NVIDIA (NVDA): Buffett/Munger-Style Fundamental Review

Prepared for: Brad Wolff | Date: Wed Oct 7, 2026 (America/New_York) | Analyst stance: owner's view, not a trader's view

> Verdict: WAIT. This is a wonderful business at a fair, not cheap, price, and there are new balance-sheet and cash-flow warning signs. Buy below about $185, or after a clean Q3 FY27 print (see §7). Do not trade it impulsively around earnings.

---

0. Quote & key sources

Primary filings used:

• Q2 FY27 press release (8-K Ex. 99.1): https://www.sec.gov/Archives/edgar/data/1045810/000104581026000073/q2fy27pr.htm
• Q2 FY27 CFO Commentary: https://s201.q4cdn.com/141608511/files/doc_financials/2027/Q227/Q2FY27-CFO-Commentary.pdf
• Q2 FY27 10-Q (quarter ended Jul 26, 2026): https://www.sec.gov/Archives/edgar/data/1045810/000104581026000075/nvda-20260726.htm
• FY26 10-K (year ended Jan 25, 2026): https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm
• FY24 10-K (for FY22–FY23 segment data): https://www.sec.gov/Archives/edgar/data/1045810/000104581024000029/nvda-20240128.htm
• Q3 FY26 10-Q: https://www.sec.gov/Archives/edgar/data/1045810/000104581025000230/nvda-20251026.htm
• SEC XBRL company facts (historical financials): https://data.sec.gov/api/xbrl/companyfacts/CIK0001045810.json
• Form 4 filings (insiders): https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001045810&type=4&dateb=&owner=include&count=40

Method note: Ratios (DSO, ROIC, multiples, FCF yield) are my calculations from the filed numbers above. Where I estimated something, I say so.

---

1. Past: what the business has done

1a. Five fiscal years (FY ends late January)

Sources: SEC XBRL company facts (above) and the 10-Ks. FY22 shares are adjusted for the 2024 10-for-1 split. \FCF = operating cash flow − purchases of PP&E and intangibles. \\*ROIC (my calculation) = operating income × (1 − 15% normalized tax) ÷ average (equity + debt + operating leases − cash & marketable debt securities). It is very high partly because the capital base nets out big cash piles, but even rough versions put it in the "extraordinary" tier.

Munger's lesson from FY23: in a down year (crypto bust plus inventory glut), revenue was flat and operating margin fell from 37% to 16%. This is still a cyclical hardware company that sells to a few huge buyers. Don't let the last three years erase that memory.

1b. Segment mix (old reporting, FY22–FY26)

Sources: FY24 10-K (FY22–FY24) and FY26 10-K (FY24–FY26; FY26 Data Center = Compute $162.4B + Networking $31.4B).

New reporting from FY27: Data Center (split into Hyperscale and AI Clouds, Industrial & Enterprise, "ACIE") plus Edge Computing. In Q2 FY27, Data Center was $89.0B (92.5% of revenue): Hyperscale $48.7B and ACIE $40.3B. Edge Computing was $7.2B. (CFO Commentary)

1c. Last four quarters

Sources: XBRL facts, Q3 FY26 10-Q, Q2 FY27 press release and CFO Commentary. †Q4 FY26 Data Center is derived as FY26 total ($193.7B) minus 9M FY26 ($131.4B). GAAP net income in Q1/Q2 FY27 includes $15.9B and $7.8B of mark-to-market gains on equity stakes (CFO Commentary reconciliation). That is investment gains, not chip profits. Non-GAAP net income was $45.5B and $54.0B.

---

2. Present: where things stand now

2a. Latest quarter (Q2 FY27, reported Aug 26, 2026) and guidance

• Revenue $96.2B, up 18% q/q and 106% y/y. Data Center $89.0B, up 117% y/y. GAAP/non-GAAP GM 75.0%. GAAP EPS $2.46, non-GAAP EPS $2.22. (PR)
• Q3 FY27 guide: revenue $108.0B ±2%, assuming no China data-center compute revenue. GM 74.0% ±50bp. Opex ~$9.2B GAAP / $9.0B non-GAAP. FY27 tax rate 16–18%. (same)
• On the call: CFO Kress said Nvidia preliminarily expects ~70% revenue growth in FY28, limited by supply. She said GM should bottom at 71–72% in Q4 FY27 because of "extreme" memory pricing, then run 72–73% in FY28. Vera Rubin is expected to be ~20% of Q3 Data Center revenue. (MarketBeat call summary; Reuters headline: https://www.reuters.com/business/media-telecom/nvidia-forecasts-quarterly-revenue-above-estimates-2026-08-26/)
• Good governance change: starting in Q1 FY27, non-GAAP results no longer exclude stock-based compensation. (PR non-GAAP note)

2b. Valuation snapshot (price $237.47, Oct 7 close)

2c. Balance sheet (Jul 26, 2026 vs Jan 25, 2026) PR

• Cash + marketable debt securities: $56.6B (cash $22.4B + debt securities $34.1B).
• Marketable equity securities: $42.8B (was $12.9B). Locked-up long-term public equity held in Other assets: $5.0B (was $4.8B). Non-marketable (private) equity: $51.2B (was $22.3B). The 10-Q MD&A totals these as $99B of equity investments, plus $25B of unfunded equity commitments (10-Q). Many of these holdings are in customers or AI labs. The CFO said on the call it has invested "nearly $50B in frontier AI labs."
• Debt: $33.4B (was $8.5B). Nvidia issued $25.0B of senior notes in June 2026, maturing 2028–2056 at 4.25%–5.625% (8-K). It is still net cash, but less so.
• Equity: $229.0B. Total assets: $320.3B.

2d. Share count, buybacks, dividend, SBC

• Diluted shares went from 25.35B (FY22) to 24.29B (Q2 FY27), about −4% in 4.5 years. (PR, XBRL)
• Buybacks: $93.4B FY23–FY26, plus $39.0B in H1 FY27. That is ~$132B to shrink the count ~4%, so much of the spend offsets dilution from employee stock rather than truly shrinking the company.
• Sep 28, 2026: Nvidia added $150B to its buyback authorization, bringing the total remaining to $235B, to be executed "through fiscal year 2028." (NVIDIA newsroom)
• The quarterly dividend is now $0.25/share (~$6.0B paid in Q2). (PR)
• SBC: $6.4B in FY26 (3.0% of revenue) and $7.2B TTM, falling as a share of revenue. Not a problem at this scale.

2e. Insider activity: Form 4s filed Apr 7 – Oct 7, 2026 (36 filings parsed from EDGAR)

• Total open-market sales ≈ $1.38B. Insider open-market purchases: zero.
• Director Mark Stevens (via his Third Millennium Trust) sold ~6.12M shares for ~$1.35B between Jun 2 and Sep 18, 2026, at ~$209.70–$234.00. The filings are not flagged as 10b5-1 plan trades. After these sales he still reports ~11.5M shares held directly plus ~15.0M in the Envy Trust. Example filing: https://www.sec.gov/Archives/edgar/data/1045810/000119903926000016/
• EVP/GC Tim Teter: ~$13.3M sold. CFO Colette Kress: ~$7.6M sold (10b5-1 plan adopted Jun 16, 2026). Directors Stephen Neal (~$3.3M) and John Dabiri (~$0.1M) also sold.
• CEO Jensen Huang made no open-market sales in this window. He filed only gifts (e.g., 400,000 shares on Jun 16 and 438,000 on Sep 17) and tax-withholding (code F) transactions. Director Tench Coxe gifted 500,000 shares/month under a 10b5-1 plan.
• Form 144s (notices of intended sales) were filed Sep 17 and Sep 21, so more sales may show up.
• Read-through: insiders selling a stock that has run this far is normal. A long-tenured director selling ~$1.35B outside a pre-set plan isn't a smoking gun, but it isn't a vote of confidence either.

2f. Customer concentration (10-Q / 10-K)

• Q2 FY27: one direct customer = 16% of revenue. H1 FY27: three direct customers = 16%, 15%, 13%, together 44%. (10-Q)
• FY26: two direct customers = 22% and 14%. FY25 was 12%, 11%, 11%. (FY26 10-K)
• Receivables are even more concentrated: five customers = 22%, 14%, 13%, 11%, 10% of AR at Jul 26, 2026 (70% total). (10-Q)
• Nvidia also says "one AI research and deployment company" (widely understood to be OpenAI; the filing doesn't name it) contributed a meaningful amount of revenue indirectly through cloud purchases.
• Revenue from customers headquartered outside the U.S. was 38% of Q2 FY27. (10-Q)

2g. Inventory, purchase commitments, receivables

• Inventory: $15.0B (Jul-25) → $19.8B → $21.4B → $25.8B → $31.6B (Jul-26). Days of inventory ≈ 119, vs ~106 a year ago (my calc). Management says the build is for the Vera Rubin launch. (CFO Commentary)
• Supply & capacity commitments: ~$45.8B (Jul-25, XBRL "purchase obligation") → $95.2B (Jan-26, 10-K) → $119B (Apr-26) → $279B (Jul-26), mostly for memory. Including cloud contracts, leases, equity commitments and capex, total commitments are $366B. Separately: $36B in AI-cloud service agreements, $20B in data-center leases meant to be reassigned to third parties, and $108.5B maximum guarantees (CFO Commentary). Definitions changed over time, so treat the early figures as approximate.
• Receivables / DSO: AR jumped from $40.7B to $63.1B in one quarter. DSO went from 45 to 60 days. Nvidia blamed "extended payment terms on large, multi-quarter agreements with certain investment-grade customers." (CFO Commentary) That is why Q2 operating cash flow was only $24.1B against $59.7B net income.

---

3. Future

3a. Consensus (Yahoo Finance, viewed Oct 7, 2026; cross-checked with Equibles)

Estimates have been going up: per Equibles, the FY28 EPS average rose from ~$12.71 ninety days ago to ~$15.8. The next earnings report is expected Tuesday, November 17, 2026, after the close (stated by NVIDIA IR on the August 26 call; not yet formally announced).

3b. Hyperscaler capex

• 2026 guidance for Microsoft, Alphabet, Amazon and Meta combined is ~$720–745B: Alphabet $195–205B (raised), Amazon ~$220B (raised, citing memory costs), Meta $130–145B, Microsoft ~$175B (lower because of a lease-accounting change, not a spending cut). (TMT Finance; EffectStory tracker)
• 2027: analyst consensus is ~$934.5B for the four (MarketScreener data cited by I/O Fund). Alphabet says 2027 capex will "increase significantly." Amazon says it will be capacity-constrained through 2027.
• New demand sources keep appearing. SpaceX is reportedly seeking ~$40B of financing to buy Nvidia chips and has said it will build exclusively on Nvidia (CNBC, Oct 7). On the call, AWS's plan to deploy 2M more Nvidia GPUs through Q2 FY29 was disclosed.
• Caution: a lot of this spending is now debt-financed: customer bond deals, private credit, and Nvidia's own guarantees. Capex that depends on credit markets can stop quickly.

3c. Competition

• Broadcom (custom ASICs): AI semiconductor revenue was $16.7B in its Aug-2026 quarter, up 221% y/y, guided to $21.7B next quarter. That is growing faster than Nvidia's Data Center (+117%) from a smaller base. (AVGO 8-K)
• Google TPU: Google now sells and rents TPUs to outside customers. Anthropic has committed to up to ~1M TPUs (SemiAnalysis). Meta signed a multibillion-dollar TPU rental deal, per The Information as reported by WinBuzzer.
• Amazon Trainium 3: in production with Anthropic and, reportedly, OpenAI workloads (industry reports; not verified against Amazon filings).
• AMD: OpenAI agreed to deploy up to 6 GW of AMD GPUs, starting with 1 GW of MI450 in 2H 2026 (AMD IR).
• Net: Nvidia's share is still dominant, but the biggest buyers are deliberately dual-sourcing. That caps how far Nvidia can push prices over time.

3d. China / export controls

• The 10-Q says Nvidia is "effectively foreclosed" from China's data-center compute market. The U.S. licensed small amounts of H200 starting Feb 2026, but Beijing restricted purchases. H200 shipments were <1% of Data Center revenue in Q2, and Nvidia took a $0.4B H200 charge in H1 FY27. Guidance assumes zero China data-center compute revenue. (10-Q; context: Tom's Hardware, Bloomberg via The Edge)
• China is not zero in revenue. Customers headquartered in China (including Hong Kong) were $7.88B, or 8.2%, of Q2 FY27 revenue, up from $4.55B in Q1. That comes from non-data-center-compute products and from purchases deployed outside China. Rules that target customers by headquarters could reach it. (10-Q geographic revenue table)
• Implication: China data-center compute is now upside optionality, not a hole in the estimates. The risk is the reverse: China builds a domestic ecosystem (Huawei and others) that later competes globally, which the 10-Q itself warns about. A further risk is the U.S. tightening rules on gaming GPUs or other regions. The 10-Q also notes the French Competition Authority is examining gaming vs. data-center GPU categories.

3e. Supply chain (TSMC, CoWoS, HBM)

• Nvidia says supply is the binding constraint: FY28's ~70% growth is "supply-constrained" and demand is "much higher" (call summary). It has locked in $279B of supply, mostly memory.
• Memory costs are pressuring margins: GM is guided down to a 71–72% trough in Q4 FY27. Amazon also cited memory costs when it raised capex.
• CoWoS packaging capacity at TSMC is still tight. Mizuho reportedly raised its CoWoS capacity forecasts (reported by Investing.com: https://www.investing.com/news/stock-market-news/mizuho-lifts-tsmc-cowos-capacity-forecasts-as-server-cpu-demand-surges-4769157). I could not load the full article, so specific wafer numbers are unverified and not used here.
• Single-point risk: essentially all leading-edge production runs through TSMC in Taiwan.

3f. Regulatory / antitrust / M&A

• China SAMR: in Sep 2025, preliminary finding that Nvidia violated conditions of its Mellanox approval (Reuters). Still unresolved as of May 2026 (TechTimes).
• U.S. DOJ (Groq deal): the Justice Department is investigating whether Nvidia structured its December 2025 Groq licensing deal to avoid antitrust review, and has sent Nvidia a formal request for information. A fine is possible; an unwinding is described as unlikely. (Reuters, Sep 10, 2026, citing the NYT; Bloomberg Law). Nvidia paid $2.9B related to Groq in H1 FY27 (cash-flow statement). The status of the broader DOJ AI-chip inquiry reported in 2024 remains unverified.
• Hugging Face acquisition: ~$11.9B plus up to ~$1.0B in retention equity, expected to close 1H 2027 subject to regulatory approval (8-K Sep 3, 2026). This adds antitrust exposure ("dominant chip vendor buys the open-model hub").
• 10-Q legal note: Nvidia says it is subject to "regulatory and government inquiries and investigations" that it does not expect to be material.

3g. Management / organization changes (8-Ks)

• New Chief Accounting Officer Scott Gawel (from Intel) started May 4, 2026, replacing retiring Donald Robertson (8-K).
• Longtime sales chief Ajay Puri is retiring; Nicholas Parker (ex-Microsoft) takes over (8-K). These look routine, but a CAO change during a period of complex new financing arrangements deserves attention.

---

4. Moat assessment

Bottom line on the moat: this is a real, wide moat with exceptional current economics: ROIC above 100% and ~45% FCF margins. But it is a technology moat, and those need constant rebuilding (an annual product cadence). Buffett historically avoided these because the 10-year outcome is hard to predict. Durability is good for the next 2–3 years and genuinely uncertain beyond that.

---

5. Red flags, ranked by severity (plain language)

1. 🔴 Nvidia is increasingly financing its own customers. It holds $99B of equity investments, plus $25B of unfunded commitments, many of them in buyers or AI labs (≈$50B "in frontier AI labs"). It spent $42.4B buying equity in H1 FY27 alone. It signed $36B of AI-cloud service agreements under which it effectively rents back capacity, plus $20B of leases meant for third parties. It also guaranteed up to $105B of SB Energy data-center leases for OpenAI in Ohio. If OpenAI can't pay, Nvidia may owe the shortfall (8-K Aug 17). Plain English: some of the demand is being made possible by Nvidia's own balance sheet. That was a hallmark of the 2000 telecom-equipment bubble (Lucent, Nortel vendor financing).
2. 🔴 Cash isn't keeping up with profits. Q2 operating cash flow was $24.1B against $59.7B net income (40% conversion, vs ~85% the prior two quarters). Receivables jumped $22B in one quarter, and DSO went from 45 to 60 days on "extended payment terms." When a seller extends terms to keep growing, watch closely.
3. 🟠 Reported earnings include big investment gains. H1 FY27 GAAP net income includes $23.7B of mark-to-market gains on equity stakes. Those gains can reverse just as quickly. Use non-GAAP EPS or core operating income, not GAAP EPS.
4. 🟠 Huge fixed commitments in a cyclical industry. $279B of supply commitments (up from $95B in January) and $366B in total. If demand pauses, as in FY23 when operating margin fell from 37% to 16%, these turn into write-downs. The $4.5B H20 charge in FY26 was a small preview.
5. 🟠 Customer concentration. Three direct customers were 44% of H1 revenue, and five customers hold 70% of receivables. A few CFOs deciding to "digest" capacity for a year would hit Nvidia hard.
6. 🟡 Margin peak and competition. GM is guided down to 71–72% (memory costs). Broadcom's AI revenue is growing ~2x faster, Google sells TPUs externally, and OpenAI and Meta are dual-sourcing. Today's ~75% margin is probably close to the peak.
7. 🟡 Balance sheet is less pristine. Debt went from $8.5B to $33.4B, and Nvidia borrowed $25B while buying back ~$20B/quarter. That's still fine, but the trend points the wrong way.
8. 🟡 Insider selling with no buying. ~$1.38B sold in 6 months, mostly by director Mark Stevens outside a 10b5-1 plan. No insider bought shares.
9. 🟡 Geopolitics. China data-center compute is already near zero, but China-headquartered customers were still 8.2% of Q2 revenue and exposed to headquarters-based rules. Add Taiwan/TSMC concentration risk, possible new U.S. rules on gaming GPUs, the unresolved SAMR antitrust case, the DOJ inquiry into the Groq deal, and Hugging Face deal scrutiny.
10. ⚪ Accounting/organization changes. New CAO, new sales chief, a re-cut segment structure, and a customer reclassified between Hyperscale and ACIE in Q2. Each is benign on its own, but together they make trends harder to compare.

---

6. Valuation: what is the price assuming?

6a. Reverse DCF (my model; 10-year explicit period, 3% terminal growth)

• Starting from TTM FCF of $126.9B and an EV of ~$5.70T, the price implies ~18% annual FCF growth for 10 years at a 10% discount rate (15.6% at 9%, 20.3% at 11%). That means FCF of roughly $660B+ by year 10, about 5x today. Using FCF after SBC ($119.7B), the requirement is ~19%.
• A second angle: if consensus is delivered (FY27 FCF ≈ $190B per StockAnalysis, and FY28 FCF ≈ $323B, which is my estimate of 85% of consensus FY28 net income ≈ $15.91 × ~23.9B shares), then the price needs only ~7.7%/yr FCF growth for FY29–FY36.
• What that means: the stock is not pricing in heroic growth beyond FY28. It is pricing in that FY28 (~$690B revenue, ~55% net margin) is a durable plateau, not a cyclical peak. If FY28-level FCF just stayed flat forever, my DCF gives about $154/share. If it then declined 5%/yr, about $117/share.

6b. Scenarios to FY31 (Jan-2031, ~4.3 years); discount rate 10%; ~$1/yr dividends included

• Probability-weighted value ≈ $263/share vs. $237.47 price → margin of safety ≈ 10%.
• Buffett/Graham want ~25–35% for a business with this much cyclicality and technology risk. A 30% margin of safety on ~$263 means ~$185. That would have been possible this year: the 52-week low was $164.27, and the stock was down 11% YTD on Mar 30 (Bloomberg).
• The bull case is real, and on forward earnings (~15x FY28) the stock looks cheap. But a low P/E on peak-cycle earnings is the classic semiconductor value trap. The question isn't "is 15x cheap?" but "will FY28 earnings still be there in FY31?"

All assumptions above are mine and illustrative, not forecasts.

---

7. Final verdict: WAIT

Rationale: Nvidia is the best business in the AI supply chain, with ROIC above 100%, ~45% FCF margins, a real CUDA-plus-systems moat, and management that has executed brilliantly. At $237.47 you are paying a fair price that assumes FY28's record earnings are permanent. That leaves only ~10% margin of safety on my numbers. Meanwhile, Q2 introduced exactly the signals an old-school owner is trained to fear: cash conversion fell to 40%, DSO rose to 60 days on extended terms, $279B of fixed supply commitments, $99B of equity investments in customers and AI labs, and a $105B lease guarantee for OpenAI. Munger's rule is "a great company at a fair price," but he also said to invert. The way to lose money here isn't missing a quarter. It's buying near a cyclical earnings peak that was partly financed by the seller.

What would change my mind:

• Price: below ~$185 (≈30% margin of safety vs. my ~$263 value) → starter BUY, in tranches.
• Event: a clean Q3 FY27 report (expected Tue Nov 17, 2026, after the close; stated by NVIDIA IR on the Aug 26 call, not yet formally announced) with: OCF ÷ net income back above ~75%, DSO back to ≤ ~50 days, GM trough held at ≥71% with FY28 ~70% growth reaffirmed, and no new large guarantees or customer-financing deals. That would move me to BUY at a fair price (even ~$230–240) for a long-term holder.
• Steer clear if: DSO or receivables keep climbing, guarantees or vendor financing expand again, any big hyperscaler cuts 2027 capex, or GM guidance breaks below ~70% while inventory keeps rising.

Student fund vs. Brad's own trading:

• Weekly student-fund pitch: suitable as a pitch, framed as "HOLD / accumulate below ~$185." It's liquid, data-rich, and has a genuine debate (forward P/E ~15x vs. customer financing and cash conversion). Show the reverse DCF and the red-flag list. Size it at or below benchmark weight. It's a great teaching case on "cheap P/E at a cyclical peak."
• Brad personally: don't trade it impulsively. This is a $5.7T stock whose price swings on one earnings print and one hyperscaler capex comment. Buying the week of earnings is gambling, not investing. If you want exposure: write the thesis down, pick your buy price (≤$185) or your event (a clean Q3), use a small starter position with no options and no margin, and re-check only at the next 10-Q.

--- Data as of Oct 7, 2026 close. Every figure is from the linked filings, releases, or quote pages, or is labeled as my calculation or estimate. Unverified items are flagged in the text. This is educational analysis, not personalized investment advice.