Tech Stocks: What Is Inside the Technology Theme

The theme applies one test: is revenue driven primarily by software, semiconductors, internet platforms, IT services, or computing hardware, with the business model dependent on technology adoption rather than on a physical commodity or a regulated rate base?

The word doing the work is primarily. Almost every large company now runs on software, employs engineers, and describes itself as a technology business given the chance. A retailer with a good app is not a technology stock, because nothing about how it earns changes if software adoption accelerates. Drop that requirement and the theme absorbs half the market, which is the failure mode of the sector label itself rather than of any particular screen.

The second and more consequential choice is that the theme is organised by business model instead of by product category. Product category is how most technology lists are built and it explains almost nothing, because two companies making similar-sounding products can earn money in completely different ways. What predicts behaviour is how the revenue arrives: from advertisers and marketplace transactions on a platform with network effects, from renewing contracts at high gross margin, from units sold on a replacement cycle, or from someone else's capital budget. Those four models turn at different points, break for different reasons, and are valued on different things. For the general idea, see thematic investing.

One consequence is worth stating up front, because it is where this theme diverges from the funds people compare it to. Sector classification puts Alphabet and Meta under communication services and Amazon under consumer discretionary, so a technology sector fund holds none of them. This theme holds all three, because a company that designs its own accelerator silicon and operates its own data centers is a technology business whatever a taxonomy says.

The first business model in the sector does not sell software or hardware at all. It assembles an audience or a marketplace, makes that gathering place more valuable to each participant as more participants join, and then charges a third party for access to it: advertisers buying attention, sellers paying a take rate on transactions. Gross margins are high because the incremental user costs almost nothing to serve, and the moat is the network rather than the code. The revenue is also the most cyclical high-margin revenue in technology, because advertising and discretionary commerce budgets are cut early and restored late.

No further constituents sit in this layer. Three platform names is a deliberate limit rather than an omission, because the model is concentrated in a handful of companies at global scale and the theme would otherwise tip further toward advertising than it already does.

How this layer relates to the rest. Platforms are where the sector's cash is generated and where an enormous share of it is spent again. Their advertising and marketplace profits fund the capital budgets that the semiconductor layer sells into, and their app stores, ad systems, and clouds are the distribution channel much of the software layer reaches customers through. That makes a weak advertising market a slow-acting problem for chip orders, not just for the platforms themselves.

The second model sells a licence or a metered service that renews. Revenue is recognised over time rather than at the moment of sale, gross margin is structurally high because copying software costs nothing, and the health of the business is read through retention and expansion rather than through units shipped. That predictability is precisely what the market pays a rich multiple for, which makes valuation the standing risk in this layer rather than demand. The other structural fact is that software revenue is somebody else's expense line, so it is negotiated at renewal, tied to headcount, and slower to break than hardware but slower to recover too.

Five further constituents fill out the layer, each chosen for a different variation on recurring revenue rather than for scale.

• Oracle (ORCL). A database and enterprise applications veteran whose cloud infrastructure arm has taken on large AI training workloads, contracted well ahead of the capacity being built. It is in the layer as the case where software economics stop applying: to serve those contracts it is spending like an infrastructure company, which gives it a capital intensity and a financing profile that no other software name here carries.
• ServiceNow (NOW). A workflow automation platform that digitises IT, HR, and operations processes inside large organisations, with retention high enough that existing customers reliably spend more each year. It represents the layer at its most expensive: the market pays a premium multiple for that predictability, so the risk sits in the valuation rather than in the business breaking.
• Adobe (ADBE). Owns the Creative Cloud and Document Cloud franchises that professional design, marketing, and document workflows run on, converted from packaged licences to subscription years ago. It is the constituent where the generative-AI question is asked in reverse: for most of this theme AI is a demand driver, while here the open question is whether cheap generative tools erode a creative-software moat.
• Intuit (INTU). Owns TurboTax, QuickBooks, Credit Karma, and Mailchimp, selling financial software to consumers and small businesses rather than to enterprise IT departments. That customer base is why it is here: its revenue tracks small-business formation and tax season, which is a completely different demand driver from the enterprise budgets and capital spending that move everything else in the theme.
• Palantir Technologies (PLTR). Builds data integration and AI operations software for government agencies and large commercial customers, sold through a small number of substantial contracts rather than through a wide seat base. It is the layer's most concentrated holding by some distance, with government revenue that follows appropriation cycles and a valuation that leaves very little room for a disappointing quarter.

How this layer relates to the rest. Software is the layer that turns computing capacity into something a business will pay a subscription for, so it is the commercial justification for the capacity the layers below build. It runs on infrastructure the platform layer rents out and the semiconductor layer supplies, and it now competes for the same enterprise budget as the AI spending those layers are funded by, which is a tension inside the theme rather than a neat dependency.

The third model sells a physical object once. Revenue is recognised at the point of sale, gross margin is set by the bill of materials and by pricing power rather than by the cost of copying, and demand arrives in cycles tied to when the previous product wears out or stops feeling current. Hardware businesses are therefore read through unit volumes, average selling prices, and replacement intervals, none of which have an equivalent in the software layer. The best of them escape the cycle partially by converting the installed base into recurring revenue, which is why the interesting question about a hardware constituent is usually how much of its revenue is no longer hardware.

One further constituent occupies the layer, and it is here for the enterprise half of hardware rather than the consumer half.

• Cisco Systems (CSCO). Supplies the switching, routing, and network security equipment that corporate and data center networks are built from, increasingly sold bundled with software and subscription support, with the Splunk acquisition pushing further in that direction. It is the theme's mature hardware ballast: slower growing than anything else in the roster, paying a meaningful dividend, and driven by enterprise refresh cycles rather than by the AI capital spending that moves the semiconductor layer.

How this layer relates to the rest. Hardware is the endpoint that gives the rest of the sector somewhere to run, and it is the largest single customer of the semiconductor layer outside the data center. It also competes with the platform layer on the terms of distribution, since whoever controls the device controls the default. Its revenue is the most immediately sensitive to consumer and enterprise conditions of anything in the theme, because a replacement purchase can simply be postponed.

The fourth model sells into a capital budget rather than an operating one, and that single fact explains most of how it behaves. Chip demand is a derivative of someone else's decision to build, so orders arrive in waves, inventory in the channel amplifies both directions, and a design win locks in revenue years before it is recognised. Fixed costs are enormous, whether the company owns fabs or pays a foundry for capacity, so operating leverage runs hard in both directions: the same business can post exceptional margins in an upcycle and losses in a downcycle without anything about it having changed. This is the most volatile layer in the theme by a wide margin, and it is also where the theme's growth has recently come from.

Three further constituents complete the layer, each covering a part of the chip economy the names above do not reach.

• Advanced Micro Devices (AMD). Supplies data center and client CPUs alongside a growing line of AI accelerators, competing against a much larger incumbent in each market. It is in the layer as the second-source position: its case rests less on the market growing than on buyers wanting an alternative, which is a different exposure from owning the leader and does not necessarily move with it.
• Qualcomm Incorporated (QCOM). Designs the processors and modems in a large share of the smartphone market and licenses a wireless patent portfolio that earns royalties on handsets it does not supply. It covers the edge and mobile end of the layer, where demand follows the device replacement cycle rather than data center construction, with customer concentration and the prospect of large buyers designing their own parts as the standing risks.
• Arteris (AIP). Licenses the on-chip interconnect that ties the blocks of a modern system-on-chip together, earning licence fees and royalties from the chip designers rather than selling any silicon itself. It is by far the smallest constituent in the theme and is here for the IP licensing model, which earns from chip design activity across the industry, though a business of this size carries execution and liquidity risk that none of the larger names do.

How this layer relates to the rest. Everything above depends on this layer and this layer depends on everything above. Platforms fund capital budgets that become chip orders, hardware ships units containing them, and software eventually has to earn enough to justify the spending. The critical point for anyone holding the whole roster is that the flow runs the other way too: because the platform layer is now among the semiconductor layer's largest customers, the two are far more correlated than a diversified technology position is supposed to be.

This is the practical risk a reader of this page actually faces, and it is worth more attention than any individual entry above. A technology position is usually far less diversified than it appears, for two separate reasons that compound.

The first is arithmetic. The largest technology companies are the largest weights in the S&P 500, a larger share still of the Nasdaq-100, and the dominant holdings of every technology sector fund. Someone holding a broad index fund, a technology ETF, and four or five individual technology names is not holding three diversified layers. They are holding the same handful of companies at three different removes, with a combined weight that no single decision ever set. Nothing about that is exotic. It is the default outcome of buying the things that are easy to buy.

The second reason is newer and less widely noticed. These companies used to fail for different reasons: an ad recession hurt the platforms, a phone cycle hurt the device makers, an inventory correction hurt the chipmakers, and enterprise software carried on renewing through all of it. Several of the largest names now share one demand driver. Platform capital spending on AI compute is revenue for the semiconductor layer, and the semiconductor layer's results are read as confirmation that the platforms are right to keep spending. A decision by a small number of buyers to slow that spending would be felt across a large part of this roster at once, in the same quarter, for the same reason.

Naming the problem is more useful than pretending an eighteen-name list solves it. The parts of this theme that genuinely behave differently are the ones that look least exciting: analog chips sold into factories and cars, consumer tax software, seat-based enterprise applications, mature networking hardware. They are in the roster for that reason. The honest starting point is not which technology stock to add, it is how much technology you already own without having chosen it.

Read as a circuit rather than a ladder, the theme makes more sense. Platforms generate the sector's cash from advertising and transactions. A large share of that cash is committed to capital budgets, which become orders in the semiconductor layer. The chips end up in data centers and in devices, the devices reach an installed base, and software is what finally has to earn enough from all of it to justify the spending. Then the cycle starts again from the platforms.

The timing along that circuit is uneven, and that is what creates most of the dispersion inside the theme. The semiconductor layer is paid out of capital budgets, in advance of any return on them, which is why its recent results have been the strongest in the roster and also why they are not independent evidence that the spending was worthwhile. Software revenue is recognised across a contract term, so it lags in both directions. Hardware is recognised at the moment of sale and is the first thing a consumer postpones. Platform advertising revenue turns fastest of all.

The practical consequence is that the eighteen names do not move for one reason, but neither do they move for eighteen reasons. A weak industrial economy hits TXN while leaving the platforms untouched. A tax-season shift moves INTU and nothing else. A slowdown in data center construction would move NVDA, AVGO, AMD, ORCL, and the platform capital budgets funding all of them at the same time. Knowing which of those descriptions applies to the weights you have chosen is more useful than any ranking of the eighteen.

A membership test is only credible if it excludes things, and with a theme this broad the exclusions do more work than usual. These are the names people most often expect to find here, and the specific reason each one does not qualify.

• Cybersecurity software (CRWD, PANW, ZS, FTNT). They are subscription software businesses and would pass the test on economics alone, which is exactly why they are separated out. Their demand is driven by threat activity, breach disclosure, and compliance requirements rather than by technology adoption or capital budgets, so they behave differently from the rest of the software layer in precisely the conditions when that matters. They sit in the cybersecurity theme, where non-discretionary security spending is the thesis.
• The AI supply chain beyond chip design (TSM, ASML, AMAT, MU, ANET). Foundry, lithography, memory, and interconnect are the manufacturing base under the semiconductor layer, and their revenue tracks multi-year fab plans and cluster buildouts rather than the technology sector as a whole. Pulling them in would make this theme a leveraged position on data center construction. They sit in the AI infrastructure and semiconductors themes, which is where that exposure is the point.
• Payment networks and fintech (V, MA, PYPL). Payment processing looks like technology and runs on it, but index providers reclassified the large payment networks out of information technology and into financials, and the underlying business is interchange economics, credit exposure, and payments regulation. That is a financial services risk profile with a software delivery mechanism, not the other way round.
• Streaming and digital media (NFLX, DIS, SPOT). Subscription revenue is not the same qualification as a subscription software business. Their largest cost is content, which does not get cheaper at scale the way software does, and their competitive position depends on what they commission rather than on any technical advantage. The economics are media economics regardless of how the product is delivered.
• Tesla and technology-adjacent manufacturers. Software and silicon are genuinely central to what Tesla builds, but the company earns its money by manufacturing and selling vehicles, with factory utilisation, unit margins, and vehicle demand setting the results. Being technically sophisticated is not the inclusion test, or the theme would eventually include most of the industrials sector.

Three of those exclusions point at narrower Walnut themes where the exposure is the thesis rather than a side effect: the semiconductors theme reaches down into foundry, memory, and materials that this theme does not hold, the AI infrastructure theme narrows to companies whose revenue tracks training and inference spending, the enterprise software theme isolates the recurring-revenue layer on its own, and the cybersecurity theme holds security software separately because its budget behaves differently from every other software line.

Several constituents of this theme appear in those narrower themes as well, and that is by design rather than duplication. NVDA, AVGO, AMD, and TXN sit in the semiconductors theme. NVDA, AVGO, AMD, MSFT, GOOGL, AMZN, and ORCL sit in the AI infrastructure theme. MSFT, GOOGL, ORCL, NOW, and PLTR sit in the enterprise software theme. A company belongs wherever its business genuinely qualifies. What changes between themes is the surrounding roster and therefore what the position expresses: NVIDIA held next to Texas Instruments and Adobe is a technology position, while NVIDIA held next to a foundry and a lithography supplier is a position on data center construction. If one of those narrower cuts is what you actually want, the deeper guides are semiconductor stocks, AI stocks, and cloud computing stocks.

All 18 constituents, grouped by the business model they run rather than ranked, so the shape of the theme is visible in one place. 9 of them are covered in full above as the names that define their layer.

Read down the layer column rather than the ticker column. Three platforms, seven software businesses, two hardware names, and six semiconductor and infrastructure names is a very different portfolio from an equal-weighted list of eighteen technology companies, and the split is the theme's central design decision.

The passive route is a sector or index fund, and here the choice between them is unusually consequential because the three obvious candidates do not define technology the same way. VGT and XLK track the information technology sector, which means neither holds Alphabet, Meta, or Amazon at all, since classification puts those three elsewhere. XLK draws only from the S&P 500 while VGT covers a broader slice of the US market, so the smaller names differ even where the largest ones overlap. QQQ is not a sector fund at all: it tracks the Nasdaq-100, which is built from where a company is listed, so it holds the platform names and omits technology companies listed on the NYSE. Buying a technology fund without knowing which of those rules it follows is how people end up surprised by what they own.

What all three share is that the largest holdings dominate the result, so the diversification implied by a holdings count is smaller than it looks. That is the same concentration problem described above, arriving through a fund rather than through individual purchases.

A theme inverts the trade. You know exactly which names you hold, which business model each one represents, and what weight each carries, and you accept a narrower roster than a fund holds along with the work of maintaining it. Neither is automatically better. The fund is the simpler instrument, the theme is the more deliberate one, and plenty of people hold a broad fund as a core with a smaller deliberate tilt beside it.

A list of eighteen names is an input, not a portfolio. What turns one into the other is structure: which business models you want exposure to, what weight each name carries, and whether the concentration you end up with was chosen or inherited.

• Count what you already own first. With this theme more than any other, the honest first step is adding up how much of the roster arrives through the index funds you already hold. Everything after that is a decision about how much more you want, not whether to own any.
• Decide the model mix before the names. The split between platforms, software, hardware, and semiconductors changes the character of the position far more than swapping one chipmaker for another.
• Set target weights that sum to 100. Equal weighting across eighteen names is a choice, and so is concentrating in a few. Both are defensible. Not deciding is what leaves you concentrated by accident after one layer runs.
• Watch for correlation, not just count. Adding a fourth name that depends on data center capital spending adds a ticker without adding much independence. The names that diversify this theme are the ones with a different customer.
• Frame it against the S&P 500. A technology tilt on top of an index that is already technology-heavy has to earn the extra concentration, and comparing against the benchmark is how you find out whether it has.
• Revisit as weights move. Dispersion inside this theme is wide enough that target weights drift quickly, and drift is how a deliberate tilt becomes an accidental bet.

This is what Walnut is built for. You describe the thesis, the AI assistant proposes constituents and weights you can edit, the portfolio tracks as one performance line against the S&P 500, and you place trades you approve yourself at your own broker. Walnut is informational and does not tell you which stocks to buy.

For the companion view of which technology names are most widely held and discussed, see best tech stocks. For the narrower cut most people actually mean when they say technology right now, see AI stocks.

The technology theme is 18 companies across four business models, and the models are the whole idea. Alphabet, Meta, and Amazon monetise attention and transactions on networks that get more valuable as they grow. Microsoft, Salesforce, Oracle, ServiceNow, Adobe, Intuit, and Palantir sell recurring revenue at high gross margin against somebody else's budget line. Apple and Cisco sell objects on a replacement cycle and try to convert the installed base into something that renews. NVIDIA, Broadcom, Texas Instruments, AMD, Qualcomm, and Arteris sell into other companies' capital budgets, with all the operating leverage that implies.

Understood as a flat list of eighteen technology stocks, the theme looks like a diversified sector holding. It is not, and that is the single most useful thing to take from this page. The largest names here already sit at the top of the index funds most people own, and several of them have come to share one demand driver, so the real question is not which technology stock to add but how much technology you already hold and which parts of it genuinely behave differently. Nothing here is a recommendation, and Walnut is not an investment adviser.