NVIDIA designs the chips and the full systems that train and run AI models. It sells GPUs (Hopper, then Blackwell, now Vera Rubin, which went into full production in August 2026), entire liquid-cooled racks (GB300 / VR200 NVL systems), the NVLink and Spectrum-X / InfiniBand networking that ties tens of thousands of GPUs into one "AI factory", and the CUDA software stack that most AI code is written for. TSMC manufactures the chips, SK hynix / Samsung / Micron supply the HBM memory, and ODMs assemble the racks. NVIDIA owns no fabs. It is the bottleneck supplier to the AI build-out: hyperscalers (Microsoft, Google, Amazon, Meta, Oracle), neoclouds (CoreWeave, Nebius), AI labs (OpenAI, Anthropic, xAI), sovereigns and enterprises.
Mix. FY26 (FYE Jan-26) revenue was $215.9B: data center $193.7B (compute $162.4B, networking $31.4B), gaming $16.0B, pro visualization $3.2B, automotive $2.3B, OEM/other $0.6B. Q2 FY27 (Jul-26) was $96.2B under the new reporting split: data center $89.0B, of which hyperscale $48.7B and AI clouds/industrial/enterprise $40.3B, plus edge computing $7.2B. Billing geography for Q2: US $60.1B, Taiwan $27.0B, China/HK $7.9B (billing location; Q3 guidance assumes zero China data-center compute), other $1.3B.
How it makes money: 75% gross margin and ~66% GAAP operating margin, on a business with almost no fixed assets. Capex is $7.4B TTM against $303B of revenue. Key sensitivities: the AI-infrastructure capex cycle (hyperscaler capex ~$870B in 2026 → ~$1.3T in 2027E per S&P), customer financing capacity, custom-silicon share (TPU, Trainium, Broadcom/Marvell ASICs, AMD), HBM/memory pricing, TSMC CoWoS capacity, and US–China export policy.
The franchise is extraordinary, and the numbers are filed, not hoped for. TTM revenue (to Jul-26) was $303.0B. GAAP EBIT was $197.6B (65.2%), or $210.6B (69.5%) with R&D capitalized. That is an R&D-adjusted after-tax ROIC of about 105% on $168B of invested capital. Q2 FY27 revenue was $96.2B (+106% y/y) and the Q3 guide is $108B ±2%. Management gave its first full-year-forward guide: ~70% revenue growth in FY28 (~$680B), which it calls supply-constrained. Gross margin troughs at 71–72% in Q4 on memory costs, then recovers to 72–73% in FY28 on price increases. On consensus, FY28 EPS is about $15.5, so the stock trades at ~14× next year's earnings. That is not a bubble multiple.
Why the DCF still lands below the price. The base case takes the FY28 guide at face value. After that it grows +15% and +7%, has one flat digestion year in FY31, and runs ~3.5–4% thereafter, reaching $1.04T in FY37. The R&D-adjusted margin glides from 67% to 50% by FY35 as TPUs, Trainium, Broadcom ASICs and AMD take inference share and memory costs bite. Even 50% is still higher than any megacap today. Discounted at a 10.8% WACC (bottom-up β 1.30, Rf 5.12%, ERP 4.40%) that glides to 9.9%, that is $193.70. The price is 12% above it. Terminal value is 47% of operating value (well under the 70% flag), because most of the value sits in the next decade of cash flows, not in perpetuity.
Reverse DCF: the gap is one assumption, not a fantasy. To get to $217.15, you need any one of these: a 58.5% steady-state margin instead of 50% (keeping roughly today's pricing power forever); revenue 14% higher every year from FY29 (FY31 ~$955B, FY37 ~$1.19T); or a WACC 78bp lower (~10.0%). Compare the rest of semis on the same framework: KLAC +193%, AMAT ~+197%, LRCX ~+245% above base. NVDA is the cheapest AI-semi name the project has run. The market isn't pricing a miracle here. It's pricing durability: that $380B+ of after-tax operating profit in FY28 doesn't roll over.
The bear case is the cycle, not the competition. Bear ($100.39) is a Cisco-2001 path: FY28 at $640B, then −15% and −15% as the capex boom outruns its funding, with margin falling to 40%. S&P projects that five of the six biggest AI spenders will be free-cash-flow negative in 2027, and the 10-year just hit a two-decade high near 5.3%. Bull ($317.31, close to the $334 Street mean target) is a supercycle: +30%/+20% to $1.2T by FY31 with a 60% margin. The probability-weighted value (25/50/25) is $201.28. The skew is unfavorable at this price: bull is +46%, bear is −54%.
Earnings-quality flags to watch. In H1 FY27, operating cash flow was $74.4B. Of that, $42.4B went into equity stakes (largely AI labs and customers; ~$50B invested in frontier labs to date) and $39.0B into buybacks. NVIDIA also issued $25B of notes in June. H1 non-operating gains of $24.1B, mostly mark-ups on those stakes, flatter GAAP net income, so this model values off EBIT and ignores them. The company is also reported to be weighing a $250B financing guarantee for an OpenAI site, alongside $500B+ financing platforms with Apollo, Blackstone, KKR and others. Vendor financing is how capex cycles extend their life, and how they end badly. Then there are $279B of supply and capacity commitments (~$366B in total) plus $31.6B of inventory, mostly non-cancellable. In a down-cycle, those turn into write-downs.
Call: HOLD what you own. No new money above ~$205 (base +5%). Accumulate at ≤$185 and go full size at ≤$165 (about a 15% margin of safety, roughly equal to base value if FY29+ revenue comes in 20% light). Size it as a cyclical, not a bond-like compounder. Trim above $250, where the price starts to discount the bull path. Get paid to wait: sell the Dec-18 $185 put (~$2.97 mark, Δ −0.14, IV 42%) for an effective entry of $182.03, about 6% below base. Or, for defined risk, the 185/175 bull put spread (~$1.13 credit vs $8.87 risk, 12.7%). Both short legs are within your Δ ≤ 0.20 rule. Note: the window includes Q3 FY27 earnings on Nov 17 (after close; EPS est $2.34 vs a ~$108B guide), and the elevated IV is what pays you for that.
| HIGH | AI-capex cycle duration. ~92% of revenue rides data-center capex. Top-6 hyperscaler capex runs ~$470B (2025) → ~$870B (2026E) → ~$1.3T (2027E), and FY28's ~$680B equals ~half the 2027 top-6 figure. A ±20% shift in the revenue level from FY29 moves value from $160 to $227. Long the cycle extending; short a 2028 digestion. |
| HIGH | Customer financing / circularity. The $94B equity portfolio ($42.8B public, $51.2B private) includes stakes in the labs that buy the chips. Add ~$50B invested in frontier labs, a reported $250B OpenAI site guarantee under consideration, and $500B+ financing platforms. NVDA CDS posted its biggest intraday jump on that report (7/27/26). The non-hyperscale half of data center (45%) has a weaker credit profile. Long customer credit. |
| HIGH | Supply-commitment "inventory". $279B of supply/capacity commitments (up from $119B a quarter earlier, mainly Rubin memory) plus $31.6B of on-hand inventory. Most of it is non-cancellable. In a demand air-pocket it becomes write-downs and gross-margin hits (the FY23 precedent, at a fraction of today's scale). Long volume. |
| MED | Custom silicon / pricing power. ASIC servers are ~28% of 2026 AI-server shipments (TrendForce). Anthropic has committed to up to 1M Google TPUs, OpenAI–Broadcom is 10 GW, and Broadcom sees line of sight to $100B+ of AI revenue in 2027. Inference (~2/3 of compute) is where ASICs win on TCO. Steady-state margin of 42% vs 58% gives $172 vs $216. Short share erosion. |
| MED | Memory / input costs & supply chain. "Extreme" HBM pricing takes gross margin from 75% to a 71–72% Q4 trough. Packaging also bites: Rubin Ultra reverted from four dies to two after CoWoS-L substrate warpage, NVDA holds ~60% of CoWoS allocation, and everything is single-sourced at TSMC Taiwan. Short memory prices; long TSMC execution. |
| MED | Discount rate. The 10Y is at 5.12% after touching ~5.3%. WACC ±100bp gives $171 / $225. Equity duration is moderate: 53% of value comes from FY27–37 cash flows. Short rates. |
| LOW | China / export policy. Guidance carries zero China data-center compute (H20/H200 shipments were <1% of DC revenue in Q2), so policy is upside optionality, not a hole in the numbers. Tail risk is Taiwan. Long easing. |
| LOW | Balance sheet. $56.6B cash and debt securities plus $94B of equity stakes against $33.4B debt and $5.5B leases. Q2 interest expense was $227M. The risk is capital allocation (stakes in customers, buybacks at 14–23× earnings), not solvency. |
| FY29+ revenue level ↓ / steady-state R&D-adj margin → | 35% | 40% | 45% | 50% | 55% | 60% | 65% |
| 0.6× base (FY31 $502B) | $101 | $109 | $118 | $126 | $134 | $142 | $151 |
| 0.8× base (FY31 $670B) | $127 | $138 | $149 | $160 | $171 | $182 | $193 |
| 1.0× base (FY31 $837B) | $152 | $166 | $180 | $194 | $207 | $221 | $235 |
| 1.2× base (FY31 $1,004B) | $177 | $194 | $210 | $227 | $243 | $260 | $276 |
| 1.4× base (FY31 $1,172B) | $202 | $221 | $241 | $260 | $279 | $299 | $318 |
| 1.6× base (FY31 $1,339B) | $227 | $249 | $271 | $293 | $315 | $337 | $359 |
| WACC (HG → stable) ↓ / g → | 2.5% | 3.0% | 3.5% | 4.0% |
| 8.8% → 7.9% | $244 | $256 | $270 | $288 |
| 9.8% → 8.9% | $209 | $216 | $225 | $235 |
| 10.8% → 9.9% | $184 | $188 | $194 | $200 |
| 11.8% → 10.9% | $164 | $167 | $171 | $175 |
| 12.8% → 11.9% | $149 | $151 | $153 | $156 |
| FY | Revenue | GAAP EBIT | Margin | R&D | SBC | D&A | Capex |
| FY2022 | $26,914 | $10,041 | 37.3% | $5,268 | — | $1,174 | $976 |
| FY2023 | $26,974 | $4,224 | 15.7% | $7,339 | — | $1,544 | $1,833 |
| FY2024 | $60,922 | $32,972 | 54.1% | $8,675 | $3,549 | $1,508 | $1,069 |
| FY2025 | $130,497 | $81,453 | 62.4% | $12,914 | $4,737 | $1,864 | $3,236 |
| FY2026 | $215,938 | $130,387 | 60.4% | $18,497 | $6,386 | $2,843 | $6,042 |
| TTM Jul-26 | $302,970 | $197,579 | 65.2% | $23,592 | $7,241 | $3,687 | $7,354 |
| 5-yr avg GAAP EBIT (FY22–26), the ginzu normalizer · vs TTM | $51,815 · vs $197,579 |
| Research asset (5-yr SL: TTM R&D + 80/60/40/20% of FY26/25/24/23) | $51,076 |
| R&D amortization (FY22–26 ÷ 5) | $10,539 |
| TTM R&D $23,592 − amortization = EBIT add-back | +$13,053 |
| TTM R&D-adjusted EBIT (starting point) | $210,632 (69.5%) |
| Invested capital: book (equity $228,984 + debt $33,366 + leases $5,494 − cash/debt sec $56,586 − equity stakes $93,940) + research asset | $117,318 + $51,076 = $168,394 |
| R&D-adjusted after-tax ROIC (16% tax) | 105% |
| H1 FY27 cash: OCF $74,421 − capex $4,434 = FCF · equity stakes bought · buybacks | $69,987 · $42,404 · $39,044 |
| Non-cash working capital (AR $63,059 + inv $31,575 + prepaid $3,409 − AP $15,059 − accrued $26,960) | $56,024 (18.5% of TTM rev; +$24,139 in H1) |
| FY | Revenue | g | Margin | EBIT | Tax | NOPAT | Reinvest | FCFF | WACC | PV |
| H2 FY27 (stub) | $222,163 | — | 67.0% | $148,849 | 16.0% | $125,033 | $36,812 | $88,221 | 10.83% | $85,983 |
| FY2028 | $680,000 | 70% | 67.0% | $455,600 | 16.0% | $382,704 | $112,000 | $270,704 | 10.83% | $244,259 |
| FY2029 | $782,000 | 15.0% | 64.6% | $504,949 | 16.0% | $424,157 | $40,800 | $383,357 | 10.83% | $312,115 |
| FY2030 | $837,000 | 7.0% | 62.1% | $520,136 | 16.0% | $436,914 | $22,000 | $414,914 | 10.83% | $304,807 |
| FY2031 | $837,000 | 0.0% | 59.7% | $499,809 | 16.0% | $419,839 | $0 | $419,839 | 10.83% | $278,295 |
| FY2032 | $870,000 | 3.9% | 57.3% | $498,386 | 16.0% | $418,644 | $13,200 | $405,444 | 10.83% | $242,498 |
| FY2033 | $905,000 | 4.0% | 54.9% | $496,457 | 16.6% | $414,045 | $14,000 | $400,045 | 10.64% | $216,265 |
| FY2034 | $941,000 | 4.0% | 52.4% | $493,353 | 17.2% | $408,496 | $14,400 | $394,096 | 10.45% | $192,896 |
| FY2035 | $974,000 | 3.5% | 50.0% | $487,000 | 17.8% | $400,314 | $13,200 | $387,114 | 10.26% | $171,849 |
| FY2036 | $1,008,000 | 3.5% | 50.0% | $504,000 | 18.4% | $411,264 | $13,600 | $397,664 | 10.07% | $160,384 |
| FY2037 | $1,043,000 | 3.5% | 50.0% | $521,500 | 19.0% | $422,415 | $14,000 | $408,415 | 9.88% | $149,911 |
| PV of FCFF (stub + FY28–37), as of 7/26/26 | $2,359,261 |
| Terminal value (g 3.5%, ROIC 30%, reinvest 11.7%, tax 19%, stable WACC 9.88%) | $6,054,016 |
| PV terminal value · share of operating value | $2,119,905 · 47% |
| Operating asset value at 7/26 → rolled to 10/20 (86 days at HG WACC) | $4,479,166 → $4,588,979 |
| Risk-free (10Y UST, 10/20/26) | 5.12% |
| Equity risk premium (project standard) | 4.40% |
| Bottom-up unlevered beta (semiconductors; same 1.30 as the AMAT/LRCX/KLAC builds) | 1.30 |
| Market D/E (debt + leases $38,860M / mkt cap $5.23T) | 0.74% |
| Levered beta | 1.308 |
| Cost of equity | 10.87% |
| Pre-tax cost of debt (AA-range, Rf + 0.60%) → after-tax at 21% | 5.72% → 4.52% |
| Debt weight | 0.74% |
| WACC: high growth (stub–FY32), gliding over FY33–37 to | 10.83% |
| Stable WACC (β 1.10 unlevered, 5% debt) | 9.88% |
| Reverse DCF: WACC that equates base case to $217.15 | 10.05% (stable 9.10%) |
| Operating asset value (rolled to 10/20/26) | $4,588,979 |
| + Cash & equivalents $22,443 + marketable debt securities $34,143 | $56,586 |
| + Publicly held equity securities (fair value) | $42,783 |
| + Non-marketable equity stakes (carrying value; AI labs etc.) | $51,157 |
| − Debt (incl. $25B notes issued Jun-26) | −$33,366 |
| − Operating lease liabilities | −$5,494 |
| = Value of equity | $4,700,645 |
| ÷ Shares: 24,100M outstanding (8/21/26) + 168M unvested RSUs/PSUs | 24,268M |
| Value per share | $193.70 |
| Market cap at $217.15 (basic) · EV | $5.23T · $5.12T |
| Cross-checks: EV / FY28 base NOPAT · P/E on FY28 consensus EPS (~$15.5) · on FY27E (~$9.2) | 13.4× · ~14× · ~24× |
| Street: 69 Buy / 2 Hold / 1 Sell · targets low / mean / high | $180 / $334 / $710 |