date. Share prices are closing prices on the trading day before the publication date, unless stated otherwise. Forecasts, valuations and price targets are Penmark Research’s own, and forecast periods are marked E. Sources are shown beneath each exhibit.
Valuation and price targets
Price targets are based on the valuation methods set out in each report and are weighted across bull, base and bear scenarios. The main risks to each rating and price target are described in the report. Penmark Research updates its coverage after each set of results but is not obliged to update a report at any other time.
Independence and conflicts of interest
Penmark Research is independent. It does not provide investment banking, brokerage, market-making or asset management services, does not trade securities, and does not accept payment from any company for research coverage. Companies do not see or review Penmark Research reports before publication.
No member of the Penmark Team holds a long or short position, directly or through derivatives, in the securities of any company Penmark Research covers.
Regulatory status
Penmark Research publishes general, impersonal investment research on a regular basis. It is not registered as an investment adviser with the US Securities and Exchange Commission or any state securities authority, is not a registered broker-dealer and is not a member of FINRA. Penmark Research does not give advice tailored to any person’s circumstances.
Not investment advice or an offer
This report is for information only. It is not personalised investment advice and is not an offer or solicitation to buy or sell any security or other financial instrument. It does not take into account any reader’s objectives, financial situation or needs. Readers should make their own investment decisions and seek independent professional advice where appropriate.
Forward-looking statements and risk
Forecasts, valuations and price targets are estimates based on assumptions that may prove wrong, and they may not be achieved. Opinions are current as of the publication date and may change without notice. Investing in equities involves risk, including the loss of capital. Share prices can be volatile, and past performance is not a reliable indicator of future results.
Data and sources
Company financials, operating data and filings come from company filings, earnings releases and earnings calls. Market data and any consensus estimates come from third-party sources named beneath the relevant exhibit; consensus figures from aggregators are unofficial. Penmark Research does not use other brokers’ research. Third-party data belong to their providers and may not be extracted from this report.
Accuracy and liability
Information is taken from sources believed to be reliable, but Penmark Research makes no representation or warranty as to its accuracy, completeness or timeliness. To the fullest extent permitted by law, Penmark Research accepts no liability for any loss arising from use of this report. Descriptions of litigation and regulatory matters reflect allegations and public records, not findings, unless stated. If a published report is corrected, the corrected version carries an “Updated” date at penmarkresearch.com.
Distribution and access
Penmark Research reports are available to registered users of penmarkresearch.com. Downloads and programmatic access are governed by the Penmark Research subscription terms. This report is not directed at, or intended for, any person in Singapore or India, and access from those countries is blocked. It is not intended for distribution to, or use by, any person in any jurisdiction where that would be contrary to law or regulation or would subject Penmark Research to registration or licensing requirements there.
Copyright and permitted use
© Penmark Research. All rights reserved. Penmark Research is a trade name of Brightriver Labs. Readers may share a link to this report but may not copy, reproduce, redistribute, sell or use its content, including for text and data mining, without written consent. Contact: research@penmarkresearch.com.