AI & Infrastructure CapexAI 与基础设施资本开支
NVIDIA Corporation
NVDA 7Price and target zones价格与目标区间
What the company does公司做什么
NVIDIA designs chips and software for artificial intelligence and computer graphics. It does not own factories; TSMC builds its chips. Most of its revenue now comes from data centres that train and run AI models, not from the gaming chips that built the company.英伟达（NVIDIA）设计用于人工智能和计算机图形处理的芯片和软件。公司本身不拥有工厂，芯片由台积电（TSMC）代工生产。如今公司的大部分营收来自训练和运行AI模型的数据中心业务，而不是让公司起家的游戏芯片。
Earns from营收来源 chip and system sales to cloud companies, AI labs, and computer makers, plus software and networking gear that ties those chips together.向云计算公司、AI实验室和电脑制造商销售芯片与整机系统，此外还销售把这些芯片连接起来的软件和网络设备。
Read the full description查看完整介绍
NVIDIA Corporation operates as a data center scale AI infrastructure company in the United States, Taiwan, China, Hong Kong, Europe, and internationally. It operates through Compute & Networking, and Graphics segments. The Compute & Networking segment provides data center accelerated computing and networking platforms and artificial intelligence solutions and software, and automotive platforms and autonomous and electric vehicle solutions, including software. The Graphics segment offers GeForce GPUs for gaming and PCs; Quadro/NVIDIA RTX GPUs for enterprise workstation graphics. The company's products are used in gaming, professional visualization, data center, and automotive markets. It sells its products to original equipment manufacturers, original device manufacturers, system integrators and distributors, independent software vendors, cloud service providers, add-in board manufacturers, distributors, automotive manufacturers and tier-1 automotive suppliers, and other ecosystem participants. NVIDIA Corporation was incorporated in 1993 and is headquartered in Santa Clara, California.
Revenue breakdown营收构成 share of revenue按营收占比
NVIDIA's overwhelming growth engine, up 117% from a year ago. Cloud companies, AI labs, and large enterprises buy these systems to train and run AI models. A small number of very large customers drive most of this revenue, and NVIDIA has started guaranteeing financing for some of them, which is the main risk to watch.英伟达增长最迅猛的引擎，同比增长117%。云计算公司、AI实验室和大型企业购买这些系统来训练和运行AI模型。少数几个大客户贡献了这项业务的大部分营收，英伟达也开始为其中一些客户提供融资担保，这是需要重点关注的主要风险。
- Data-centre GPUs (Blackwell, Vera Rubin)数据中心GPU（Blackwell、Vera Rubin） — The processor that does the heavy math behind training and running AI models. Each new generation is faster than the last, which keeps customers upgrading.负责AI模型训练和运行背后繁重运算的处理器。每一代新产品都比上一代更快，促使客户不断升级换代。
- NVLink and networkingNVLink与网络设备 — The wiring and switches that connect thousands of GPUs so they work as one machine. AI training needs this speed, not just fast chips.把数千个GPU连接起来、使其协同运作如同一台机器的线路和交换设备。AI训练需要的正是这种互联速度，而不仅仅是单颗芯片的算力。
- CUDA softwareCUDA软件 — The programming toolset every major AI program is written against. Switching to a rival chip means rewriting years of software, which is the deepest part of the moat.几乎所有主要AI程序都基于此编写的编程工具集。换用对手芯片就意味着要重写多年积累的软件，这是护城河中最深的一层。
The smaller, older part of NVIDIA's business, now a rounding error next to Data Center but still growing (up 27% from a year ago). It is more cyclical than Data Center because it depends on consumer and auto demand rather than one continuous AI buildout.英伟达业务中规模较小、历史更久的部分，如今相较数据中心业务已显得微不足道，但仍在增长（同比增长27%）。相比数据中心业务，这部分业务周期性更强，因为它依赖消费者和汽车需求，而非持续不断的AI建设浪潮。
- GeForce / RTX gaming GPUsGeForce / RTX 游戏GPU — Graphics chips for gaming PCs. This is the business NVIDIA was built on.面向游戏电脑的图形芯片，也是英伟达起家的业务。
- Professional visualization (RTX workstation, Omniverse)专业可视化（RTX工作站、Omniverse） — Chips and software for designers and engineers who build 3D models and simulations.面向设计师和工程师、用于构建3D模型和仿真的芯片与软件。
- Automotive (DRIVE)汽车（DRIVE） — Chips and software for self-driving and driver-assist systems, sold to carmakers.销售给汽车厂商、用于自动驾驶和驾驶辅助系统的芯片与软件。
Company facts公司资料
- Founded成立时间
- 1993
- Headquarters总部
- Santa Clara, California
- Employees员工人数
- 42,000
- Industry行业
- Semiconductors
- Country国家
- United States
Investment thesis投资逻辑 from the watchlist来自关注列表
🔄 A2026-09-10 — Hugging Face ($12.93B) + MediaTek $3.5B: ACCUMULATE [6.5]→[7.0]. Moat-positive (CUDA into open-model distribution) but MediaTek deepens the circular-financing flag; China H20 reversal was old news. FV $210-290, entry $195-215, trim 32x fwd. Buy near zone, don't chase. ——
Red flags
- The reason it's ACCUMULATE not BUY — a real, growing, not-yet-acute circular-financing risk the +8.7% print-day pop did not price: net margin now exceeds operating margin on non-cash equity marks, a +15-day DSO extension NVIDIA itself attributes to financing customer growth, and a $108.5B guarantee book explicitly built to backstop a customer whose "balance sheet can't support" its own buildout. FV $210-290. Entry $195-215 (add on a pullback, don't chase). Trim 32x fwd (current-FY, non-GAAP). Recheck Q3 FY27 ~11/18
Watch
- Same AI-capex bucket as AVGO/KLAC — size the bucket, not the name
Would upgrade on
- ANALYZED 8/28 — ACCUMULATE, first file. The seller at the center of the buildout. Emphatically a great business — 75% gross margin, 85% ROIC, Data Center revenue +117% YoY, a real multi-generation CUDA/NVLink software moat the hyperscalers' own custom silicon hasn't dented at the frontier (evergreen test passes). But the market has mostly priced it: the screener's "15x forward" that would scream BUY is a FY2028 number two years out ([[pitfall-vendor-forward-eps-is-the-wrong-fiscal-year]]); the true current-FY multiple is ~24-25x, full for a mega-cap even against 100%+ growth, and sits inside — not below — the $210-290 FV band
🔄 A2026-09-10 —— 收购 Hugging Face（$12.93B）+ MediaTek 交易 $3.5B：增持 [6.5]→[7.0]。 护城河利好（CUDA 延伸到开源模型分发），但 MediaTek 交易加深了循环融资的担忧；中国 H20 禁令逆转是旧闻。 公允价值 $210-290，入场 $195-215，减持 32x fwd。 接近区间可买，不要追高。 —— 🟢 8/28 分析——增持，首份报告。 处于这轮建设中心的『卖方』。 毫无疑问是一门好生意——毛利率 75%，ROIC 85%，数据中心营收同比 +117%，跨代的 CUDA/NVLink 软件护城河尚未被超大规模厂商的自研芯片撼动（常青测试通过）。 但市场已基本定价：筛选器上的『15x 前瞻』其实是两年后的 FY2028 数字；真实当年倍数约 24-25x，落在 $210-290 公允价值区间之内而非之下。 🚩 一个真实、正在扩大、尚不急迫的循环融资风险：净利率因非现金股权计价已超过营业利润率，NVIDIA 把 +15 天的应收账款周转天数延长归因于为客户增长提供融资，还有一本 $108.5B 的担保账本，为一个『资产负债表无法支撑』自身建设的客户兜底。 公允价值 $210-290，入场 $195-215（回调时加仓，不要追高），减持 32x fwd（当年，非公认会计准则）。 FY27 第三季度约 11/18 复查。 ⚠️ 与 AVGO/KLAC 同属 AI 资本支出仓位——按仓位定规模。 A2026-08-28
Historical results历史业绩 past fiscal years — these values do not change过去财年——这些数值不会变动
Profit margins利润率 the share of revenue that stays as profit营收中留存为利润的比例
of every $1 of revenue每 $1 营收中的占比
- Return on equity净资产收益率
- 117%
- Revenue growth营收增长
- +106%
- Dividend yield股息率
- 0.44%
Competitors竞争对手
Why customers stay客户为什么留下
CUDA software and NVLink networking took over a decade to build, and every major AI lab writes its code against them. A rival chip has to be much better to justify the cost of rewriting that software, which is why NVIDIA's data-centre lead has held even as well-funded rivals compete hard on chip specs alone.CUDA软件和NVLink网络技术历时十多年才打造完成，几乎所有主要AI实验室都基于它们编写代码。竞争对手的芯片必须显著更优，才能让客户觉得值得为此重写这些软件——这也是为什么即便资金雄厚的对手在芯片规格上正面竞争，英伟达在数据中心领域的领先地位依然稳固。
Related companies相关公司 🔴 AI capex — SELLERS into the buildout
Datacenter buildout pace. SOX −28.6% (peak 6/22 → 7/29), $3.3T of sector cap gone in 5 weeks — every name fell regardless of its own results.
Size the bucket, not the name. (ORCL's own analysis says this.)