--- title: "Equity Research: Nvidia Corporation (NVDA) - Q3 2025" author: "DeepValues Research" publisher: "DeepValues Research (DeepValues / LuLu Ventures LLC)" type: "Research article" ticker: "NVDA" company: "Nvidia" sector: "Technology" industry: "Semiconductors" wikidata: "https://www.wikidata.org/wiki/Q182477" wikipedia: "https://en.wikipedia.org/wiki/Nvidia" date: "2025-11-19" language: "en" section: "Company Research" answer: "DeepValues Research published “Equity Research: Nvidia Corporation (NVDA) - Q3 2025” on Nvidia (NVDA) on Nov 19, 2025. 英伟达 (NVDA) 发布的2026财年第三季度财报不仅是一份财务成绩单，更是全球人工智能经济健康状况的宏观晴雨表。" canonical_url: "https://research.deepvalues.ai/company/nvda/nvidia-q3-2025-analysis-2/" markdown_url: "https://research.deepvalues.ai/company/nvda/nvidia-q3-2025-analysis-2/index.md" license: "CC BY 4.0" license_url: "https://creativecommons.org/licenses/by/4.0/" cite_as: "DeepValues Research, Equity Research: Nvidia Corporation (NVDA) - Q3 2025, https://research.deepvalues.ai/company/nvda/nvidia-q3-2025-analysis-2/" publisher_url: "https://deepvalues.ai" app: "https://deepvalues.ai" api: "https://app.deepvalues.ai/developers" data_note: "Reports are dated snapshots. Current data via the DeepValues API: https://app.deepvalues.ai/developers" disclaimer: "Educational research, not investment advice." --- # Equity Research: Nvidia Corporation (NVDA) \- Q3 2025 ## **The AI Infrastructure Supercycle: From Component Supplier to Sovereign Platform** ### **Date: November 21, 2025** ### **Fiscal Period: Q3 FY2026 (Actuals) & Forward Outlook** ### **Investment Rating: BUY** ### **12-Month Target Price: $235.00** --- ## 1\. Executive Summary: The Industrial Revolution of Intelligence Nvidia Corporation (NVDA) stands at the epicenter of a technological transformation that rivals the industrial revolution in scale and economic impact. The company’s Third Quarter Fiscal Year 2026 results, delivered in November 2025, serve not merely as a financial update but as a macroeconomic indicator for the health of the global artificial intelligence economy. Reporting record revenue of **$57.0 billion**, a staggering **62% year-over-year increase**, and **22% sequential growth** 1, Nvidia has demonstrated that the demand for accelerated computing is accelerating rather than plateauing. This performance defies the gravitational pull of the "law of large numbers" and effectively rebuts the "AI Bubble" thesis that has permeated market discourse throughout late 2025\. The central thesis of this report is that Nvidia has successfully transitioned from a hardware component vendor to a full-stack data center platform provider. This transition effectively insulates the company from the commoditization risks that typically plague semiconductor manufacturers. The arrival of the **Blackwell** architecture marks the beginning of a new product supercycle, characterized not just by faster chips, but by "rack-scale" system integration that fundamentally changes how data centers are architected. With Data Center revenue reaching **$51.2 billion**—comprising nearly 90% of total sales—Nvidia is no longer a gaming company with an AI side business; it is the sovereign foundry of the AI age.1 Our valuation analysis, utilizing a Discounted Cash Flow (DCF) model anchored by a **9.14% Weighted Average Cost of Capital (WACC)** 3, suggests that the market continues to underestimate the durability of this infrastructure build-out phase. While skeptics point to the inevitable digestion phase of hyperscaler capital expenditures (CapEx), our research indicates a diversification of demand vectors—specifically "Sovereign AI" and "Enterprise Inference"—that will sustain growth through FY2028 and beyond. The "virtuous cycle" described by CEO Jensen Huang, where AI is used to create more advanced AI, is creating a compounding demand for compute that linear forecasting models fail to capture.1 However, risks remain elevated. The geopolitical landscape is fracturing, evidenced by the effective cessation of China-specific revenue following the H20 silicon restrictions.4 Furthermore, high-profile institutional exits, such as **Thiel Macro's** divestment in Q3 5, highlight the intense debate regarding valuation multiples at these historic highs. Despite these headwinds, Nvidia’s operational execution—maintaining **73.4% gross margins** amidst a complex product transition—signals a competitive moat that is deepening, not eroding.1 We initiate coverage with a **BUY** rating and a price target of **$235.00**, implying significant upside from the current trading level of \~$181. --- ## **2\. Financial Performance Analysis: Q3 FY2026 Deep Dive** ### **2.1 Revenue Dynamics: The Scale of Hypergrowth** In the third quarter of Fiscal Year 2026, ending October 26, 2025, Nvidia delivered a top-line performance that redefines the ceiling for semiconductor companies. Total revenue of **$57.006 billion** beat consensus estimates which hovered around $54 billion to $55 billion.6 To contextualize this scale: Nvidia now generates more revenue in a single quarter than it did in the entire Fiscal Year 2023\. The primary driver remains the **Data Center** segment, which grew **66% year-over-year** and **25% sequentially** to $51.2 billion.1 This sequential acceleration—from 16% growth in Q2 to 25% in Q3—is a critical indicator. It suggests that supply chain constraints, particularly regarding CoWoS (Chip-on-Wafer-on-Substrate) packaging and HBM3e (High Bandwidth Memory) availability, are resolving just as the Blackwell architecture begins its volume ramp. The market had feared a "gap quarter" or an "air pocket" between the peak of the Hopper (H100/H200) cycle and the onset of Blackwell volume; instead, demand for the H200 remained robust even as Blackwell production lines spun up.8 | Segment Performance | Q3 FY2026 ($M) | Q2 FY2026 ($M) | Q3 FY2025 ($M) | Q/Q Growth | Y/Y Growth | | :---- | :---- | :---- | :---- | :---- | :---- | | **Data Center** | $51,200 | $41,100 | $30,800 | \+25% | \+66% | | **Gaming** | $4,300 | $4,358 | $3,300 | \-1% | \+30% | | **Professional Vis.** | $1,100 | $1,050 | $800 | \+5% | \+38% | | **Automotive** | $406 | $235 | $182 | \+73% | \+123% | | **Total Revenue** | **$57,006** | **$46,743** | **$35,082** | **\+22%** | **\+62%** | Table Source: Consolidated from Nvidia Q3 FY2026 Press Release and historical data.1 The **Gaming** segment, while no longer the primary valuation driver, delivered **$4.3 billion** in revenue, a **30% year-over-year increase**.1 This sector is experiencing a cyclical recovery, aided by the anticipation of the RTX 50-series launch. The "Install-to-Play" metrics and the integration of generative AI into gaming workflows (DLSS 4\) are creating a new upgrade cycle for consumer GPUs, ensuring that this segment remains a highly profitable cash cow that funds broader R\&D initiatives. ### **2.2 Profitability Profile: Managing Complexity at Scale** A key focal point for analysts in Q3 FY2026 was gross margin performance. Historically, new product ramps weigh on margins due to initial yield inefficiencies. The Blackwell architecture is significantly more complex than Hopper, involving multi-die packaging and higher HBM content. Nvidia reported a **GAAP Gross Margin of 73.4%**, a sequential expansion of 100 basis points from 72.4% in Q2 FY2026, though down from the peak of 74.6% in Q3 FY2025.1 The year-over-year compression of roughly 120 basis points reflects the "mix shift" towards these more complex systems. However, the sequential improvement is the more salient data point for investors. It indicates that Nvidia’s manufacturing partners (primarily TSMC) are achieving better-than-expected yields on the new nodes sooner than anticipated. Operating leverage continues to be a story of historic efficiency. **Operating Expenses (OpEx)** grew only **8% sequentially** to **$5.84 billion**, while revenue grew 22%.1 This disparity drove **Operating Income** to **$36.0 billion**, a **65% year-over-year increase**. The company is effectively printing cash, with an operating margin exceeding 63%. This level of profitability is rare for hardware companies and is more akin to a mature software monopoly, underscoring the value of the CUDA software ecosystem which locks customers into the Nvidia hardware stack. ### **2.3 Capital Allocation and Shareholder Returns** Nvidia’s balance sheet is a fortress, ending the quarter with cash, cash equivalents, and marketable securities of **$56.8 billion**.11 In a high-interest-rate environment, this cash pile generates substantial interest income, further boosting the bottom line. Management has adopted an aggressive capital return strategy, clearly signaling their view that the stock remains undervalued despite its run-up. In the first nine months of Fiscal 2026, Nvidia returned **$37.0 billion** to shareholders through share repurchases and cash dividends.1 Notably, the board authorized an additional **$60 billion** share repurchase program in August 2025, without expiration.4 This buyback velocity—annualizing at nearly $50 billion—provides a structural bid for the stock and significantly accretive support to Earnings Per Share (EPS). The weighted average shares outstanding (diluted) have remained relatively stable at roughly **24.8 billion shares** despite employee stock-based compensation, thanks to these offset repurchases.4 --- ## **3\. Strategic Analysis: The Blackwell Supercycle & Sovereign AI** ### **3.1 The Blackwell Architecture: A Platform Shift** The transition from Hopper (H100/H200) to Blackwell (GB200) represents more than a generational speed bump; it is a fundamental architectural shift toward "rack-scale" computing. The **GB200 NVL72** system connects 72 Blackwell GPUs and 36 Grace CPUs into a single, massive computer acting as one GPU. Research indicates that Blackwell offers a **10x throughput per megawatt** improvement for inference workloads compared to Hopper.1 This metric is crucial. As hyperscalers like Microsoft, Meta, and Google fill their data centers, they are increasingly power-constrained rather than capital-constrained. A chip that delivers ten times the performance for the same power envelope is not just an upgrade; it is an operational necessity to meet carbon goals and electricity grid limitations. Supply dynamics for Blackwell are critical. CEO Jensen Huang stated that production is "ramping at full speed" and that demand is "extraordinary" and "off the charts".1 This phrasing suggests that the entire supply for FY2027 (calendar 2026\) may already be allocated. The fear of "double ordering"—where customers over-order to secure allocation—is a historical risk in semiconductors. However, the high unit cost of Blackwell systems (estimated at millions of dollars per rack) and the requisite complex data center infrastructure make speculative ordering less likely than in previous component cycles. ### **3.2 Sovereign AI: The New Pillar of Demand** A significant evolution in Nvidia’s FY2026 narrative is the emergence of "Sovereign AI." Nations are increasingly viewing artificial intelligence infrastructure as a matter of national security, akin to energy independence or defense capabilities. They are unwilling to rely solely on US-based public clouds for their critical state data and cultural/linguistic models. The **RIKEN** deployment in Japan serves as a prime case study. RIKEN announced it will deploy two AI systems built on the **Nvidia GB200 NVL4** platform, utilizing 1,600 Blackwell GPUs connected via Quantum-X800 InfiniBand.12 This system is dedicated to scientific research in climate science, life sciences, and industrial applications. Similar initiatives are underway in France, India, Singapore, and the UAE. This "Sovereign AI" demand vector is distinct from hyperscaler demand. It is less sensitive to quarterly advertising revenue fluctuations or consumer interest rates. It is driven by government budgets and long-term strategic industrial policy, adding a layer of resilience and diversification to Nvidia’s revenue base. It effectively reduces customer concentration risk, a key concern for investors worried about the dominance of the "Magnificent Seven" in Nvidia’s order book. ### **3.3 The "Virtuous Cycle" of Generative AI** Nvidia management describes a "virtuous cycle" wherein AI is used to accelerate the development of the next generation of AI. This is not marketing hyperbole but a technical reality. The development of "Agentic AI"—autonomous software agents that can reason, plan, and execute multi-step workflows—requires exponentially more compute for "inference" (running the model) than simple chatbots. As models get smarter, they are used to generate synthetic data to train even larger models, which in turn require more GPUs. Furthermore, the "inference" market is exploding. Inference is now growing "exponentially" and compounding across industries.1 The implication is that for every dollar spent on training a model (CapEx), there is a multiple of that revenue generated in inference (OpEx) over the model's life. Nvidia’s continued dominance in inference benchmarks reinforces the "stickiness" of its CUDA software platform; even if competitors can train models, deploying them efficiently often leads customers back to Nvidia for inference. --- ## **4\. Geopolitical Landscape & The China Decoupling** ### **4.1 The End of the China Growth Story?** Fiscal Year 2026 marked a decisive decoupling of Nvidia’s high-end business from China. Following successive rounds of US Department of Commerce export controls, Nvidia attempted to maintain market access via the compliant **H20** chip. However, Q3 FY2026 data and commentary suggest this strategy has been effectively sunsetted or rendered negligible. Reports indicate "zero share" for the H20 in the Chinese market in recent quarters, largely due to competitive pressure from domestic Chinese alternatives (like Huawei’s Ascend series) and regulatory ambiguity.2 In Q2 FY2026, Nvidia explicitly excluded H20 shipments from its guidance and took inventory write-downs related to the product, though this was financially offset by a $180 million inventory release from sales to non-China customers.4 ### **4.2 Strategic Implications of the Decoupling** The cessation of significant China revenue is a double-edged sword. On the negative side, it permanently removes a massive Total Addressable Market (TAM)—historically 20-25% of data center revenue—from Nvidia’s reach. It also accelerates the development of a rival ecosystem within China that could eventually spill over into emerging markets. However, on the positive side, the Q3 FY2026 results prove that Nvidia does not *need* China to grow at hyper-scale rates. The fact that the company grew Data Center revenue by 66% YoY while effectively exiting its largest historical export market is a testament to the voraciousness of demand in the US, Europe, and allied Asian nations. It also "de-risks" the stock from a regulatory perspective; further US sanctions can no longer hurt a revenue stream that has already been effectively zeroed out. --- ## **5\. Valuation & Discounted Cash Flow (DCF) Analysis** ### **5.1 Methodology and Market Data** To determine the intrinsic value of Nvidia, we employ a 5-Year Discounted Cash Flow (DCF) model. This approach is preferred over Price-to-Earnings (P/E) multiples, which can be distorted during periods of hypergrowth. Our valuation date is November 21, 2025, utilizing the most current market data available. **WACC Calculation Inputs:** * **Risk-Free Rate ($R\_f$):** We utilize the **US 10-Year Treasury Yield**, which stood at **4.13%** as of mid-November 2025\.13 This rate reflects the "higher for longer" interest rate environment. * **Beta ($\\beta$):** We apply a Beta of **1.22**.3 This reflects Nvidia’s high volatility relative to the S\&P 500, although this beta has compressed slightly as the company’s cash flows have become more robust. * **Equity Risk Premium (ERP):** We assume an ERP of **4.12%** 3, consistent with current equity market risk assessments. * Cost of Equity ($K\_e$): Using the Capital Asset Pricing Model (CAPM): $$K\_e \= 4.13\\% \+ 1.22(4.12\\%) \= 9.16\\%$$ * **Cost of Debt ($K\_d$):** Nvidia’s effective pre-tax cost of debt is estimated at **5.13%**, reflecting its A-rated credit status in the current yield environment.3 * **Tax Rate:** We use the non-GAAP effective tax rate guidance of **16.5%**.15 * **Capital Structure:** Nvidia is almost entirely equity-funded. With a market cap approaching $4.5 trillion and debt of roughly $8.5 billion 16, the debt weight is a negligible **0.19%**. * **WACC:** The calculated Weighted Average Cost of Capital is **9.14%**.3 ### **5.2 Forecast Assumptions: The Growth Trajectory** Our revenue projections are based on a "Base Case" scenario that assumes a successful Blackwell ramp followed by a moderation in growth rates as the law of large numbers takes effect. * **FY2026 (Current Year):** We project full-year revenue of **$204 billion**, based on actuals for Q1-Q3 and a conservative Q4 estimate. * **FY2027:** We forecast **$310 billion** in revenue (+52% growth). This aligns with consensus estimates 17 and assumes the peak of the Blackwell deployment cycle across hyperscalers and sovereign entities. * **FY2028-FY2031:** We model a deceleration to **15% growth** by FY2031. This assumes that while AI infrastructure build-out continues, the frantic pace of initial deployment moderates into a replacement and upgrade cycle. | Fiscal Year | Revenue ($B) | Growth Rate | EBIT Margin | Tax Rate | FCF ($B) | PV of FCF ($B) | | :---- | :---- | :---- | :---- | :---- | :---- | :---- | | **2026E** | 204.0 | \- | 65.0% | 16.5% | 110.0 | \- | | **2027E** | 310.0 | 52.0% | 65.0% | 16.5% | 143.1 | **131.1** | | **2028E** | 420.0 | 35.5% | 65.0% | 16.5% | 195.0 | **163.6** | | **2029E** | 525.0 | 25.0% | 65.0% | 16.5% | 243.0 | **186.8** | | **2030E** | 630.0 | 20.0% | 65.0% | 16.5% | 289.4 | **203.9** | | **2031E** | 725.0 | 15.1% | 65.0% | 16.5% | 330.5 | **213.3** | *Note: EBIT Margin assumes sustained operating leverage. FCF is proxied as NOPAT minus Reinvestment, assuming conversion rates remain high due to the fabless model.* **Cumulative Present Value (PV) of Free Cash Flows (2027-2031):** **$898.7 Billion**. ### **5.3 Terminal Value and Intrinsic Price** We apply a **Perpetual Growth Rate of 3.5%** for the terminal value. This is higher than global GDP growth (typically 2-3%) but justified by the belief that AI will be a primary driver of productivity growth for the next several decades, allowing Nvidia to grow faster than the broader economy in perpetuity. * **Terminal FCF (FY2032):** $330.5B \* 1.035 \= $342.1 Billion. * **Terminal Value (TV):** $342.1B / (9.14% \- 3.5%) \= **$6,065.6 Billion**. * **PV of Terminal Value:** $6,065.6B / (1.0914)^5 \= **$3,915.4 Billion**. Enterprise Value Calculation: $$EV \= PV(FCF) \+ PV(TV) \= 898.7B \+ 3,915.4B \= \\$4,814.1 \\text{ Billion}$$ **Equity Value Calculation:** * **Add Cash & Equivalents:** \+$56.8 Billion.11 * **Less Debt:** \-$8.5 Billion.16 * **Net Cash Adjustment:** \+$48.3 Billion. * **Total Equity Value:** **$4,862.4 Billion**. **Per Share Value:** * **Fully Diluted Shares Outstanding:** 24.87 Billion.4 * **Intrinsic Value per Share:** **$195.51**. ### **5.4 Sensitivity and Scenario Analysis** The base case yields a value slightly above the current trading price. However, the "Bull Case" reflects the potential for a prolonged "supercycle" where revenue reaches **$1 trillion+** by FY2031, as suggested by some analyst models.17 * **Bull Case ($1.18T Revenue in FY31):** Implies a share price of **\~$285.00**. * **Bear Case (Stagnation at $400B):** Implies a share price of **\~$135.00**. Given the momentum of the Blackwell launch and the unlocking of Sovereign AI budgets, we apply a probability-weighted target: **40% Bull / 60% Base**, resulting in our 12-month Price Target of **$235.00**. --- ## **6\. Risks and Investment Considerations** ### **6.1 The "Air Pocket" and Cyclicality Risk** The most prevalent bear thesis is the "Air Pocket" risk. This theory posits that once the initial training clusters are built by the hyperscalers, demand will fall off a cliff before inference demand can ramp up to fill the void. This occurred in the networking gear boom of the late 1990s. While Nvidia argues that the "virtuous cycle" mitigates this, investors must remain vigilant for any signs of CapEx pausing among the "Magnificent Seven" (Microsoft, Meta, Google, Amazon). ### **6.2 Competition: The ASIC Threat** While AMD is a merchant silicon competitor, the deeper threat comes from Nvidia’s own customers. Google (TPU), Amazon (Trainium/Inferentia), and Microsoft (Maia) are all designing custom Application Specific Integrated Circuits (ASICs). These chips are optimized for their specific workloads and offer better unit economics than general-purpose GPUs. Nvidia’s defense is the **CUDA software moat** and the versatility of its GPUs. An ASIC designed today for Transformers might be obsolete if the dominant model architecture shifts to State Space Models (SSMs) or another novelty next year. Nvidia GPUs remain the "programmable" safe bet in a rapidly evolving algorithmic landscape. ### **6.3 Institutional Sentiment and Volatility** The exit of **Thiel Macro** 5 signals that some smart money sees the current valuation as stretched or the AI narrative as a bubble. Options markets imply massive volatility (±7% moves on earnings).18 Nvidia is now a "macro" stock; its movements dictate the direction of the S\&P 500\. This high correlation means the stock can be punished for broader macro failures (e.g., interest rate spikes) regardless of its own execution. ### **6.4 Supply Chain Concentration** Nvidia is a fabless designer. It relies almost exclusively on **TSMC** (Taiwan Semiconductor Manufacturing Company) for advanced wafer fabrication and, crucially, for CoWoS packaging. Any geopolitical disruption in the Taiwan Strait represents an existential risk to Nvidia’s ability to deliver product. While TSMC is diversifying its footprint (Arizona, Japan), the leading-edge capacity remains concentrated in Taiwan for the foreseeable future. --- ## **简体中文报告 (Simplified Chinese Section)** ### **股票研究报告：英伟达 (NVIDIA Corp.) —— AI 基础设施的超级周期与主权算力平台** 日期： 2025年11月21日 财季： 2026财年第三季度 (Q3 FY2026) 评级： 买入 (BUY) 12个月目标价： $235.00 ### **1\. 执行摘要：智能工业革命的核心引擎** 英伟达 (NVDA) 发布的2026财年第三季度财报不仅是一份财务成绩单，更是全球人工智能经济健康状况的宏观晴雨表。公司报告了 **570亿美元** 的创纪录营收，同比激增 **62%**，环比增长 **22%** 1。这一业绩彻底打破了市场对于“大数定律”将限制其增长的担忧，并有力回击了2025年末弥漫市场的“AI泡沫”论调。 本报告的核心逻辑在于，英伟达已成功从单纯的硬件组件供应商转型为全栈式数据中心平台提供商。随着 **Blackwell** 架构的推出，公司开启了一个新的产品超级周期。Blackwell 不仅仅是更快的芯片，它是通过 NVLink 技术实现的“机架级”计算系统，根本性地改变了数据中心的构建方式。数据中心业务营收达到 **512亿美元**，占总营收的近90%，标志着英伟达已成为AI时代不可撼动的算力“铸造厂” 1。 基于 **9.14% 加权平均资本成本 (WACC)** 3 的现金流折现 (DCF) 模型分析显示，市场仍低估了这一基础设施建设周期的持久性。尽管存在对科技巨头资本开支 (CapEx) 周期性回调的担忧，但我们的研究发现，“主权AI (Sovereign AI)”和“企业级推理 (Enterprise Inference)”正在成为新的增长引擎，这将支撑公司至2028财年及以后的增长。 然而，风险依然存在。地缘政治的裂痕日益加深，受 H20 芯片出口管制影响，中国市场的收入已几乎归零 4。此外，知名机构如 **Thiel Macro** 在第三季度的清仓离场 5，也凸显了市场在历史高估值下的分歧。尽管如此，英伟达在极其复杂的产品转型期仍保持了 **73.4% 的毛利率** 1，显示出其坚不可摧的竞争护城河。我们给予 **买入** 评级，目标价 **$235.00**。 ### **2\. 财务表现深度解析 (Q3 FY2026)** #### **2.1 营收动态：超大规模增长的常态化** 在截至2025年10月26日的第三季度，英伟达录得 **570.06亿美元** 的总营收，远超华尔街普遍预期的540亿-550亿美元 6。这意味着英伟达现在的单季度营收已超过了2023全财年的总和。 **数据中心 (Data Center)** 业务是绝对的核心驱动力，同比增长 **66%**，环比增长 **25%**，达到 **512亿美元**。环比增速的加快（从Q2的16%提升至Q3的25%）是一个关键信号，表明供应链瓶颈（特别是 CoWoS 封装和 HBM3e 内存）正在缓解，恰逢 Blackwell 架构开始大规模量产。市场此前担心的 Hopper (H100/H200) 周期见顶与 Blackwell 放量之间的“空窗期”并未出现，H200 的需求依然强劲 8。 | 业务部门 | Q3 FY2026 营收 ($M) | 环比增长 (Q/Q) | 同比增长 (Y/Y) | | :---- | :---- | :---- | :---- | | **数据中心** | $51,200 | \+25% | \+66% | | **游戏** | $4,300 | \-1% | \+30% | | **专业可视化** | $1,100 | \+5% | \+38% | | **汽车** | $406 | \+73% | \+123% | | **总营收** | **$57,006** | **\+22%** | **\+62%** | | 数据来源：英伟达 Q3 FY2026 财报及历史数据整理 1。 | | | | #### **2.2 盈利能力：在复杂性中保持高效** **毛利率 (Gross Margin)** 是分析师关注的焦点。GAAP 毛利率为 **73.4%**，虽较去年同期的峰值（74.6%）有所下降，但较上一季度（72.4%）提升了100个基点 1。 * **同比下降原因：** 产品组合向 Blackwell 等更复杂的系统转移，这些系统包含更高的组件成本（如 HBM3e 内存）和复杂的封装工艺。 * **环比上升意义：** 这表明新产品的良率提升速度快于预期，供应链效率正在优化。 **运营利润 (Operating Income)** 飙升至 **360亿美元**，同比增长 **65%**。运营费用 (OpEx) 仅环比增长8%，而营收环比增长22%，体现了极强的经营杠杆效应。这种类似软件公司的利润率水平，凸显了 CUDA 软件生态系统的价值——它成功将客户锁定在英伟达的硬件平台上。 #### **2.3 资本配置与股东回报** 英伟达的资产负债表坚如磐石，期末持有现金及有价证券 568亿美元 11。在高利率环境下，这笔巨额现金产生了可观的利息收入。 公司管理层实施了积极的股东回报策略。在2026财年的前九个月，英伟达通过回购和分红向股东返还了 370亿美元 1。此外，董事会在2025年8月批准了额外 600亿美元 的无到期日股票回购计划 4。这种大规模回购（年化约500亿美元）为股价提供了强有力的支撑，并显著增厚了每股收益 (EPS)。 ### **3\. 战略分析：Blackwell 超级周期与主权 AI** #### **3.1 Blackwell 架构：平台级的代际飞跃** 从 Hopper 到 Blackwell (GB200) 的转变不仅仅是速度的提升，而是向“机架级计算”的架构转移。GB200 NVL72 系统将72个 GPU 和36个 Grace CPU 连接成一个巨大的计算单元。 研究显示，Blackwell 在推理任务上的 每兆瓦吞吐量 (Throughput per Megawatt) 是 Hopper 的 10倍 1。这对于面临电力瓶颈的超大规模云厂商至关重要。CEO 黄仁勋表示，Blackwell 的生产“正全速推进”，需求“疯狂 (off the charts)” 1。这暗示2027财年（日历年2026）的产能可能已被预订一空。 #### **3.2 主权 AI (Sovereign AI)：新的需求支柱** 2026财年英伟达叙事的一个重要演变是“主权 AI”的崛起。各国政府开始将 AI 基础设施视为国家安全问题，不愿完全依赖美国的公共云。 日本 RIKEN 项目 是一个典型案例。RIKEN 宣布部署两套基于 Nvidia GB200 NVL4 平台的 AI 系统，包含1600个 Blackwell GPU，用于气候科学和工业研究 12。法国、印度、中东等地也有类似举措。这种由国家预算驱动的需求，降低了英伟达对少数几家科技巨头的依赖风险。 ### **4\. 地缘政治与中国市场的脱钩** 2026财年标志着英伟达高端业务与中国市场的决定性脱钩。受美国出口管制影响，合规芯片 H20 的销售在Q3已降至微不足道的水平。报道称 H20 在中国市场的份额已接近“归零”，主要受制于国内竞争对手（如华为昇腾）的崛起和监管的不确定性 2。 从积极的一面看，Q3 财报证明了英伟达不需要中国市场也能实现超高速增长。在完全剔除这一历史最大出口市场的情况下，数据中心业务仍实现了66%的增长，这证明了全球其他地区需求的强劲，同时也从监管层面“去风险化”了这只股票。 ### **5\. 估值与 DCF 模型分析** 我们采用5年现金流折现 (DCF) 模型来评估英伟达的内在价值。 **WACC 计算关键假设：** * **无风险利率 ($R\_f$)：** 4.13% (美国10年期国债收益率) 13。 * **Beta ($\\beta$)：** 1.22 3。 * **股权风险溢价 (ERP)：** 4.12%。 * **加权平均资本成本 (WACC)：** **9.14%**。 增长预测： 我们假设 FY2027 营收达到 3100亿美元（同比增长52%），随后几年增速逐渐放缓至 FY2031 的 15%。 * **终值增长率 (Terminal Growth)：** 3.5%。 * **计算结果：** * 终值现值 (PV of Terminal Value)：3.91万亿美元。 * 未来现金流现值 (PV of FCF)：8987亿美元。 * 企业价值 (EV)：4.81万亿美元。 * **每股内在价值：$195.51**。 目标价： 考虑到 Blackwell 带来的“超级周期”潜力，我们结合看涨情景（FY2031 营收破万亿），给予 $235.00 的12个月目标价。 ### **6\. 风险提示** 1. **周期性“气穴” (Air Pocket)：** 如果科技巨头在完成初步训练集群建设后暂停资本开支，需求可能出现断崖式下跌。 2. **ASIC 竞争：** 客户（Google, AWS, Microsoft）自研芯片的长期威胁。 3. **供应链集中：** 对台积电 (TSMC) CoWoS 产能和地缘政治稳定性的高度依赖。 --- ## **繁體中文報告 (Traditional Chinese Section)** ### **股票研究報告：輝達 (NVIDIA Corp.) —— AI 基礎設施的超級週期與主權算力平台** 日期： 2025年11月21日 財季： 2026會計年度第三季度 (Q3 FY2026) 評級： 買入 (BUY) 12個月目標價： $235.00 ### **1\. 執行摘要：智能工業革命的核心引擎** 輝達 (NVDA) 發布的2026會計年度第三季度財報不僅是一份財務成績單，更是全球人工智能經濟健康狀況的宏觀晴雨表。公司報告了 **570億美元** 的創紀錄營收，同比激增 **62%**，環比增長 **22%** 1。這一業績徹底打破了市場對於「大數定律」將限制其增長的擔憂，並有力回擊了2025年末瀰漫市場的「AI泡沫」論調。 本報告的核心邏輯在於，輝達已成功從單純的硬體組件供應商轉型為全棧式資料中心平台提供商。隨著 **Blackwell** 架構的推出，公司開啟了一個新的產品超級週期。Blackwell 不僅僅是更快的晶片，它是通過 NVLink 技術實現的「機架級」運算系統，根本性地改變了資料中心的構建方式。資料中心業務營收達到 **512億美元**，占總營收的近90%，標誌著輝達已成為AI時代不可撼動的算力「鑄造廠」 1。 基於 **9.14% 加權平均資本成本 (WACC)** 3 的現金流折現 (DCF) 模型分析顯示，市場仍低估了這一基礎設施建設週期的持久性。儘管存在對科技巨頭資本開支 (CapEx) 週期性回調的擔憂，但我們的研究發現，「主權AI (Sovereign AI)」和「企業級推論 (Enterprise Inference)」正在成為新的增長引擎，這將支撐公司至2028會計年度及以後的增長。 然而，風險依然存在。地緣政治的裂痕日益加深，受 H20 晶片出口管制影響，中國市場的收入已幾乎歸零 4。此外，知名機構如 **Thiel Macro** 在第三季度的清倉離場 5，也凸顯了市場在歷史高估值下的分歧。儘管如此，輝達在極其複雜的產品轉型期仍保持了 **73.4% 的毛利率** 1，顯示出其堅不可摧的競爭護城河。我們給予 **買入** 評級，目標價 **$235.00**。 ### **2\. 財務表現深度解析 (Q3 FY2026)** #### **2.1 營收動態：超大規模增長的常態化** 在截至2025年10月26日的第三季度，輝達錄得 **570.06億美元** 的總營收，遠超華爾街普遍預期的540億-550億美元 6。這意味著輝達現在的單季度營收已超過了2023全會計年度的總和。 **資料中心 (Data Center)** 業務是絕對的核心驅動力，同比增長 **66%**，環比增長 **25%**，達到 **512億美元**。環比增速的加快（從Q2的16%提升至Q3的25%）是一個關鍵信號，表明供應鏈瓶頸（特別是 CoWoS 封裝和 HBM3e 記憶體）正在緩解，恰逢 Blackwell 架構開始大規模量產。市場此前擔心的 Hopper (H100/H200) 週期見頂與 Blackwell 放量之間的「空窗期」並未出現，H200 的需求依然強勁 8。 | 業務部門 | Q3 FY2026 營收 ($M) | 環比增長 (Q/Q) | 同比增長 (Y/Y) | | :---- | :---- | :---- | :---- | | **資料中心** | $51,200 | \+25% | \+66% | | **遊戲** | $4,300 | \-1% | \+30% | | **專業視覺化** | $1,100 | \+5% | \+38% | | **汽車** | $406 | \+73% | \+123% | | **總營收** | **$57,006** | **\+22%** | **\+62%** | | 數據來源：輝達 Q3 FY2026 財報及歷史數據整理 1。 | | | | #### **2.2 盈利能力：在複雜性中保持高效** **毛利率 (Gross Margin)** 是分析師關注的焦點。GAAP 毛利率為 **73.4%**，雖較去年同期的峰值（74.6%）有所下降，但較上一季度（72.4%）提升了100個基點 1。 * **同比下降原因：** 產品組合向 Blackwell 等更複雜的系統轉移，這些系統包含更高的組件成本（如 HBM3e 記憶體）和複雜的封裝工藝。 * **環比上升意義：** 這表明新產品的良率提升速度快於預期，供應鏈效率正在優化。 **運營利潤 (Operating Income)** 飆升至 **360億美元**，同比增長 **65%**。運營費用 (OpEx) 僅環比增長8%，而營收環比增長22%，體現了極強的經營槓桿效應。這種類似軟體公司的利潤率水平，凸顯了 CUDA 軟體生態系統的價值——它成功將客戶鎖定在輝達的硬體平台上。 #### **2.3 資本配置與股東回報** 輝達的資產負債表堅如磐石，期末持有現金及有價證券 568億美元 11。在高利率環境下，這筆鉅額現金產生了可觀的利息收入。 公司管理層實施了積極的股東回報策略。在2026會計年度的前九個月，輝達通過回購和分紅向股東返還了 370億美元 1。此外，董事會在2025年8月批准了額外 600億美元 的無到期日股票回購計畫 4。這種大規模回購（年化約500億美元）為股價提供了強有力的支撐，並顯著增厚了每股盈餘 (EPS)。 ### **3\. 戰略分析：Blackwell 超級週期與主權 AI** #### **3.1 Blackwell 架構：平台級的代際飛躍** 從 Hopper 到 Blackwell (GB200) 的轉變不僅僅是速度的提升，而是向「機架級運算」的架構轉移。GB200 NVL72 系統將72個 GPU 和36個 Grace CPU 連接成一個巨大的運算單元。 研究顯示，Blackwell 在推論任務上的 每兆瓦吞吐量 (Throughput per Megawatt) 是 Hopper 的 10倍 1。這對於面臨電力瓶頸的超大規模雲廠商至關重要。CEO 黃仁勳表示，Blackwell 的生產「正全速推進」，需求「瘋狂 (off the charts)」 1。這暗示2027會計年度（日曆年2026）的產能可能已被預訂一空。 #### **3.2 主權 AI (Sovereign AI)：新的需求支柱** 2026會計年度輝達敘事的一個重要演變是「主權 AI」的崛起。各國政府開始將 AI 基礎設施視為國家安全問題，不願完全依賴美國的公共雲。 日本 RIKEN 專案 是一個典型案例。RIKEN 宣布部署兩套基於 Nvidia GB200 NVL4 平台的 AI 系統，包含1600個 Blackwell GPU，用於氣候科學和工業研究 12。法國、印度、中東等地也有類似舉措。這種由國家預算驅動的需求，降低了輝達對少數幾家科技巨頭的依賴風險。 ### **4\. 地緣政治與中國市場的脫鉤** 2026會計年度標誌著輝達高端業務與中國市場的決定性脫鉤。受美國出口管制影響，合規晶片 H20 的銷售在Q3已降至微不足道的水平。報導稱 H20 在中國市場的份額已接近「歸零」，主要受制於國內競爭對手（如華為昇騰）的崛起和監管的不確定性 2。 從積極的一面看，Q3 財報證明了輝達不需要中國市場也能實現超高速增長。在完全剔除這一歷史最大出口市場的情況下，資料中心業務仍實現了66%的增長，這證明了全球其他地區需求的強勁，同時也從監管層面「去風險化」了這隻股票。 ### **5\. 估值與 DCF 模型分析** 我們採用5年現金流折現 (DCF) 模型來評估輝達的內在價值。 **WACC 計算關鍵假設：** * **無風險利率 ($R\_f$)：** 4.13% (美國10年期國債收益率) 13。 * **Beta ($\\beta$)：** 1.22 3。 * **股權風險溢價 (ERP)：** 4.12%。 * **加權平均資本成本 (WACC)：** **9.14%**。 增長預測： 我們假設 FY2027 營收達到 3100億美元（同比增長52%），隨後幾年增速逐漸放緩至 FY2031 的 15%。 * **終值增長率 (Terminal Growth)：** 3.5%。 * **計算結果：** * 終值現值 (PV of Terminal Value)：3.91兆美元。 * 未來現金流現值 (PV of FCF)：8987億美元。 * 企業價值 (EV)：4.81兆美元。 * **每股內在價值：$195.51**。 目標價： 考慮到 Blackwell 帶來的「超級週期」潛力，我們結合看漲情境（FY2031 營收破萬億），給予 $235.00 的12個月目標價。 ### **6\. 風險提示** 1. **週期性「氣穴」 (Air Pocket)：** 如果科技巨頭在完成初步訓練集群建設後暫停資本開支，需求可能出現斷崖式下跌。 2. **ASIC 競爭：** 客戶（Google, AWS, Microsoft）自研晶片的長期威脅。 3. **供應鏈集中：** 對台積電 (TSMC) CoWoS 產能和地緣政治穩定性的高度依賴。 #### **Works cited** 1. NVIDIA Announces Financial Results for Third Quarter Fiscal 2026, accessed November 19, 2025, [https://www.globenewswire.com/news-release/2025/11/19/3191444/0/en/NVIDIA-Announces-Financial-Results-for-Third-Quarter-Fiscal-2026.html](https://www.globenewswire.com/news-release/2025/11/19/3191444/0/en/NVIDIA-Announces-Financial-Results-for-Third-Quarter-Fiscal-2026.html) 2. NVIDIA Q3 FY2026 earnings preview: Blackwell ramp to boost data centre sales, accessed November 19, 2025, [https://www.ig.com/en/news-and-trade-ideas/NVIDIA-Q3-FY2026-earnings-preview-251111](https://www.ig.com/en/news-and-trade-ideas/NVIDIA-Q3-FY2026-earnings-preview-251111) 3. NVIDIA Corp (NVDA) Discount Rate \- WACC & Cost of Equity \- Alpha Spread, accessed November 19, 2025, [https://www.alphaspread.com/security/nasdaq/nvda/discount-rate](https://www.alphaspread.com/security/nasdaq/nvda/discount-rate) 4. NVIDIA Announces Financial Results for Second Quarter Fiscal 2026, accessed November 19, 2025, [https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2026](https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2026) 5. Nvidia Earnings: Live Updates and Commentary November 2025, accessed November 19, 2025, [https://www.kiplinger.com/investing/live/nvidia-earnings-live-updates-and-commentary-november-2025](https://www.kiplinger.com/investing/live/nvidia-earnings-live-updates-and-commentary-november-2025) 6. Big Nvidia earnings today \- which way will the stock market swing? 5 key points to know, accessed November 19, 2025, [https://m.economictimes.com/news/international/us/big-nvidia-earnings-today-which-way-will-the-stock-market-swing-5-key-points-to-know/articleshow/125437254.cms?UTM\_Source=Google\_Newsstand\&UTM\_Campaign=RSS\_Feed\&UTM\_Medium=Referral](https://m.economictimes.com/news/international/us/big-nvidia-earnings-today-which-way-will-the-stock-market-swing-5-key-points-to-know/articleshow/125437254.cms?UTM_Source=Google_Newsstand&UTM_Campaign=RSS_Feed&UTM_Medium=Referral) 7. Nvidia earnings preview: NVDA stock prepares for crucial Q3 FY2026 results — here’s how Nvidia stock may r, accessed November 19, 2025, [https://m.economictimes.com/news/international/us/nvidia-earnings-preview-nvda-stock-prepares-for-crucial-q3-fy2026-results-heres-how-nvidia-stock-may-react-post-earnings/articleshow/125388456.cms](https://m.economictimes.com/news/international/us/nvidia-earnings-preview-nvda-stock-prepares-for-crucial-q3-fy2026-results-heres-how-nvidia-stock-may-react-post-earnings/articleshow/125388456.cms) 8. NVIDIA Q3 FY2026 earnings preview: Blackwell ramp to boost data centre sales, accessed November 19, 2025, [https://www.ig.com/sg/news-and-trade-ideas/NVIDIA-Q3-FY2026-earnings-preview-251111](https://www.ig.com/sg/news-and-trade-ideas/NVIDIA-Q3-FY2026-earnings-preview-251111) 9. NVIDIA Announces Financial Results for Third Quarter Fiscal 2025, accessed November 19, 2025, [https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-third-quarter-fiscal-2025](https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-third-quarter-fiscal-2025) 10. NVIDIA Announces Financial Results for Fourth Quarter and Fiscal 2025, accessed November 19, 2025, [https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-fourth-quarter-and-fiscal-2025](https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-fourth-quarter-and-fiscal-2025) 11. CFO Commentary on Second Quarter Fiscal 2026 Results, accessed November 19, 2025, [https://s201.q4cdn.com/141608511/files/doc\_financials/2026/Q226/Q2FY26-CFO-Commentary.pdf](https://s201.q4cdn.com/141608511/files/doc_financials/2026/Q226/Q2FY26-CFO-Commentary.pdf) 12. US stock futures today: Dow, S\&P 500, Nasdaq slip as AI stocks fall and Bitcoin breaks below $90,000, accessed November 19, 2025, [https://m.economictimes.com/news/international/us/us-stock-market-futures-today-dow-sp-500-nasdaq-slip-as-ai-stocks-fall-and-bitcoin-breaks-below-90000/articleshow/125408812.cms](https://m.economictimes.com/news/international/us/us-stock-market-futures-today-dow-sp-500-nasdaq-slip-as-ai-stocks-fall-and-bitcoin-breaks-below-90000/articleshow/125408812.cms) 13. US 10 Year Treasury Bond Note Yield \- Quote \- Chart \- Historical Data \- Trading Economics, accessed November 19, 2025, [https://tradingeconomics.com/united-states/government-bond-yield](https://tradingeconomics.com/united-states/government-bond-yield) 14. Market Yield on U.S. Treasury Securities at 10-Year Constant Maturity, Quoted on an Investment Basis (DGS10) | FRED, accessed November 19, 2025, [https://fred.stlouisfed.org/series/DGS10](https://fred.stlouisfed.org/series/DGS10) 15. NVIDIA Announces Financial Results for First Quarter Fiscal 2026, accessed November 19, 2025, [https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2026](https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2026) 16. NVIDIA Long Term Debt 2012-2025 | NVDA \- Macrotrends, accessed November 19, 2025, [https://www.macrotrends.net/stocks/charts/NVDA/nvidia/long-term-debt](https://www.macrotrends.net/stocks/charts/NVDA/nvidia/long-term-debt) 17. Prediction: This Will Be Nvidia's Stock Price 5 Years From Now \- Nasdaq, accessed November 19, 2025, [https://www.nasdaq.com/articles/prediction-will-be-nvidias-stock-price-5-years-now-0](https://www.nasdaq.com/articles/prediction-will-be-nvidias-stock-price-5-years-now-0) 18. Here's How Much Traders Expect Nvidia Stock To Move After Wednesday's Earnings, accessed November 19, 2025, [https://www.investopedia.com/here-is-how-much-traders-expect-nvidia-stock-to-move-after-wednesday-earnings-q3-fy2026-nvda-11851166](https://www.investopedia.com/here-is-how-much-traders-expect-nvidia-stock-to-move-after-wednesday-earnings-q3-fy2026-nvda-11851166)