The next two years are largely visible and only partly priced. The risk has moved from demand to who pays for it.
Revenue nearly quadrupled in nine quarters. The last bar is management’s Q3 guide, not a result. Sources: NVIDIA CFO commentaries and 10-Qs.
Quarter to July 26, 2026. Total $96.2B, up 106%. Competitive position
The Street is far more bullish than I am: 59 of 62 analysts rate the stock a buy, and the median target is $315 (StockAnalysis, October 9, 2026). My $240 is lower for three reasons. None of them is a weaker view of demand.
NVIDIA’s measured beta against the S&P 500 is 1.91 over two years. Even after the standard Blume adjustment to 1.61, its cost of equity is 13.2%, high for a company this large and cash-rich. Each point off the WACC adds about $15 a share to my base case.
My base case follows the Street through fiscal 2029, then slows growth to 4-5% and lets operating margin drift from 66% to about 60% by fiscal 2035. That is roughly NVIDIA’s fiscal 2026 level, before memory costs and custom chips push back.
I haircut the equity stakes (85% of listed, 50% of private value) and deduct expected losses on guarantees and capacity backstops: $8 billion in the base case and $35 billion in the bear case. Most models add the stakes at book value and ignore the guarantees.
NVIDIA designs the accelerated computers that train and run AI models: GPUs (Blackwell now, Vera Rubin from August 2026), the CPUs, networking (NVLink, InfiniBand, Spectrum-X Ethernet) and software (CUDA and its libraries) that turn chips into full racks. Taiwan Semiconductor makes the chips; NVIDIA owns no fabs. Since Q1 fiscal 2027 it reports two market platforms: Data Center, split into Hyperscale and AI Clouds, Industrial & Enterprise (ACIE), and Edge Computing (PCs, game consoles, workstations, robotics and cars).
Data Center was 93% of revenue in Q2. Networking was the fastest-growing part of it: $31.4 billion in fiscal 2026, up 142%, and about 20% of Data Center revenue by Q1 fiscal 2027 (NVIDIA stopped splitting it out after that quarter). The fiscal 2026 gross margin of 71.1% includes a $4.5 billion charge for H20 chips that export rules stopped it selling to China.
Reported results, USD billions. Fiscal years end in late January.
| Line item | FY2024 | FY2025 | FY2026 | Q1 FY2027 | Q2 FY2027 |
|---|---|---|---|---|---|
| Data Center revenue | 47.53 | 115.19 | 193.74 | 75.25 | 89.02 |
| Edge Computing and other | 13.40 | 15.31 | 22.20 | 6.37 | 7.20 |
| Total revenue | 60.92 | 130.50 | 215.94 | 81.62 | 96.22 |
| Gross margin | 72.7% | 75.0% | 71.1% | 74.9% | 75.0% |
| Operating expenses (incl. stock comp) | 11.33 | 16.41 | 23.08 | 7.62 | 8.41 |
| Operating income | 32.97 | 81.45 | 130.39 | 53.54 | 63.73 |
| Net income (incl. investment gains) | 29.76 | 72.88 | 120.07 | 58.32 | 59.69 |
| Free cash flow | 26.95 | 60.72 | 96.58 | 48.55 | 21.34 |
| Share repurchases | 9.53 | 33.71 | 40.09 | 19.31 | 19.73 |
| Diluted shares (millions) | 24,940 | 24,804 | 24,514 | 24,391 | 24,285 |
Sources: NVIDIA FY2026 10-K (filed Feb 25, 2026); Q1 FY2027 CFO commentary (May 20, 2026); Q2 FY2027 10-Q and CFO commentary (Aug 26, 2026). For FY2024-FY2026, “Edge Computing and other” is total revenue less Data Center (Gaming, Professional Visualization, Automotive, OEM). Free cash flow is NVIDIA’s definition: operating cash flow less capital spending and principal payments on equipment.
For most of the boom, NVIDIA’s customers were the richest companies in the world, paying cash. That is changing. The fastest-growing buyers are AI clouds (neoclouds) and AI labs that are, in NVIDIA’s words, “growing faster than their balance sheets and long-term credit profiles can support”. NVIDIA has responded by lending its own balance sheet. This section is the main reason my fair value is below the Street.
| Exposure, July 26, 2026 | USD billions | What it is |
|---|---|---|
| Equity investments held | 99.0 | Listed stakes $47.7B (of which $36.9B locked up) and private stakes $51.2B, mostly AI model makers and infrastructure financiers. CFO: “nearly $50 billion in the frontier AI labs”. |
| Further equity commitments | 25.0 | Mostly due in the rest of fiscal 2027. |
| SB Energy lease guarantees for OpenAI | 105.0 | Cap on residual value guarantees for about 4.25 GW at the PORTS campus in Ohio, 20-year leases to OpenAI, phases from fiscal 2029. Paid only if OpenAI defaults; ends if OpenAI earns a satisfactory credit rating. |
| Other land, power and shell guarantees | 3.5 | For AI cloud partners’ data center leases. |
| AI cloud capacity backstops | 36.0 | Six-year cloud agreements: NVIDIA buys capacity the AI clouds cannot sell to others, and shares revenue if they can. |
| Third-party data center leases | 20.0 | Leases NVIDIA signed and expects to reassign to others. |
| Supply and capacity commitments | 279.0 | Up from $119B one quarter earlier, mostly memory, spread over fiscal 2027-2029. |
My reading: real demand, partly self-financed. The guarantees make sense commercially: the PORTS site will host only NVIDIA systems, and the CFO says each hardware generation there could be worth $150-200 billion of revenue. But they tie NVIDIA’s results to the credit of a few AI labs. If those labs cannot raise money, NVIDIA loses the sale, marks down the stake and may pay on the guarantee, all at once. That is why my bear case is a demand bust with a $35 billion financing loss, not just slower growth.
Sources: Q2 FY2027 10-Q, Notes 5, 6 and 10, MD&A and Item 1A (Aug 26, 2026); NVIDIA 8-K on the SB Energy residual value guaranties (Aug 17, 2026); Q2 FY2027 earnings call (Aug 26, 2026, third-party transcript). In August 2026 NVIDIA also signed memoranda with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR for financing platforms targeting more than $500 billion of outside capital; they are not definitive agreements.
NVIDIA expects the top five hyperscalers to spend nearly $800 billion on capital projects in 2026 and $1.3 trillion in 2027. FactSet notes that most of them now spend more than their operating cash flow; Alphabet raised $84.75 billion of equity in June 2026 partly to fund it. Spending can keep growing only while AI revenue grows to pay for it.
Vera Rubin began shipping in August and should be about 20% of Q3 Data Center revenue. Each generation raises NVIDIA’s content per gigawatt. Broadcom’s AI revenue grew 221% last quarter. The split between general-purpose GPUs and custom chips in inference is the biggest long-run swing factor.
Memory (HBM and DRAM) is the largest bill-of-materials item NVIDIA does not control. Micron’s revenue is expected to roughly double this year. NVIDIA has raised its own prices from Q1 fiscal 2028; whether they hold decides if margin recovers to 72-73% or keeps sliding.
Receivables rose $22 billion in Q2 as large customers got up to a year to pay. NVIDIA now backstops capacity, guarantees leases and invests in its customers. As long as AI labs can raise capital this is a growth tool. If they cannot, it becomes a loss.
The model is driver-based: Data Center and Edge Computing revenue by fiscal year, gross margin, operating expenses in dollars (stock-based compensation stays in, as NVIDIA’s own non-GAAP figures now do), then cash tax at 17%, capital spending and working capital at 15% of each year’s revenue growth. Fiscal 2028 and 2029 in the base case match the Street; the years after that are mine.
Base case, USD billions
Consensus: StockAnalysis (FY2027 $411.7B, FY2028 $692.2B, updated Oct 9, 2026); MarketScreener (FY2029 $903.2B, Oct 16, 2026). Guidance: Q2 FY2027 CFO commentary and call. Scenario paths are the author’s.
NVIDIA reported $118 billion of net income in the first half of fiscal 2027 but generated $70 billion of free cash flow. The gap is worth understanding before choosing a method, because it decides whether earnings or cash flow is the right thing to value.
Reduces cash Investment in growth Adds cash Result
Partly. $23.7 billion of pre-tax income was gains on equity stakes, mostly unrealized, about a fifth of first-half net income. NVIDIA’s own non-GAAP figures exclude them; so does my model.
PartlyMostly. Receivables grew $24.6 billion on longer payment terms for investment-grade buyers, and inventory $10.2 billion ahead of Vera Rubin. These reverse if growth slows, but longer terms are now policy, so I hold working capital at 15% of each year’s revenue growth.
MostlyNo. Capital spending was $4.4 billion, 2.5% of revenue. The investment NVIDIA makes is in its customers, through stakes and guarantees, which is why I value those separately.
NoMy reading: high-quality cash flow with a financing layer on top. The operating business converts well and needs little capital. The equity gains and the guarantees are a separate, riskier business attached to it. I value the first on free cash flow and the second at a discount.
Source: Q2 FY2027 10-Q, condensed consolidated statement of cash flows, six months to July 26, 2026. “Stock comp, D&A and other items” is the balancing figure: stock-based compensation $4.0B, depreciation and amortization $2.1B, deferred taxes, payables, accruals and other working-capital lines.
NVIDIA is profitable, so an earnings multiple is possible. But its profits are near a cyclical peak, its margins are guided down, and a fifth of its first-half net income was investment gains. A multiple of this year’s earnings would bake all of that in. A cash-flow model built from the drivers lets each of those questions be answered separately, under three scenarios.
Data Center and Edge revenue, gross margin and operating costs, fiscal 2028 to 2035.
Why: these are what each quarterly report and each hyperscaler capex update change.
Operating profit after 17% cash tax, plus depreciation, less capital spending and working capital. Stock compensation stays a cost.
Why: earnings include investment gains and miss the receivables build.
Cost of equity 13.15% from a Blume-adjusted beta of 1.61. WACC of 13.1%.
Why: the risk charge comes from how the stock trades, adjusted only by the standard rule.
13 times fiscal 2035 EBITDA, valued at the end of fiscal 2035, cross-checked against 3% perpetual growth.
Why: below what mature chip companies trade at today, because 2035 is far away and the industry is cyclical.
Net cash, plus stakes after haircuts, less expected guarantee and backstop losses for each scenario.
Why: they are real assets and real liabilities, but neither is worth book value.
Bullish if the expected return over 3-5 years beats the cost of equity by 2 points a year; Bearish if it trails by 2.
Why: the call follows the numbers, and anyone can check it.
| Method | How I used it | Reason |
|---|---|---|
| Driver-based DCF, exit multiple, three scenarios | Primary | Ties value to the variables that move the stock and separates the operating business from the financing layer. |
| DCF with perpetual growth | Cross-check | Gives $203.27 for the base case, and $191 probability-weighted. At a 13.1% discount rate, 3% growth is a harsh assumption for a business still growing 4-5% in 2035. |
| P/E and EV / EBITDA against peers | Cross-check | NVIDIA trades at 13.6x the Street’s fiscal 2028 earnings, below Broadcom and AMD, but at peak margins. |
| Reverse DCF | Cross-check | Solves for what today’s price assumes about Data Center revenue. |
| Sum of the parts | Not used | NVIDIA does not report costs by platform. Investments are valued separately instead. |
| Monte Carlo simulation | Not used | Needs probability distributions nobody can observe. Three explicit stories are easier to check. |
I discount unlevered free cash flow for the second half of fiscal 2027 and for fiscal 2028 to 2035 at a 13.1% WACC, using mid-year discounting, and capitalize fiscal 2035 EBITDA at 13x as a value at the end of that year. Debt is under 1% of capital, so WACC is essentially the cost of equity.
| Input | Value | Source or reason |
|---|---|---|
| Risk-free rate | 5.10% | US 10-year Treasury yield, October 16, 2026. |
| Beta, measured | 1.91 | Two years of daily returns against IVV (501 days). 1.85 against SPY; 2.22 over five years (StockAnalysis). |
| Beta, used | 1.61 | Blume adjustment: two-thirds measured, one-third 1.0. The past two years were an extreme period for this stock; betas tend to drift toward 1. Damodaran’s semiconductor industry beta is 1.52 (January 2026), close to the adjusted figure. |
| Equity risk premium | 5.0% | House assumption, the same as my Rivian report. Damodaran’s implied premium was 4.2% on October 1, 2026; I did not use the lower figure. |
| Cost of equity | 13.15% | 5.10% + 1.61 × 5.0%. |
| WACC | 13.1% | 99.4% equity at market value; debt at 5.0% before 17% tax. |
The beta adjustment matters, so here is the call under each reasonable choice:
| Choice | Beta | ERP | Cost of equity | Fair value | Call | Bullish below |
|---|
The ERP column is my house 5.0% or Damodaran’s implied 4.2%. With the raw beta the call is Neutral; with the adjusted or industry beta it is Bullish. I publish the adjusted beta because it is the standard forward-looking estimate, and I show the raw result so readers can disagree.
| Company | EV / sales, next FY | Fwd P/E | Growth, next FY |
|---|---|---|---|
| Broadcom | 9.3x | 19.0x | 64% |
| AMD | 9.6x | 50.6x | 75% |
| Marvell | 10.2x | 40.5x | 63% |
| TSMC | n/a | 18.8x | 38% |
| Arm | 29.1x | 102.6x | 36% |
| Micron | 3.4x | 5.3x | 16% |
| NVIDIA at $214.13 | 7.4x | 17.7x | 68% |
Sources: StockAnalysis statistics and forecast pages, Oct 16, 2026 (estimates updated Sep 18 to Oct 16). Forward P/E is on the next twelve months. Prices moved sharply that day, so treat multiples as approximate. Multiples are context, not inputs.
The exit multiple of 13x sits below where mature chip companies trade today: Qualcomm and TSMC at about 16x trailing EBITDA, Texas Instruments and Analog Devices near 28-30x. As a check, 13x implies 6.7% perpetual growth at my discount rate, which is generous; the same base case on a 3% perpetuity is worth $203. That gap is the honest measure of how much of NVIDIA’s value rests on the market still paying a growth multiple in 2035.
Holding everything else in my base case fixed, $214.13 implies Data Center revenue about 16% below my path in every year.
Fair value needs to be at least $233.74 for the rule to read Bullish at $214.13, so the cushion is about $6.60 a share, under 3%. The call turns Neutral with any one of these: an exit multiple below 12.4x, scenario revenue more than about 3% below my paths, a discount rate above 13.6%, gross margin 2 points lower in every year, or scenario weights of 30/50/20 (bear/base/bull) as in my Rivian report. On the perpetual-growth method alone, the rule would read Bearish. I rate it Bullish because each published assumption is defensible on its own, not because the result is comfortable.
Each slider applies to all three scenarios. The scenario values and the probability-weighted fair value update instantly; the published call does not change.
NVIDIA is run by its co-founder, who still owns about 3.6% of it, with a long-serving finance chief. The leadership is stable; the open questions are succession and how far the company should go in financing its customers.
Co-founder (1993), President and Chief Executive Officer
What it means: his wealth is overwhelmingly in the stock, which aligns him with long-term value. It also means there is no obvious successor of similar standing.
Executive Vice President and Chief Financial Officer
What it means: guidance has been conservative: NVIDIA beat its revenue outlook by about 5% in each of the last two quarters.
Executive Vice President, Worldwide Field Operations, from August 24, 2026
What it means: NVIDIA hired an enterprise software seller to run sales as the customer base moves beyond the hyperscalers.
| Holder | Shares | Stake | Note |
|---|---|---|---|
| BlackRock | 1,941.9M | 8.1% | 13F, June 30, 2026. Mostly index funds. |
| Vanguard (two reporting entities) | 2,058.6M | 8.5% | Vanguard Capital Management 1,540.8M and Vanguard Portfolio Management 517.8M, 13F, June 30, 2026. |
| FMR (Fidelity) | 1,026.1M | 4.3% | 13F, June 30, 2026. |
| State Street | 1,009.8M | 4.2% | 13F, June 30, 2026. |
| Jensen Huang | 870.6M | 3.6% | Proxy, March 23, 2026, held mostly through family trusts. |
| All directors and officers | 957.3M | 4.0% | Proxy, March 23, 2026. |
Stakes use 24.1 billion shares outstanding (August 21, 2026), so they differ slightly from the proxy’s percentages. Director Mark Stevens sold about 6.9 million shares between December 2025 and September 2026, about 4.2 million of them between August 31 and September 18; no insider bought shares in the open market in that period.
Sources: NVIDIA 2026 proxy statement (SEC DEF 14A, May 12, 2026); 8-K on the Worldwide Field Operations change (SEC, Jul 2, 2026); Form 4 filings (SEC); trading plans in the FY2026 10-K and FY2027 10-Qs; 13F holdings via Nasdaq, retrieved Oct 16, 2026.
| When | Event | What I am watching |
|---|---|---|
| Late Oct 2026 | Hyperscaler Q3 results | 2027 capital spending plans from Microsoft, Alphabet, Amazon, Meta and Oracle against NVIDIA’s $1.3 trillion top-five estimate. |
| Nov 17, 2026 | Q3 fiscal 2027 results | Revenue against the $108B guide, the Q4 margin trough, receivables and cash flow, and whether the fiscal 2028 view of about 70% growth holds. |
| Q4 FY2027 | Vera Rubin volume ramp; Groq 3 LPX shipments | Rubin’s share of Data Center and its gross margin; early inference demand for the Groq-based LPX. |
| Feb 2027 | Q4 results and fiscal 2028 outlook | Whether price increases from Q1 fiscal 2028 lift gross margin back to 72-73%. |
| Mar 2027 | GTC developer conference | Roadmap after Rubin and an update to the $1 trillion Blackwell and Rubin demand figure given in March 2026. |
| First half 2027 | Hugging Face acquisition closes | Regulatory approval and whether the platform stays open to other chips, as NVIDIA committed. |
| Fiscal 2029 | First PORTS campus phase in service | The SB Energy guarantees start to apply; OpenAI’s credit rating decides how long they last. |
Q3 date: Robinhood earnings calendar, marked as company-verified. Other dates are company statements or typical timing.
| Risk | Trigger I would act on | Today |
|---|---|---|
| Customer financing | Days sales outstanding above 70, or a guarantee or backstop payment | 60 days; none paid |
| Earnings quality | Operating cash flow below 60% of net income for two quarters | 63% in H1 FY2027 |
| Customer concentration | One direct customer above 20% of revenue, or a top-five capex cut | 16% in Q2 |
| Custom chips | Broadcom AI revenue above a quarter of NVIDIA Data Center | About 19% |
| Memory and supply | Gross margin below 71%, or supply commitments rising faster than revenue | 74% guided for Q3; $279B commitments |
| Export controls | New US limits on networking or Rubin-class chips | China Data Center assumed at zero |
| Power and sites | Neocloud capacity well short of the 8 GW NVIDIA expects | About 3 GW at end-2025 |
Here are my estimates for Q3 fiscal 2027, so the call can be checked against the result.
The most important signal will not be revenue. It will be operating cash flow against net income and any change in the guarantees and capacity backstops.
Download the Excel model. It holds the inputs, three scenario blocks, the valuation with the 3-5 year rule, a 30-cell sensitivity grid and reported history, all as live formulas. Every hardcoded input carries a source note. It reproduces this page’s numbers exactly.
This is personal research by Phat Ngo, prepared as a work sample. It is not investment advice or a recommendation to buy or sell any security. Disclosure: the portfolio shown on this site holds NVDA at a 10% target weight and AMD, a direct competitor, at 9% (from the October 16, 2026 close; 12% and 10% before), so the author has a financial interest in this company and its peers. Forecasts, scenario probabilities and the call are the author’s judgments and may be wrong. Figures are as reported in public filings and news coverage on the dates cited.