NVDA (NVIDIA): Themes, ETFs, and Basket Ideas
This isn't a NVIDIA stock-price page, those exist already. Instead, this guide answers: which themes does NVIDIA fit in, which ETFs hold it most heavily, what other stocks are similar, and how would you size NVDA inside a basket. The angle is portfolio thinking, not price-checking.
What does NVIDIA do?
NVIDIA designs graphics processing units (GPUs) and the software stack that runs on them. The company operates across four reporting segments. Data Center sells GPUs to cloud providers and AI labs for training and running large language models; this is now around 85% of revenue. Gaming covers GeForce consumer GPUs for PC gaming, NVIDIA's original core market. Professional Visualization sells workstation GPUs for design, simulation, and content creation. Automotive ships the DRIVE platform for assisted and autonomous driving.
The company also builds the CUDA software platform that lets developers write code that runs on NVIDIA GPUs. CUDA has become the de facto target for AI frameworks like PyTorch, TensorFlow, and JAX, which is much of the reason NVIDIA's competitive position has been so durable. Founded in 1993, headquartered in Santa Clara, California, led by co-founder and CEO Jensen Huang.
Where is NVIDIA heading?
Four major themes drive NVIDIA's strategy.
1. Continued AI infrastructure dominance.
The Hopper generation (H100, H200) trained the current frontier of large language models. The Blackwell generation (B100, B200) is shipping in volume with massive demand from hyperscalers (Microsoft, Amazon, Google, Meta, Oracle). NVIDIA has signaled the Rubin platform for 2026, extending the roughly annual cadence of new architectures. Each generation has been roughly 2 to 5x more performant per watt than the last, which keeps the upgrade cycle aggressive.
2. Beyond hyperscalers: sovereign AI and enterprise.
NVIDIA has been actively expanding past its top four to five hyperscaler customers. National governments are building “sovereign AI” data centers using NVIDIA hardware (the company has cited dozens of country-level deals). Enterprises are increasingly deploying NVIDIA-powered AI on-premises or in dedicated clouds rather than relying solely on hyperscaler-hosted models.
3. Adjacent revenue layers.
Automotive (DRIVE for self-driving stacks at Mercedes, Volvo, Hyundai, and others), robotics (Isaac platform for industrial humanoid robotics), and digital twins (Omniverse for industrial simulation). These layers are small today relative to data center, but they represent long-duration optionality if AI permeates these industries the way many expect.
4. The CUDA moat.
NVIDIA's competitive position depends as much on its software ecosystem as its hardware. CUDA has been the standard target for AI frameworks for over a decade. Switching off CUDA is technically possible (AMD has ROCm, the major frameworks have alternative backends), but it costs real engineering time and risk. Customers who have invested in CUDA-targeted code keep coming back to NVIDIA hardware.
Risks worth tracking: customer concentration (the top 5 hyperscalers represent roughly 50% of revenue), competition from custom AI silicon at the hyperscalers (Google TPU, AWS Trainium, Microsoft Maia, Meta MTIA, plus AMD's MI300X and MI400), and geopolitical (US export restrictions to China have already cut a meaningful revenue stream).
Earnings and valuation (approximate, early 2026)
A simple financial snapshot. These are approximations as of early 2026 and refresh quarterly; for current figures see NVIDIA's investor relations page or your broker.
- Revenue (FY2026 ending Jan): approximately $130 billion, having grown 100%+ in recent quarters
- Operating margin: ~63%, exceptionally high for a hardware company (the CUDA moat and pricing power show up here)
- Net income: approximately $70 billion
- EPS (TTM): ~$2.80
- P/E (TTM): ~50x
- Price to sales: ~28x
- Dividend yield: <0.1% (a token quarterly dividend; capital is reinvested in R&D and capex)
- Free cash flow: ~$60 billion annually
- Customer concentration: top 4 hyperscaler customers account for roughly 50% of revenue (Microsoft, Amazon, Meta, Google)
The P/E of around 50x is the headline most investors focus on. The reason the market tolerates it is the combination of triple-digit revenue growth, 60%+ operating margins, and a software moat (CUDA) that competitors keep failing to match. The risk is that if hyperscaler capex slows or the AI buildout decelerates, the valuation compresses very quickly.
Themes NVIDIA belongs to
NVIDIA is unusual in how many distinct theses point at it. AI infrastructure is the obvious one, but the stock sits at the intersection of semiconductors, the CUDA software moat, autonomous vehicles, robotics, and mega-cap growth. Each card is one investment theme NVDA credibly belongs to, with one-sentence rationale:
AI infrastructure
The defining stock of the AI era. NVIDIA's GPUs train and run essentially every modern large language model; the company is the picks-and-shovels play for generative AI.
Semiconductors
Largest US semiconductor company by market cap. Designs chips; manufacturing is outsourced primarily to TSMC.
Data center power and cooling
Massive GPU demand is driving an electrical-grid + cooling buildout. NVIDIA is the upstream beneficiary; power names like VRT, ETN are downstream.
Data center / hyperscaler capex
Data-center revenue (sold to Microsoft, Amazon, Google, Meta, Oracle) is now the bulk of the business. Direct beneficiary of hyperscaler AI capex.
Generative AI
Hopper (H100) and Blackwell (B100/B200) architectures power the training of frontier models. Demand is multi-year, capacity-constrained.
Mag 7 / mega-cap tech
Joined the Magnificent 7 narrative once it crossed $1T market cap in 2023. Now one of the top three US companies by market cap.
Trillion-dollar club
Crossed $1T in 2023, $2T in 2024, $3T in 2024, fastest path to $3T in history. Sits in the very small group of mega-mega-caps.
GPU computing
Defined the modern GPU computing category. NVIDIA's lead is hardware + software (CUDA + cuDNN + Triton), a moat built over 20 years.
CUDA ecosystem
CUDA is the proprietary software platform that makes NVIDIA's chips usable. Every AI framework targets CUDA first. The lock-in is real and durable.
Gaming
Original core market, GeForce GPUs. Still meaningful revenue, but now a small fraction of total compared to data center.
Autonomous vehicles
DRIVE platform supplies the compute for self-driving stacks at Mercedes, Volvo, Hyundai, and many others. Long-tail option value.
Robotics
Isaac platform + Omniverse for industrial digital twins. Emerging revenue stream as humanoid robotics scales.
High-growth large-caps
Revenue growth of 100%+ in some recent years, extraordinary for a company of NVIDIA's size. Quality-growth factor.
Semiconductor equipment ecosystem (adjacent)
Designs chips made by TSMC, using equipment from ASML, Applied Materials, KLA. NVIDIA sits at the top of a long supply chain it doesn't own.
ETFs that hold NVDA
You can get NVIDIA exposure many ways. Semi-focused ETFs concentrate it heavily; tech-sector ETFs hold it as a top weight; broad-index ETFs hold it as one of the top three positions:
| ETF | Name | % in NVDA | Expense ratio | |
|---|---|---|---|---|
| SMH | VanEck Semiconductor ETF | 21.4% | 0.35% | |
| XLK | Technology Select Sector SPDR | 16.8% | 0.09% | |
| VGT | Vanguard Information Tech ETF | 14.5% | 0.10% | |
| IGM | iShares Expanded Tech Sector ETF | 13.1% | 0.41% | |
| MGK | Vanguard Mega Cap Growth ETF | 11.7% | 0.07% | |
| VUG | Vanguard Growth ETF | 11.2% | 0.04% | |
| SOXX | iShares Semiconductor ETF | 10.9% | 0.35% | |
| QQQ | Invesco QQQ Trust (Nasdaq-100) | 8.0% | 0.20% | |
| VOO | Vanguard S&P 500 ETF | 6.2% | 0.03% |
A common pattern: hold SMH for diversified semi exposure (and an indirect NVDA position), plus NVDA directly to express higher conviction on the company specifically. Walnut surfaces the combined exposure so you don't end up with 40% of your portfolio in one chip company by accident.
Stocks similar to NVIDIA
“Similar to NVIDIA” depends on which theme you care about, AI infrastructure peers, GPU competitors, semiconductor supply-chain peers, AI data-center beneficiaries:
AI infrastructure peer, custom AI ASICs for hyperscalers, plus networking silicon. Often paired with NVDA in AI baskets.
Direct GPU competitor, MI300X targets AI training. Smaller share but a real second source for hyperscalers.
The fab that physically manufactures NVIDIA's chips. Critical supply-chain peer; geopolitical risk pairs.
Makes the EUV lithography machines TSMC uses to produce NVIDIA's chips. Two-step upstream from NVDA.
Lagging in AI accelerators (Gaudi has not gained share). Different position, more CPU than GPU, but in the same broad semi category.
Largest NVIDIA customer (Azure). Owns the AI infrastructure stack on top of NVIDIA's chips.
Networking gear inside AI data centers, sells the high-speed switches that connect NVIDIA GPU clusters.
Cooling and power for AI data centers, direct beneficiary of NVIDIA-driven AI capex.
How to invest in NVIDIA
NVDA trades on Nasdaq at $0 commission across every major US broker. Fractional shares are supported at Robinhood, Fidelity, Schwab, Public, M1, and others.
Patterns for holding NVIDIA thoughtfully:
- Through a broad ETF: VOO or VTI gives you a few percent NVDA inside a diversified large-cap fund. Simple, low concentration.
- Through a semi-focused ETF: SMH (~21% NVDA) or SOXX (~11% NVDA) expresses the broader semis thesis with NVIDIA as the largest position.
- Direct + ETF: Own NVDA directly to express specific conviction, plus a broad ETF for the rest of your portfolio. Watch combined concentration carefully, NVDA can swing 5-10% in a day.
- In a thematic basket: Build a Walnut basket around “AI infrastructure”, “hyperscaler capex beneficiaries”, or “AI power buildout”. NVDA is usually 20-30%, leave room for AVGO, MSFT, AMZN, VRT, ANET, and others depending on the thesis.
Build a basket around NVDA with Walnut
Use NVIDIA as one constituent in an AI infrastructure or semiconductor basket. Walnut's AI proposes the rest of the basket (AVGO, AMD, TSM, MSFT, ANET, VRT, etc.) with target weights, and tells you when the overall basket is functionally just a NVDA bet in disguise.
FAQ
What is NVIDIA's ticker symbol?
+
NVDA, listed on Nasdaq. Officially NVIDIA Corporation. Founded 1993 by Jensen Huang, Chris Malachowsky, and Curtis Priem, headquartered in Santa Clara, California. Trades during US market hours, available at every major US brokerage with $0 commission.
Who are NVIDIA's main competitors?
+
Several categories. Direct GPU competition: AMD (MI300X today, MI400 series coming), Intel (Gaudi accelerators, though they have struggled to gain share). Custom AI silicon from the hyperscalers themselves: Google TPU, AWS Trainium and Inferentia, Microsoft Maia, Meta MTIA. NVIDIA holds roughly 90% market share of AI training accelerators today; the real moat is the CUDA software ecosystem, not just the hardware. Switching off CUDA is technically possible but expensive in engineering effort, which is why competitors take years rather than quarters to gain meaningful share.
What is NVIDIA's P/E ratio?
+
Approximately 50x trailing twelve months as of early 2026. That's elevated compared to the S&P 500 (around 22x), but supported by triple-digit revenue growth in recent quarters. The premium prices in the assumption that AI infrastructure demand from hyperscalers continues for several more years.
Why is NVIDIA stock so expensive?
+
High P/E reflects extraordinary growth: revenue more than doubled in 2024 and continued growing strongly into 2026. The market is pricing in continued AI capex from major hyperscalers (Microsoft, Amazon, Meta, Google, Oracle), plus the growing pipeline of sovereign AI deals. The valuation is reasonable if AI infrastructure demand persists; it compresses quickly if hyperscaler capex slows. The CUDA software moat is the structural reason NVIDIA can sustain pricing power even as competitors catch up on hardware.
Who owns the most NVIDIA stock?
+
Major institutional holders include Vanguard (around 9% of shares outstanding), BlackRock (around 7%), and FMR / Fidelity. CEO Jensen Huang personally owns around 3.5%, which is unusually high for a company NVIDIA's size and represents a meaningful alignment between founder and shareholders. Insider ownership overall is around 4%, mostly concentrated in Huang.
What does NVIDIA do?
+
NVIDIA designs the graphics processing units (GPUs) that have become the standard compute platform for AI training and inference. Its data-center business sells GPUs to the major cloud providers (Microsoft Azure, AWS, Google Cloud, Oracle, Meta) and to AI labs (OpenAI, Anthropic, xAI). It also has a gaming GPU business (GeForce), an automotive platform (DRIVE), and a robotics platform (Isaac). Data center is now the dominant revenue driver.
Why is NVIDIA so important for AI?
+
Two reasons: hardware and software. NVIDIA's GPUs (H100, H200, B100/B200) are by a wide margin the most performant chips for training large language models. And CUDA, NVIDIA's proprietary software layer, is what every major AI framework (PyTorch, TensorFlow, JAX) targets first. Competitors have to match both the chip and the software ecosystem; that's a real moat.
Which ETFs have the most NVIDIA exposure?
+
Semi-focused ETFs hold NVDA heaviest: SMH (~21%), SOXX (~11%). Broader tech-sector ETFs are next: XLK (~17%), VGT (~14%), MGK (~12%). Broad-index ETFs hold less but still meaningful: QQQ (~8%), VOO (~6%). If you want concentrated NVDA exposure through a fund, SMH is the most direct.
Which thematic baskets typically include NVIDIA?
+
NVIDIA is the centerpiece of any AI infrastructure basket. It also fits in semiconductor baskets, mega-cap tech, growth, GPU computing, autonomous-vehicle compute, and hyperscaler-capex baskets. Walnut's AI can build any of these in conversation and tell you when adding NVDA on top of an ETF that already holds it heavily creates concentration risk.
How much of QQQ is NVIDIA?
+
Approximately 8.0% as of early 2026, NVIDIA is typically the third-largest holding in QQQ, behind Microsoft and Apple. Buying QQQ gives you meaningful but not dominant NVIDIA exposure.
How much of VOO is NVIDIA?
+
Approximately 6.2% as of early 2026, NVIDIA is typically the third-largest holding in VOO, in the same range as Apple and Microsoft. Day-to-day shifts in market cap reorder the top three.
Does NVIDIA pay a dividend?
+
Yes, but it's minimal, a token quarterly dividend yielding well under 0.1%. NVIDIA prioritizes reinvesting cash flow into R&D and capex; dividends are a small share-return mechanism alongside buybacks.
Is NVIDIA in the S&P 500?
+
Yes. NVIDIA is one of the largest constituents of the S&P 500 (top three by weight). It joined the index in 2001.
What's the difference between NVIDIA and AMD?
+
Both make GPUs, but NVIDIA leads decisively in AI accelerator share, roughly 90%+ of the market. AMD's MI300X has won some hyperscaler workloads (especially at Microsoft and Meta) and is a real second source, but it's a distant second. NVIDIA also has the CUDA software moat that AMD's ROCm doesn't fully match.
What is CUDA and why does it matter?
+
CUDA is NVIDIA's parallel computing platform and programming model, the software layer that lets developers write code that runs on NVIDIA GPUs. Over 20 years, it has become the default target for AI frameworks. Switching off NVIDIA hardware also means switching off CUDA, which is technically possible but expensive in engineering effort. The CUDA lock-in is one of the most-cited reasons NVIDIA's competitive position is durable.
Should I buy NVIDIA directly or through SMH?
+
Depends on conviction. Buying NVDA directly gives you full single-stock exposure. SMH gives you ~21% NVDA inside a diversified semi-ETF, you get the AI thesis but also exposure to TSM, AVGO, AMD, ASML. SMH is the simpler way to express “I believe in semis broadly”; direct NVDA expresses “I have a specific view on NVIDIA”. Many investors do both, sized deliberately.
How concentrated is the AI infrastructure thesis in NVIDIA?
+
Very. A naive AI infrastructure basket built today would be heavily NVIDIA, by market cap, NVDA is several times larger than AVGO, AMD, TSM, ANET, or ASML. A balanced AI infrastructure basket usually caps NVDA at 20-30% to leave room for power (VRT, ETN), networking (ANET), and software (MSFT, ORCL). Walnut's AI will warn you when a basket is functionally just a NVIDIA bet.
Stats, ETF weights, and market-cap figures are approximations as of early 2026 and refresh quarterly. For live price and current data, see your broker or connect one to Walnut. Walnut is informational, not investment advice.