{"allowContribute":false,"item":{"artifactHtml":"\u003cstyle\u003e:root{--paper:#F4F7F4;--ink:#131A15;--ink2:#4D5A50;--acc:#0E7A3D;--acc2:#8A6D1B;--soft:#DDEEE2;--soft2:#F0EAD3;--hair:#D3DED6}:root[data-theme=dark]{--paper:#101511;--ink:#E8EEE9;--ink2:#A9B7AC;--acc:#5BD98A;--acc2:#D9BE6A;--soft:#14301F;--soft2:#2E2A16;--hair:#26332A}.doc{background:var(--paper);color:var(--ink);padding:28px 30px;font:16px/1.72 -apple-system,Segoe UI,Helvetica,Arial,sans-serif}.eyebrow{font:600 11px ui-monospace,Menlo,monospace;letter-spacing:.16em;color:var(--acc)}h1{font-size:27px;line-height:1.25;margin:12px 0 0}.lede{color:var(--ink2);max-width:64ch}h2{font-size:19px;margin:26px 0 8px}h3{font-size:15px;margin:18px 0 6px}p{color:var(--ink2);max-width:66ch}p strong,li strong{color:var(--ink)}.stats{display:grid;grid-template-columns:repeat(auto-fit,minmax(150px,1fr));gap:12px;margin:18px 0}.stat{background:var(--soft);border-radius:8px;padding:12px 14px}.stat .k{font-size:12px;color:var(--acc);font-weight:700}.stat .v{font-size:22px;font-weight:700;margin-top:4px}.stat .n{font-size:12px;color:var(--ink2);margin-top:2px}.callout{display:flex;gap:13px;padding:14px 16px;background:var(--soft2);border-radius:6px;margin:18px 0}.callout .tag{flex:none;font:700 11px ui-monospace,monospace;color:var(--acc2);padding-top:3px}table{border-collapse:collapse;width:100%;font-size:14px}th,td{text-align:left;padding:9px 12px;border-bottom:1px solid var(--hair)}th{font-size:11px;color:var(--ink2)}.wrap{overflow-x:auto}.src{font-size:13px}@media(max-width:640px){.doc{padding:20px 18px}h1{font-size:23px}}\u003c/style\u003e\u003cdiv class=doc\u003e\u003cheader\u003e\u003cdiv class=eyebrow\u003eDEEP DIVE - US LARGE CAP - SEP 9 2026\u003c/div\u003e\u003ch1\u003eDeep Dive NVDA Sep 9 2026\u003c/h1\u003e\u003cp class=lede\u003eNvidia is priced as the full-stack toll road for AI compute. Q2 FY2027 posted 96.2B revenue (+106 pct) with Data Center at 89B (+117 pct), plus a first-ever year-ahead guide of about 70 pct growth for FY2028. Long-term reasoning support for a 5-year hold frame. No buy or sell call.\u003c/p\u003e\u003c/header\u003e\u003csection data-cube=stats class=stats\u003e\u003cdiv class=stat\u003e\u003cdiv class=k\u003eNVDA Sep 8 close\u003c/div\u003e\u003cdiv class=v\u003e225.73\u003c/div\u003e\u003cdiv class=n\u003eDown 2.0 pct on day\u003c/div\u003e\u003c/div\u003e\u003cdiv class=stat\u003e\u003cdiv class=k\u003eQ2 FY2027 revenue\u003c/div\u003e\u003cdiv class=v\u003e96.2B USD\u003c/div\u003e\u003cdiv class=n\u003eUp 106 pct YoY; beat 92.2B est\u003c/div\u003e\u003c/div\u003e\u003cdiv class=stat\u003e\u003cdiv class=k\u003eData Center\u003c/div\u003e\u003cdiv class=v\u003e89.0B USD\u003c/div\u003e\u003cdiv class=n\u003eUp 117 pct YoY; 93 pct of total\u003c/div\u003e\u003c/div\u003e\u003cdiv class=stat\u003e\u003cdiv class=k\u003eQ3 guide\u003c/div\u003e\u003cdiv class=v\u003e108B USD\u003c/div\u003e\u003cdiv class=n\u003ePlus/minus 2 pct; GM 74 pct\u003c/div\u003e\u003c/div\u003e\u003c/section\u003e\u003cdiv class=callout\u003e\u003cspan class=tag\u003eFRAME\u003c/span\u003e\u003cdiv\u003e\u003cstrong\u003eWhat this is:\u003c/strong\u003e fundamentals-first read for a multi-year holder. Fed path and oil tape from the Sep 9 digest are context, not drivers.\u003c/div\u003e\u003c/div\u003e\u003ch2\u003e1 - Thesis\u003c/h2\u003e\u003cp\u003eAI compute is becoming revenue-generating infrastructure, in CEO Jensen Huang words from the August call: compute is revenue. Nvidia sells the full factory: Vera CPU plus Rubin GPU, NVLink and InfiniBand networking, racks and systems, plus the CUDA software layer. Each generation extracts more dollars per gigawatt: Hopper about 18B per GW, Blackwell about 25B, Vera Rubin about 40B (per Sep 7 24/7 Wall St synthesis). With FY2028 guided to about 70 pct on a supply-constrained basis and hyperscale backlogs fully utilized, the bull case is multi-year compounding on AI capex. The bear case is concentration, competition, and circularity compressing margin and multiple.\u003c/p\u003e\u003ch2\u003e2 - Business\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003eFull-stack AI factory.\u003c/strong\u003e Data Center is the company now (93 pct of Q2 FY2027). Moat stack in order: (1) \u003cstrong\u003eCUDA ecosystem\u003c/strong\u003e, hardest to replicate; (2) \u003cstrong\u003enetworking plus systems integration\u003c/strong\u003e that makes GPU clusters work at scale; (3) \u003cstrong\u003esilicon cadence\u003c/strong\u003e with Vera Rubin in full production. Gaming, ProViz, Auto are rounding errors for valuation.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eDemand visibility.\u003c/strong\u003e Management: NVIDIA Compute in every cloud served is fully utilized. FY2028 guide implies roughly 690 to 700B revenue, 100B above 570B consensus (Fortune, Aug 26-27), framed as supply-limited. Supply commitments jumped 119B to \u003cstrong\u003e279B\u003c/strong\u003e. Third-party capital over \u003cstrong\u003e500B\u003c/strong\u003e mobilized with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, KKR; plus a 20-year Ohio compute campus with SoftBank Energy for OpenAI.\u003c/p\u003e\u003ch2\u003e3 - Fundamentals\u003c/h2\u003e\u003cdiv class=wrap\u003e\u003ctable\u003e\u003ctr\u003e\u003cth\u003eItem\u003c/th\u003e\u003cth\u003eFigure\u003c/th\u003e\u003cth\u003eSource\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eRevenue\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e96.22B, +106 pct YoY\u003c/td\u003e\u003ctd\u003eNvidia newsroom Aug 26 2026\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eData Center\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e89.0B, +117 pct YoY\u003c/td\u003e\u003ctd\u003eNvidia newsroom Aug 26 2026\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eGross margin\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e75.0 pct GAAP and non-GAAP\u003c/td\u003e\u003ctd\u003eNvidia newsroom Aug 26 2026\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eEPS\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003eGAAP 2.46; non-GAAP 2.22 vs 2.10 est\u003c/td\u003e\u003ctd\u003eNewsroom + CNBC Aug 26\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eQ3 guide\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e108B; GM 74 pct\u003c/td\u003e\u003ctd\u003eNewsroom Aug 26\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eCapital returns\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e26B buybacks + dividends; 0.25 div Oct 1\u003c/td\u003e\u003ctd\u003eNewsroom Aug 26\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eFY2028\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003eAbout 70 pct growth, supply-constrained\u003c/td\u003e\u003ctd\u003eFortune Aug 26-27\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eFwd P/E framing\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003eNVDA cheaper than AVGO; both sub-20x\u003c/td\u003e\u003ctd\u003eFool Sep 5 via digest\u003c/td\u003e\u003c/tr\u003e\u003c/table\u003e\u003c/div\u003e\u003cp\u003e\u003cstrong\u003eValuation read:\u003c/strong\u003e the Sep 5 Fool comparison frames Nvidia as the cheaper AI-hardware comp vs Broadcom, with a 30x-multiple exercise implying about 100 pct upside for NVDA vs 50 pct plus for AVGO. Treat as arithmetic, not forecast. Margin path: Q3 74 pct, Q4 trough 71-72, settling 72-73 next fiscal (Fortune read of call). Elite but directionally down on memory costs and mix.\u003c/p\u003e\u003ch2\u003e4 - Drivers\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003e1. Buildout breadth.\u003c/strong\u003e Multiple frontier labs scaling in parallel plus new AI labs; labs to be about a quarter of business next year (CFO Kress). \u003cstrong\u003e2. Revenue density per watt\u003c/strong\u003e rising each generation. \u003cstrong\u003e3. CUDA lock-in\u003c/strong\u003e: ASIC share gains do not equal Nvidia revenue loss unless workloads leave the ecosystem. \u003cstrong\u003e4. Supply chain as weapon\u003c/strong\u003e: early upstream engagement pulls supply to Nvidia first.\u003c/p\u003e\u003ch2\u003e5 - Risks\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003e1. Custom silicon (structural).\u003c/strong\u003e Broadcom AI semi +221 pct to 16.7B with Alphabet, Meta, Anthropic, OpenAI wins; Meta Iris in production Sep 2026; Nvidia about 85 pct of AI-processor revenue (Mar 2026) with ASICs above 10 pct. Question is GPU vs ASIC share of incremental training and inference. \u003cstrong\u003e2. China.\u003c/strong\u003e Q3 outlook assumes zero China Data Center compute revenue; Q2 took about 8B hit; Bernstein cited: China share 40 pct (2025) to 8 pct (2026); domestic alternatives advancing. \u003cstrong\u003e3. Circular financing.\u003c/strong\u003e Investments and guarantees across AI labs drew critiques; Kress: low-risk, high-reward, paid via hardware plus rental revenue, not loans. If lab funding tightens, backlog becomes receivable risk. \u003cstrong\u003e4. Supply concentration and cost.\u003c/strong\u003e Taiwan dependence, memory costs, 279B obligation book. \u003cstrong\u003e5. Expectations.\u003c/strong\u003e Bar is the guide, not consensus; minus 2 pct post-earnings drift shows valuation debate dominates even on beats.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eNoise to discount:\u003c/strong\u003e 13-F headlines (Loeb/Third Point exiting NVDA and AVGO in Q2 per Barron mid-Aug; Tepper adding; Soros puts per 24/7 Wall St Aug 16) are position management, not thesis evidence.\u003c/p\u003e\u003ch2\u003e6 - What changes my view\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003eBullish if:\u003c/strong\u003e Rubin dollars-per-GW holds with margins above 73 pct; backlog diversifies beyond hyperscalers; China optionality returns without margin sacrifice. \u003cstrong\u003eBearish if:\u003c/strong\u003e ASICs inflect in inference with CUDA bypass; Data Center sequentials decelerate while obligations stay elevated; guarantee terms worsen; margins settle durably below 70 pct. \u003cstrong\u003eWatch:\u003c/strong\u003e Q3 108B delivery and Q4 margin trough; FY2028 70 pct trajectory; per-generation dollar-per-GW disclosure; China language; 279B book changes.\u003c/p\u003e\u003ch2\u003e7 - Sources\u003c/h2\u003e\u003cul class=src\u003e\u003cli\u003eNvidia newsroom Aug 26 2026 (qtr ended Jul 26): revenue, Data Center, margin, EPS, Q3 guide, dividend, buybacks.\u003c/li\u003e\u003cli\u003eCNBC Aug 26 2026: EPS 2.22 vs 2.10, revenue 96.22B vs 92.17B.\u003c/li\u003e\u003cli\u003eFortune Aug 26-27 2026: FY2028 70 pct guide, circular defense, 279B obligations, margin path.\u003c/li\u003e\u003cli\u003e24/7 Wall St Sep 7 2026: platform framing, per-GW ladder, 500B capital, Ohio project.\u003c/li\u003e\u003cli\u003eFool Sep 5 2026 via Sep 9 digest: NVDA vs AVGO forward P/E.\u003c/li\u003e\u003cli\u003eRisks: AIMultiple (85 pct share Mar 2026); tech-insider (Meta Iris Sep 2026); Bernstein via press (China 40 to 8 pct); Barron plus 24/7 Wall St Aug 16 (13-F moves); S and P Global preview Aug 24.\u003c/li\u003e\u003cli\u003eQuote: Sep 9 digest snapshot (Sep 8 close 225.73, minus 2.0 pct) cross-checked vs platform adapter (last 225.73). No price target cited; none verified from primary source in window.\u003c/li\u003e\u003cli\u003eRecent sources Aug-Sep 2026 plus timeless primary docs. Reasoning support only, not a buy or sell recommendation.\u003c/li\u003e\u003c/ul\u003e\u003c/div\u003e","content":"\u003cdiv class=\"doc\"\u003e\u003ctable\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd\u003e\u003cp\u003eNVDA Sep 8 close\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003e225.73\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003eDown 2.0 pct on day\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cp\u003eQ2 FY2027 revenue\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003e96.2B USD\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003eUp 106 pct YoY; beat 92.2B est\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cp\u003eData Center\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003e89.0B USD\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003eUp 117 pct YoY; 93 pct of total\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cp\u003eQ3 guide\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003e108B USD\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003ePlus/minus 2 pct; GM 74 pct\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003cdiv class=\"callout\"\u003e\u003cspan class=\"tag\"\u003eFRAME\u003c/span\u003e\u003cdiv\u003e\u003cstrong\u003eWhat this is:\u003c/strong\u003e fundamentals-first read for a multi-year holder. Fed path and oil tape from the Sep 9 digest are context, not drivers.\u003c/div\u003e\u003c/div\u003e\u003ch2\u003e1 - Thesis\u003c/h2\u003e\u003cp\u003eAI compute is becoming revenue-generating infrastructure, in CEO Jensen Huang words from the August call: compute is revenue. Nvidia sells the full factory: Vera CPU plus Rubin GPU, NVLink and InfiniBand networking, racks and systems, plus the CUDA software layer. Each generation extracts more dollars per gigawatt: Hopper about 18B per GW, Blackwell about 25B, Vera Rubin about 40B (per Sep 7 24/7 Wall St synthesis). With FY2028 guided to about 70 pct on a supply-constrained basis and hyperscale backlogs fully utilized, the bull case is multi-year compounding on AI capex. The bear case is concentration, competition, and circularity compressing margin and multiple.\u003c/p\u003e\u003ch2\u003e2 - Business\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003eFull-stack AI factory.\u003c/strong\u003e Data Center is the company now (93 pct of Q2 FY2027). Moat stack in order: (1) \u003cstrong\u003eCUDA ecosystem\u003c/strong\u003e, hardest to replicate; (2) \u003cstrong\u003enetworking plus systems integration\u003c/strong\u003e that makes GPU clusters work at scale; (3) \u003cstrong\u003esilicon cadence\u003c/strong\u003e with Vera Rubin in full production. Gaming, ProViz, Auto are rounding errors for valuation.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eDemand visibility.\u003c/strong\u003e Management: NVIDIA Compute in every cloud served is fully utilized. FY2028 guide implies roughly 690 to 700B revenue, 100B above 570B consensus (Fortune, Aug 26-27), framed as supply-limited. Supply commitments jumped 119B to \u003cstrong\u003e279B\u003c/strong\u003e. Third-party capital over \u003cstrong\u003e500B\u003c/strong\u003e mobilized with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, KKR; plus a 20-year Ohio compute campus with SoftBank Energy for OpenAI.\u003c/p\u003e\u003ch2\u003e3 - Fundamentals\u003c/h2\u003e\u003cdiv class=\"wrap\"\u003e\u003ctable\u003e\u003ctbody\u003e\u003ctr\u003e\u003cth\u003eItem\u003c/th\u003e\u003cth\u003eFigure\u003c/th\u003e\u003cth\u003eSource\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eRevenue\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e96.22B, +106 pct YoY\u003c/td\u003e\u003ctd\u003eNvidia newsroom Aug 26 2026\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eData Center\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e89.0B, +117 pct YoY\u003c/td\u003e\u003ctd\u003eNvidia newsroom Aug 26 2026\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eGross margin\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e75.0 pct GAAP and non-GAAP\u003c/td\u003e\u003ctd\u003eNvidia newsroom Aug 26 2026\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eEPS\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003eGAAP 2.46; non-GAAP 2.22 vs 2.10 est\u003c/td\u003e\u003ctd\u003eNewsroom + CNBC Aug 26\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eQ3 guide\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e108B; GM 74 pct\u003c/td\u003e\u003ctd\u003eNewsroom Aug 26\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eCapital returns\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e26B buybacks + dividends; 0.25 div Oct 1\u003c/td\u003e\u003ctd\u003eNewsroom Aug 26\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eFY2028\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003eAbout 70 pct growth, supply-constrained\u003c/td\u003e\u003ctd\u003eFortune Aug 26-27\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eFwd P/E framing\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003eNVDA cheaper than AVGO; both sub-20x\u003c/td\u003e\u003ctd\u003eFool Sep 5 via digest\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e\u003cp\u003e\u003cstrong\u003eValuation read:\u003c/strong\u003e the Sep 5 Fool comparison frames Nvidia as the cheaper AI-hardware comp vs Broadcom, with a 30x-multiple exercise implying about 100 pct upside for NVDA vs 50 pct plus for AVGO. Treat as arithmetic, not forecast. Margin path: Q3 74 pct, Q4 trough 71-72, settling 72-73 next fiscal (Fortune read of call). Elite but directionally down on memory costs and mix.\u003c/p\u003e\u003ch2\u003e4 - Drivers\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003e1. Buildout breadth.\u003c/strong\u003e Multiple frontier labs scaling in parallel plus new AI labs; labs to be about a quarter of business next year (CFO Kress). \u003cstrong\u003e2. Revenue density per watt\u003c/strong\u003e rising each generation. \u003cstrong\u003e3. CUDA lock-in\u003c/strong\u003e: ASIC share gains do not equal Nvidia revenue loss unless workloads leave the ecosystem. \u003cstrong\u003e4. Supply chain as weapon\u003c/strong\u003e: early upstream engagement pulls supply to Nvidia first.\u003c/p\u003e\u003ch2\u003e5 - Risks\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003e1. Custom silicon (structural).\u003c/strong\u003e Broadcom AI semi +221 pct to 16.7B with Alphabet, Meta, Anthropic, OpenAI wins; Meta Iris in production Sep 2026; Nvidia about 85 pct of AI-processor revenue (Mar 2026) with ASICs above 10 pct. Question is GPU vs ASIC share of incremental training and inference. \u003cstrong\u003e2. China.\u003c/strong\u003e Q3 outlook assumes zero China Data Center compute revenue; Q2 took about 8B hit; Bernstein cited: China share 40 pct (2025) to 8 pct (2026); domestic alternatives advancing. \u003cstrong\u003e3. Circular financing.\u003c/strong\u003e Investments and guarantees across AI labs drew critiques; Kress: low-risk, high-reward, paid via hardware plus rental revenue, not loans. If lab funding tightens, backlog becomes receivable risk. \u003cstrong\u003e4. Supply concentration and cost.\u003c/strong\u003e Taiwan dependence, memory costs, 279B obligation book. \u003cstrong\u003e5. Expectations.\u003c/strong\u003e Bar is the guide, not consensus; minus 2 pct post-earnings drift shows valuation debate dominates even on beats.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eNoise to discount:\u003c/strong\u003e 13-F headlines (Loeb/Third Point exiting NVDA and AVGO in Q2 per Barron mid-Aug; Tepper adding; Soros puts per 24/7 Wall St Aug 16) are position management, not thesis evidence.\u003c/p\u003e\u003ch2\u003e6 - What changes my view\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003eBullish if:\u003c/strong\u003e Rubin dollars-per-GW holds with margins above 73 pct; backlog diversifies beyond hyperscalers; China optionality returns without margin sacrifice. \u003cstrong\u003eBearish if:\u003c/strong\u003e ASICs inflect in inference with CUDA bypass; Data Center sequentials decelerate while obligations stay elevated; guarantee terms worsen; margins settle durably below 70 pct. \u003cstrong\u003eWatch:\u003c/strong\u003e Q3 108B delivery and Q4 margin trough; FY2028 70 pct trajectory; per-generation dollar-per-GW disclosure; China language; 279B book changes.\u003c/p\u003e\u003ch2\u003e7 - Sources\u003c/h2\u003e\u003cul class=\"src\"\u003e\u003cli\u003eNvidia newsroom Aug 26 2026 (qtr ended Jul 26): revenue, Data Center, margin, EPS, Q3 guide, dividend, buybacks.\u003c/li\u003e\u003cli\u003eCNBC Aug 26 2026: EPS 2.22 vs 2.10, revenue 96.22B vs 92.17B.\u003c/li\u003e\u003cli\u003eFortune Aug 26-27 2026: FY2028 70 pct guide, circular defense, 279B obligations, margin path.\u003c/li\u003e\u003cli\u003e24/7 Wall St Sep 7 2026: platform framing, per-GW ladder, 500B capital, Ohio project.\u003c/li\u003e\u003cli\u003eFool Sep 5 2026 via Sep 9 digest: NVDA vs AVGO forward P/E.\u003c/li\u003e\u003cli\u003eRisks: AIMultiple (85 pct share Mar 2026); tech-insider (Meta Iris Sep 2026); Bernstein via press (China 40 to 8 pct); Barron plus 24/7 Wall St Aug 16 (13-F moves); S and P Global preview Aug 24.\u003c/li\u003e\u003cli\u003eQuote: Sep 9 digest snapshot (Sep 8 close 225.73, minus 2.0 pct) cross-checked vs platform adapter (last 225.73). No price target cited; none verified from primary source in window.\u003c/li\u003e\u003cli\u003eRecent sources Aug-Sep 2026 plus timeless primary docs. Reasoning support only, not a buy or sell recommendation.\u003c/li\u003e\u003c/ul\u003e\u003c/div\u003e","createdAt":1788924873360,"displayMode":"artifact","embedRenders":{},"id":"a4088ffcb008c54e766d5257","isNew":false,"isPublic":true,"itemType":"NOTE","name":"Deep Dive NVDA Sep 9 2026","parents":{"4600249c072694ac4a99304c":1788924873360},"updatedAt":1788924925717,"updatedBy":{"userId":"6aa0d1ae01029037a723cc","userName":"min chi Hsu"},"version":4},"ownerName":"min chi Hsu","subtree":[{"artifactHtml":"\u003cstyle\u003e:root{--paper:#F4F7F4;--ink:#131A15;--ink2:#4D5A50;--acc:#0E7A3D;--acc2:#8A6D1B;--soft:#DDEEE2;--soft2:#F0EAD3;--hair:#D3DED6}:root[data-theme=dark]{--paper:#101511;--ink:#E8EEE9;--ink2:#A9B7AC;--acc:#5BD98A;--acc2:#D9BE6A;--soft:#14301F;--soft2:#2E2A16;--hair:#26332A}.doc{background:var(--paper);color:var(--ink);padding:28px 30px;font:16px/1.72 -apple-system,Segoe UI,Helvetica,Arial,sans-serif}.eyebrow{font:600 11px ui-monospace,Menlo,monospace;letter-spacing:.16em;color:var(--acc)}h1{font-size:27px;line-height:1.25;margin:12px 0 0}.lede{color:var(--ink2);max-width:64ch}h2{font-size:19px;margin:26px 0 8px}h3{font-size:15px;margin:18px 0 6px}p{color:var(--ink2);max-width:66ch}p strong,li strong{color:var(--ink)}.stats{display:grid;grid-template-columns:repeat(auto-fit,minmax(150px,1fr));gap:12px;margin:18px 0}.stat{background:var(--soft);border-radius:8px;padding:12px 14px}.stat .k{font-size:12px;color:var(--acc);font-weight:700}.stat .v{font-size:22px;font-weight:700;margin-top:4px}.stat .n{font-size:12px;color:var(--ink2);margin-top:2px}.callout{display:flex;gap:13px;padding:14px 16px;background:var(--soft2);border-radius:6px;margin:18px 0}.callout .tag{flex:none;font:700 11px ui-monospace,monospace;color:var(--acc2);padding-top:3px}table{border-collapse:collapse;width:100%;font-size:14px}th,td{text-align:left;padding:9px 12px;border-bottom:1px solid var(--hair)}th{font-size:11px;color:var(--ink2)}.wrap{overflow-x:auto}.src{font-size:13px}@media(max-width:640px){.doc{padding:20px 18px}h1{font-size:23px}}\u003c/style\u003e\u003cdiv class=doc\u003e\u003cheader\u003e\u003cdiv class=eyebrow\u003eDEEP DIVE - US LARGE CAP - SEP 9 2026\u003c/div\u003e\u003ch1\u003eDeep Dive NVDA Sep 9 2026\u003c/h1\u003e\u003cp class=lede\u003eNvidia is priced as the full-stack toll road for AI compute. Q2 FY2027 posted 96.2B revenue (+106 pct) with Data Center at 89B (+117 pct), plus a first-ever year-ahead guide of about 70 pct growth for FY2028. Long-term reasoning support for a 5-year hold frame. No buy or sell call.\u003c/p\u003e\u003c/header\u003e\u003csection data-cube=stats class=stats\u003e\u003cdiv class=stat\u003e\u003cdiv class=k\u003eNVDA Sep 8 close\u003c/div\u003e\u003cdiv class=v\u003e225.73\u003c/div\u003e\u003cdiv class=n\u003eDown 2.0 pct on day\u003c/div\u003e\u003c/div\u003e\u003cdiv class=stat\u003e\u003cdiv class=k\u003eQ2 FY2027 revenue\u003c/div\u003e\u003cdiv class=v\u003e96.2B USD\u003c/div\u003e\u003cdiv class=n\u003eUp 106 pct YoY; beat 92.2B est\u003c/div\u003e\u003c/div\u003e\u003cdiv class=stat\u003e\u003cdiv class=k\u003eData Center\u003c/div\u003e\u003cdiv class=v\u003e89.0B USD\u003c/div\u003e\u003cdiv class=n\u003eUp 117 pct YoY; 93 pct of total\u003c/div\u003e\u003c/div\u003e\u003cdiv class=stat\u003e\u003cdiv class=k\u003eQ3 guide\u003c/div\u003e\u003cdiv class=v\u003e108B USD\u003c/div\u003e\u003cdiv class=n\u003ePlus/minus 2 pct; GM 74 pct\u003c/div\u003e\u003c/div\u003e\u003c/section\u003e\u003cdiv class=callout\u003e\u003cspan class=tag\u003eFRAME\u003c/span\u003e\u003cdiv\u003e\u003cstrong\u003eWhat this is:\u003c/strong\u003e fundamentals-first read for a multi-year holder. Fed path and oil tape from the Sep 9 digest are context, not drivers.\u003c/div\u003e\u003c/div\u003e\u003ch2\u003e1 - Thesis\u003c/h2\u003e\u003cp\u003eAI compute is becoming revenue-generating infrastructure, in CEO Jensen Huang words from the August call: compute is revenue. Nvidia sells the full factory: Vera CPU plus Rubin GPU, NVLink and InfiniBand networking, racks and systems, plus the CUDA software layer. Each generation extracts more dollars per gigawatt: Hopper about 18B per GW, Blackwell about 25B, Vera Rubin about 40B (per Sep 7 24/7 Wall St synthesis). With FY2028 guided to about 70 pct on a supply-constrained basis and hyperscale backlogs fully utilized, the bull case is multi-year compounding on AI capex. The bear case is concentration, competition, and circularity compressing margin and multiple.\u003c/p\u003e\u003ch2\u003e2 - Business\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003eFull-stack AI factory.\u003c/strong\u003e Data Center is the company now (93 pct of Q2 FY2027). Moat stack in order: (1) \u003cstrong\u003eCUDA ecosystem\u003c/strong\u003e, hardest to replicate; (2) \u003cstrong\u003enetworking plus systems integration\u003c/strong\u003e that makes GPU clusters work at scale; (3) \u003cstrong\u003esilicon cadence\u003c/strong\u003e with Vera Rubin in full production. Gaming, ProViz, Auto are rounding errors for valuation.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eDemand visibility.\u003c/strong\u003e Management: NVIDIA Compute in every cloud served is fully utilized. FY2028 guide implies roughly 690 to 700B revenue, 100B above 570B consensus (Fortune, Aug 26-27), framed as supply-limited. Supply commitments jumped 119B to \u003cstrong\u003e279B\u003c/strong\u003e. Third-party capital over \u003cstrong\u003e500B\u003c/strong\u003e mobilized with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, KKR; plus a 20-year Ohio compute campus with SoftBank Energy for OpenAI.\u003c/p\u003e\u003ch2\u003e3 - Fundamentals\u003c/h2\u003e\u003cdiv class=wrap\u003e\u003ctable\u003e\u003ctr\u003e\u003cth\u003eItem\u003c/th\u003e\u003cth\u003eFigure\u003c/th\u003e\u003cth\u003eSource\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eRevenue\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e96.22B, +106 pct YoY\u003c/td\u003e\u003ctd\u003eNvidia newsroom Aug 26 2026\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eData Center\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e89.0B, +117 pct YoY\u003c/td\u003e\u003ctd\u003eNvidia newsroom Aug 26 2026\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eGross margin\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e75.0 pct GAAP and non-GAAP\u003c/td\u003e\u003ctd\u003eNvidia newsroom Aug 26 2026\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eEPS\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003eGAAP 2.46; non-GAAP 2.22 vs 2.10 est\u003c/td\u003e\u003ctd\u003eNewsroom + CNBC Aug 26\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eQ3 guide\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e108B; GM 74 pct\u003c/td\u003e\u003ctd\u003eNewsroom Aug 26\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eCapital returns\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e26B buybacks + dividends; 0.25 div Oct 1\u003c/td\u003e\u003ctd\u003eNewsroom Aug 26\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eFY2028\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003eAbout 70 pct growth, supply-constrained\u003c/td\u003e\u003ctd\u003eFortune Aug 26-27\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eFwd P/E framing\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003eNVDA cheaper than AVGO; both sub-20x\u003c/td\u003e\u003ctd\u003eFool Sep 5 via digest\u003c/td\u003e\u003c/tr\u003e\u003c/table\u003e\u003c/div\u003e\u003cp\u003e\u003cstrong\u003eValuation read:\u003c/strong\u003e the Sep 5 Fool comparison frames Nvidia as the cheaper AI-hardware comp vs Broadcom, with a 30x-multiple exercise implying about 100 pct upside for NVDA vs 50 pct plus for AVGO. Treat as arithmetic, not forecast. Margin path: Q3 74 pct, Q4 trough 71-72, settling 72-73 next fiscal (Fortune read of call). Elite but directionally down on memory costs and mix.\u003c/p\u003e\u003ch2\u003e4 - Drivers\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003e1. Buildout breadth.\u003c/strong\u003e Multiple frontier labs scaling in parallel plus new AI labs; labs to be about a quarter of business next year (CFO Kress). \u003cstrong\u003e2. Revenue density per watt\u003c/strong\u003e rising each generation. \u003cstrong\u003e3. CUDA lock-in\u003c/strong\u003e: ASIC share gains do not equal Nvidia revenue loss unless workloads leave the ecosystem. \u003cstrong\u003e4. Supply chain as weapon\u003c/strong\u003e: early upstream engagement pulls supply to Nvidia first.\u003c/p\u003e\u003ch2\u003e5 - Risks\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003e1. Custom silicon (structural).\u003c/strong\u003e Broadcom AI semi +221 pct to 16.7B with Alphabet, Meta, Anthropic, OpenAI wins; Meta Iris in production Sep 2026; Nvidia about 85 pct of AI-processor revenue (Mar 2026) with ASICs above 10 pct. Question is GPU vs ASIC share of incremental training and inference. \u003cstrong\u003e2. China.\u003c/strong\u003e Q3 outlook assumes zero China Data Center compute revenue; Q2 took about 8B hit; Bernstein cited: China share 40 pct (2025) to 8 pct (2026); domestic alternatives advancing. \u003cstrong\u003e3. Circular financing.\u003c/strong\u003e Investments and guarantees across AI labs drew critiques; Kress: low-risk, high-reward, paid via hardware plus rental revenue, not loans. If lab funding tightens, backlog becomes receivable risk. \u003cstrong\u003e4. Supply concentration and cost.\u003c/strong\u003e Taiwan dependence, memory costs, 279B obligation book. \u003cstrong\u003e5. Expectations.\u003c/strong\u003e Bar is the guide, not consensus; minus 2 pct post-earnings drift shows valuation debate dominates even on beats.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eNoise to discount:\u003c/strong\u003e 13-F headlines (Loeb/Third Point exiting NVDA and AVGO in Q2 per Barron mid-Aug; Tepper adding; Soros puts per 24/7 Wall St Aug 16) are position management, not thesis evidence.\u003c/p\u003e\u003ch2\u003e6 - What changes my view\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003eBullish if:\u003c/strong\u003e Rubin dollars-per-GW holds with margins above 73 pct; backlog diversifies beyond hyperscalers; China optionality returns without margin sacrifice. \u003cstrong\u003eBearish if:\u003c/strong\u003e ASICs inflect in inference with CUDA bypass; Data Center sequentials decelerate while obligations stay elevated; guarantee terms worsen; margins settle durably below 70 pct. \u003cstrong\u003eWatch:\u003c/strong\u003e Q3 108B delivery and Q4 margin trough; FY2028 70 pct trajectory; per-generation dollar-per-GW disclosure; China language; 279B book changes.\u003c/p\u003e\u003ch2\u003e7 - Sources\u003c/h2\u003e\u003cul class=src\u003e\u003cli\u003eNvidia newsroom Aug 26 2026 (qtr ended Jul 26): revenue, Data Center, margin, EPS, Q3 guide, dividend, buybacks.\u003c/li\u003e\u003cli\u003eCNBC Aug 26 2026: EPS 2.22 vs 2.10, revenue 96.22B vs 92.17B.\u003c/li\u003e\u003cli\u003eFortune Aug 26-27 2026: FY2028 70 pct guide, circular defense, 279B obligations, margin path.\u003c/li\u003e\u003cli\u003e24/7 Wall St Sep 7 2026: platform framing, per-GW ladder, 500B capital, Ohio project.\u003c/li\u003e\u003cli\u003eFool Sep 5 2026 via Sep 9 digest: NVDA vs AVGO forward P/E.\u003c/li\u003e\u003cli\u003eRisks: AIMultiple (85 pct share Mar 2026); tech-insider (Meta Iris Sep 2026); Bernstein via press (China 40 to 8 pct); Barron plus 24/7 Wall St Aug 16 (13-F moves); S and P Global preview Aug 24.\u003c/li\u003e\u003cli\u003eQuote: Sep 9 digest snapshot (Sep 8 close 225.73, minus 2.0 pct) cross-checked vs platform adapter (last 225.73). No price target cited; none verified from primary source in window.\u003c/li\u003e\u003cli\u003eRecent sources Aug-Sep 2026 plus timeless primary docs. Reasoning support only, not a buy or sell recommendation.\u003c/li\u003e\u003c/ul\u003e\u003c/div\u003e","content":"\u003cdiv class=\"doc\"\u003e\u003ctable\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd\u003e\u003cp\u003eNVDA Sep 8 close\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003e225.73\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003eDown 2.0 pct on day\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cp\u003eQ2 FY2027 revenue\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003e96.2B USD\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003eUp 106 pct YoY; beat 92.2B est\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cp\u003eData Center\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003e89.0B USD\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003eUp 117 pct YoY; 93 pct of total\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cp\u003eQ3 guide\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003e108B USD\u003c/p\u003e\u003c/td\u003e\u003ctd\u003e\u003cp\u003ePlus/minus 2 pct; GM 74 pct\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003cdiv class=\"callout\"\u003e\u003cspan class=\"tag\"\u003eFRAME\u003c/span\u003e\u003cdiv\u003e\u003cstrong\u003eWhat this is:\u003c/strong\u003e fundamentals-first read for a multi-year holder. Fed path and oil tape from the Sep 9 digest are context, not drivers.\u003c/div\u003e\u003c/div\u003e\u003ch2\u003e1 - Thesis\u003c/h2\u003e\u003cp\u003eAI compute is becoming revenue-generating infrastructure, in CEO Jensen Huang words from the August call: compute is revenue. Nvidia sells the full factory: Vera CPU plus Rubin GPU, NVLink and InfiniBand networking, racks and systems, plus the CUDA software layer. Each generation extracts more dollars per gigawatt: Hopper about 18B per GW, Blackwell about 25B, Vera Rubin about 40B (per Sep 7 24/7 Wall St synthesis). With FY2028 guided to about 70 pct on a supply-constrained basis and hyperscale backlogs fully utilized, the bull case is multi-year compounding on AI capex. The bear case is concentration, competition, and circularity compressing margin and multiple.\u003c/p\u003e\u003ch2\u003e2 - Business\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003eFull-stack AI factory.\u003c/strong\u003e Data Center is the company now (93 pct of Q2 FY2027). Moat stack in order: (1) \u003cstrong\u003eCUDA ecosystem\u003c/strong\u003e, hardest to replicate; (2) \u003cstrong\u003enetworking plus systems integration\u003c/strong\u003e that makes GPU clusters work at scale; (3) \u003cstrong\u003esilicon cadence\u003c/strong\u003e with Vera Rubin in full production. Gaming, ProViz, Auto are rounding errors for valuation.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eDemand visibility.\u003c/strong\u003e Management: NVIDIA Compute in every cloud served is fully utilized. FY2028 guide implies roughly 690 to 700B revenue, 100B above 570B consensus (Fortune, Aug 26-27), framed as supply-limited. Supply commitments jumped 119B to \u003cstrong\u003e279B\u003c/strong\u003e. Third-party capital over \u003cstrong\u003e500B\u003c/strong\u003e mobilized with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, KKR; plus a 20-year Ohio compute campus with SoftBank Energy for OpenAI.\u003c/p\u003e\u003ch2\u003e3 - Fundamentals\u003c/h2\u003e\u003cdiv class=\"wrap\"\u003e\u003ctable\u003e\u003ctbody\u003e\u003ctr\u003e\u003cth\u003eItem\u003c/th\u003e\u003cth\u003eFigure\u003c/th\u003e\u003cth\u003eSource\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eRevenue\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e96.22B, +106 pct YoY\u003c/td\u003e\u003ctd\u003eNvidia newsroom Aug 26 2026\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eData Center\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e89.0B, +117 pct YoY\u003c/td\u003e\u003ctd\u003eNvidia newsroom Aug 26 2026\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eGross margin\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e75.0 pct GAAP and non-GAAP\u003c/td\u003e\u003ctd\u003eNvidia newsroom Aug 26 2026\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eEPS\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003eGAAP 2.46; non-GAAP 2.22 vs 2.10 est\u003c/td\u003e\u003ctd\u003eNewsroom + CNBC Aug 26\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eQ3 guide\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e108B; GM 74 pct\u003c/td\u003e\u003ctd\u003eNewsroom Aug 26\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eCapital returns\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003e26B buybacks + dividends; 0.25 div Oct 1\u003c/td\u003e\u003ctd\u003eNewsroom Aug 26\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eFY2028\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003eAbout 70 pct growth, supply-constrained\u003c/td\u003e\u003ctd\u003eFortune Aug 26-27\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd\u003e\u003cstrong\u003eFwd P/E framing\u003c/strong\u003e\u003c/td\u003e\u003ctd\u003eNVDA cheaper than AVGO; both sub-20x\u003c/td\u003e\u003ctd\u003eFool Sep 5 via digest\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e\u003cp\u003e\u003cstrong\u003eValuation read:\u003c/strong\u003e the Sep 5 Fool comparison frames Nvidia as the cheaper AI-hardware comp vs Broadcom, with a 30x-multiple exercise implying about 100 pct upside for NVDA vs 50 pct plus for AVGO. Treat as arithmetic, not forecast. Margin path: Q3 74 pct, Q4 trough 71-72, settling 72-73 next fiscal (Fortune read of call). Elite but directionally down on memory costs and mix.\u003c/p\u003e\u003ch2\u003e4 - Drivers\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003e1. Buildout breadth.\u003c/strong\u003e Multiple frontier labs scaling in parallel plus new AI labs; labs to be about a quarter of business next year (CFO Kress). \u003cstrong\u003e2. Revenue density per watt\u003c/strong\u003e rising each generation. \u003cstrong\u003e3. CUDA lock-in\u003c/strong\u003e: ASIC share gains do not equal Nvidia revenue loss unless workloads leave the ecosystem. \u003cstrong\u003e4. Supply chain as weapon\u003c/strong\u003e: early upstream engagement pulls supply to Nvidia first.\u003c/p\u003e\u003ch2\u003e5 - Risks\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003e1. Custom silicon (structural).\u003c/strong\u003e Broadcom AI semi +221 pct to 16.7B with Alphabet, Meta, Anthropic, OpenAI wins; Meta Iris in production Sep 2026; Nvidia about 85 pct of AI-processor revenue (Mar 2026) with ASICs above 10 pct. Question is GPU vs ASIC share of incremental training and inference. \u003cstrong\u003e2. China.\u003c/strong\u003e Q3 outlook assumes zero China Data Center compute revenue; Q2 took about 8B hit; Bernstein cited: China share 40 pct (2025) to 8 pct (2026); domestic alternatives advancing. \u003cstrong\u003e3. Circular financing.\u003c/strong\u003e Investments and guarantees across AI labs drew critiques; Kress: low-risk, high-reward, paid via hardware plus rental revenue, not loans. If lab funding tightens, backlog becomes receivable risk. \u003cstrong\u003e4. Supply concentration and cost.\u003c/strong\u003e Taiwan dependence, memory costs, 279B obligation book. \u003cstrong\u003e5. Expectations.\u003c/strong\u003e Bar is the guide, not consensus; minus 2 pct post-earnings drift shows valuation debate dominates even on beats.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eNoise to discount:\u003c/strong\u003e 13-F headlines (Loeb/Third Point exiting NVDA and AVGO in Q2 per Barron mid-Aug; Tepper adding; Soros puts per 24/7 Wall St Aug 16) are position management, not thesis evidence.\u003c/p\u003e\u003ch2\u003e6 - What changes my view\u003c/h2\u003e\u003cp\u003e\u003cstrong\u003eBullish if:\u003c/strong\u003e Rubin dollars-per-GW holds with margins above 73 pct; backlog diversifies beyond hyperscalers; China optionality returns without margin sacrifice. \u003cstrong\u003eBearish if:\u003c/strong\u003e ASICs inflect in inference with CUDA bypass; Data Center sequentials decelerate while obligations stay elevated; guarantee terms worsen; margins settle durably below 70 pct. \u003cstrong\u003eWatch:\u003c/strong\u003e Q3 108B delivery and Q4 margin trough; FY2028 70 pct trajectory; per-generation dollar-per-GW disclosure; China language; 279B book changes.\u003c/p\u003e\u003ch2\u003e7 - Sources\u003c/h2\u003e\u003cul class=\"src\"\u003e\u003cli\u003eNvidia newsroom Aug 26 2026 (qtr ended Jul 26): revenue, Data Center, margin, EPS, Q3 guide, dividend, buybacks.\u003c/li\u003e\u003cli\u003eCNBC Aug 26 2026: EPS 2.22 vs 2.10, revenue 96.22B vs 92.17B.\u003c/li\u003e\u003cli\u003eFortune Aug 26-27 2026: FY2028 70 pct guide, circular defense, 279B obligations, margin path.\u003c/li\u003e\u003cli\u003e24/7 Wall St Sep 7 2026: platform framing, per-GW ladder, 500B capital, Ohio project.\u003c/li\u003e\u003cli\u003eFool Sep 5 2026 via Sep 9 digest: NVDA vs AVGO forward P/E.\u003c/li\u003e\u003cli\u003eRisks: AIMultiple (85 pct share Mar 2026); tech-insider (Meta Iris Sep 2026); Bernstein via press (China 40 to 8 pct); Barron plus 24/7 Wall St Aug 16 (13-F moves); S and P Global preview Aug 24.\u003c/li\u003e\u003cli\u003eQuote: Sep 9 digest snapshot (Sep 8 close 225.73, minus 2.0 pct) cross-checked vs platform adapter (last 225.73). No price target cited; none verified from primary source in window.\u003c/li\u003e\u003cli\u003eRecent sources Aug-Sep 2026 plus timeless primary docs. Reasoning support only, not a buy or sell recommendation.\u003c/li\u003e\u003c/ul\u003e\u003c/div\u003e","createdAt":1788924873360,"displayMode":"artifact","embedRenders":{},"id":"a4088ffcb008c54e766d5257","isNew":false,"isPublic":true,"itemType":"NOTE","name":"Deep Dive NVDA Sep 9 2026","parents":{"4600249c072694ac4a99304c":1788924873360},"updatedAt":1788924925717,"updatedBy":{"userId":"6aa0d1ae01029037a723cc","userName":"min chi Hsu"},"version":4}]}