Cloud Stocks: What Is Inside the Cloud Computing Theme

The theme applies one test: does meaningful revenue come from delivering computing resources over the internet? In practice that means hyperscale cloud infrastructure, cloud databases and applications, cloud data and analytics platforms, and the networking and edge layers built on top of the cloud.

The words doing the work are delivering computing. Nearly all enterprise software now runs in a data center somewhere, so hosting is not a distinguishing feature of anything. A company that sells seats of an application is selling software, and its revenue moves with headcount, budgets, and renewal cycles. A company that sells computing sells a metered resource, and its revenue moves with how much customers run. Drop that distinction and the theme quietly becomes a list of enterprise software vendors, which is the failure mode of most screens carrying the cloud label.

The second structural choice is that the theme spans layers rather than picking one. Owning only the hyperscalers is a position in three mega-caps whose cloud divisions are a minority of what moves them. Owning only the specialists is a concentrated bet on premium-multiple software with the hyperscalers competing directly against every one of them. Holding both means the theme contains the landlord and the tenant at once, which is unusual and is the reason the roster behaves the way it does. For the general idea, see thematic investing.

Three companies own most of the world's rentable computing capacity, and none of them is a cloud company. Cloud is a division inside a search advertising business, a division inside a retailer, and a division inside a software licensing franchise. That is the single most important thing to understand about this theme, because it means the biggest, most obvious, most widely held cloud names are also the most diluted exposure to cloud in the entire roster. You buy them for the segment and you get the whole company, including the parts that have nothing to do with renting compute and that usually decide the share price.

How this layer relates to the rest. Everything above this layer runs on it and pays rent to it, which makes the relationship uncomfortable in a way worth naming. The data platforms and delivery networks in the layers above are customers of these three companies and competitors of them at the same time, because each hyperscaler ships its own data warehouse, its own content delivery, and its own security products alongside the raw capacity it rents to the companies selling those things independently.

Renting a server solves nothing on its own. The layer above hyperscale infrastructure is where the customer's data lives and where their applications run: databases, data warehouses, analytics engines, and the enterprise applications built on top of them. This is the layer with the real switching costs, because moving a workload between clouds is an inconvenience while moving a company's accumulated data and the queries written against it is a project. It is also the layer where the theme's economics are cleanest, since these companies sell the thing they were founded to sell rather than a division of something else.

How this layer relates to the rest. This layer is what turns rented capacity into a recurring bill. It depends on the infrastructure layer for the machines and, more awkwardly, competes with it for the customer, because every hyperscaler sells a first-party alternative to what these companies offer. It supplies the delivery layer below with traffic worth accelerating and protecting.

Between the person using an application and the data center running it sits a layer almost nobody buys deliberately and almost every internet business pays for: content delivery, traffic routing, attack mitigation, secure access, and increasingly code that executes at the network edge rather than at an origin server. This layer is included because its demand is driven by traffic and by the number of places a company needs to be fast and protected, which is a different variable from how much compute anyone rents. It is the part of the theme least dependent on which hyperscaler wins.

How this layer relates to the rest. This layer is the theme's hedge against platform concentration. It is paid for traffic regardless of where that traffic terminates, so it grows with internet activity rather than with any one platform's share, and its edge-compute products push execution outward at the margin, which is a mild counter-position to the centralisation the layer above it depends on.

This is the most important change in the theme, and it is easy to miss because the constituents have not changed while the character of the largest ones has. For most of the past two decades, the investment case for cloud rested on an asymmetry. The customer stopped buying servers and started paying a metered bill, converting a capital expense into an operating one. The provider spent heavily up front, then earned high incremental margins on capacity it had already built. Compared with the on-premise software and hardware model it replaced, cloud looked like the capital-light end of technology, and plenty of people bought it for exactly that reason.

The AI buildout inverted that at the provider. Data centers full of accelerators, the power to run them, and the cooling to keep them alive are expensive in a way conventional cloud capacity was not, and the hyperscalers have committed to that spending at a scale that shows up as one of the largest capital programs in the market. The consequence for a holder of this theme is not a narrative point, it is an accounting one. Capital spending lands on the cash flow statement now and on the income statement later as depreciation, so free cash flow and reported margins both look different from the software-like profile the theme used to offer. Reported profitability now depends partly on assumptions about how long that equipment remains useful, which is a different kind of judgment from the ones software revenue involves.

This is also where the theme touches the AI infrastructure theme without being it. The same spending is a revenue line for the companies that sell accelerators, networking, and fabrication capacity, and a cash outflow for the companies here. If you own both themes, you own both sides of the same transaction, and the overlap is worth holding deliberately rather than discovering later. The AI theme's composition guide covers the sellers in detail, so this page does not repeat it.

The final point is that the change is not uniform across the roster. Snowflake and Cloudflare remain far less capital-intensive than the hyperscalers, and Oracle is carrying the heaviest relative burden of all because it is building comparable capacity from a smaller base. So a reader who bought cloud for asset-light compounding should know that the description still fits parts of this theme and no longer fits its three largest members. That is a change in what the position is, not a verdict on whether it is a good one.

Read top to bottom, the theme is a landlord, a tenant, and a doorway. The hyperscale layer owns the machines and rents them by the unit. The data and platform layer runs the workloads and the databases that make renting worth doing, and it is where the customer's switching costs actually accumulate, because moving a workload is an inconvenience while moving a decade of accumulated data and the queries written against it is a project. The delivery layer sits between the end user and both of them, paid for traffic no matter where that traffic terminates.

What makes this roster unusual is that the layers are not only dependent on each other, they compete with each other. Every hyperscaler sells a first-party data warehouse against Snowflake, a content delivery and edge product against Cloudflare, and a database against Oracle, while simultaneously renting all three of them the infrastructure they operate on. There is no clean supplier-and-customer chain here, and pretending otherwise would misdescribe the theme.

The practical consequence is that the six names do not move for one reason, but they also do not diversify each other as much as six names suggest. A weak advertising quarter moves GOOGL and leaves SNOW untouched. A round of customer cost-optimisation shows up immediately in consumption-billed revenue and barely at all in a licensing renewal. And a decision by the hyperscalers to compete harder on first-party data and delivery products would be good for the largest constituents and bad for the smallest at the same time, which is the one scenario where holding both ends of the theme is a genuine hedge rather than a doubled bet.

A membership test is only credible if it excludes things. These are the names people most often expect to find here, and the specific reason each one does not qualify.

• Salesforce and ServiceNow. Both deliver their products from the cloud, and neither is a cloud company under this test. Their revenue is subscription software sold by seat and by module to enterprise buyers, so it tracks software budgets and renewal cycles rather than how much computing anyone consumes. Being hosted somewhere is a delivery detail, not a business model. They sit in the enterprise software theme, where that subscription economics is the thesis.
• NVIDIA and the semiconductor supply chain. They sell the equipment this theme's constituents buy. Their revenue is recognised from capital budgets while data centers are being built, which is a capital-goods exposure on an entirely different timetable from metered usage revenue. That is the AI infrastructure theme's subject, not this one's.
• CrowdStrike, Palo Alto Networks, and Zscaler. Cloud-delivered, subscription-billed, and genuinely excellent businesses, but the product is protection. What drives them is the threat environment and security budgets, which are treated as non-discretionary and behave unlike consumption spending. They belong to the cybersecurity theme.
• Datadog and MongoDB. These two do pass the criteria test, which makes them the most debatable absence on the page. They are excluded on structure rather than definition: they express the same exposure the theme already holds through Snowflake and Cloudflare, at the same premium software multiple, so adding them would deepen one factor rather than add a layer. The roundup covers them because it is answering a different question.
• Data-center REITs and colocation providers. They own the buildings the cloud runs inside and rent space, power, and cooling under long leases. That is real-estate economics driven by lease terms, power availability, and financing costs, not by metered computing demand, so it fails a test that asks whether revenue comes from delivering computing over the internet.

Datadog and MongoDB are the exclusions worth arguing with, because they are the ones excluded on structure rather than on definition. Both are cloud-native, both grow with cloud consumption, and both would fit the criteria comfortably. The reason they are absent is that a roster is a set of decisions about what you already own, and adding two more premium-multiple cloud software names to a theme that already holds Snowflake and Cloudflare buys correlation rather than coverage. The companion roundup at best cloud stocks covers them, because it is answering the question of what people hold rather than the question of what belongs.

Cloud is the theme with the most contested edges on the site, because almost everything in technology now touches it somewhere. Three borders are worth stating explicitly, since four of these six tickers appear in an adjacent theme as well and that is a design decision rather than an oversight.

• AI infrastructure. MSFT, GOOGL, AMZN, and ORCL are in both, on opposite sides of the same transaction. There they are the buyers whose capital spending funds the equipment makers, and the theme is a position on the construction of AI capacity. Here they are the operators of a rental business whose workloads are overwhelmingly not AI, and the theme is a position on metered consumption. The cleanest way to hold the difference is to decide which side of the spending you actually want.
• Enterprise software. That theme is defined by subscription economics: recurring revenue from enterprise customers, expanding over multi-year contracts. This one is defined by consumption economics. MSFT, GOOGL, and ORCL span both because they genuinely sell both, and the names that sit only in the software theme, such as ServiceNow, sell seats and modules rather than computing.
• Cybersecurity. Cloudflare is in both, and it is the only constituent here that is. Its network is cloud infrastructure and security is one of the products it sells across that network, which is the reverse of a dedicated security vendor whose product is protection delivered from someone else's cloud. Security budgets are treated as non-discretionary and behave differently from consumption spending, which is why the two themes exist separately at all.

The point of naming the borders is overlap control. If you hold this theme alongside either of the other two, you are holding several of the same companies twice, and the second position is not adding what it looks like it is adding.

The same six names, grouped by the layer they occupy rather than ranked, so the shape of the theme is visible in one view.

Three of the 6 are mega-caps where cloud is a segment, and three are companies where cloud is the entire business. Reading down the table, exposure to the theme gets purer as market value gets smaller, which is the inverse of how most thematic rosters are shaped.

The passive route answers a different question. An index defines what counts as cloud, then assigns weights you do not control, and cloud definitions vary more than most: some funds are dominated by the hyperscalers, others deliberately exclude them to concentrate on higher-growth cloud software, and the two produce very different results from the same label. Dedicated cloud funds exist in the broader market, and the best cloud ETFs guide compares what each one actually holds. The proxies this theme names are the broad technology funds QQQ, VGT, and XLK, which hold the hyperscalers heavily and dilute the cloud read with a great deal of unrelated technology.

A theme inverts the trade. You know exactly which six names you own, which layer each represents, and what weight each carries, and you accept a narrower roster and the work of maintaining it. Neither is automatically better. The fund is the simpler instrument and the theme is the more deliberate one, and plenty of people hold a broad fund as a core with a small thematic tilt beside it. The thing worth checking either way is overlap, because the three largest constituents here are already among the largest positions in most broad index funds.

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

• Decide how much dilution you are willing to accept. Tilting toward the hyperscalers buys profitability and stability and gives up most of the cloud specificity. Tilting toward the specialists buys specificity and takes on multiple risk and direct competition from your other holdings. That decision changes the position far more than swapping one name for another.
• Set target weights that sum to 100. Equal weighting across six names is a choice, and so is anchoring on the infrastructure layer. Both are defensible. Not deciding is what leaves you concentrated by accident after one name runs.
• Check what you already own. Three of these six are usually significant positions inside any broad index fund, so a cloud tilt can concentrate an existing portfolio while feeling like diversification.
• Frame it against the S&P 500. A narrow thematic position should be judged against a broad benchmark, particularly this one, where several constituents are among the benchmark's largest members already.
• Revisit as weights move. The dispersion between a mega-cap and a consumption-billed software name is wide enough that target weights drift quickly.

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 cloud names are most widely held and discussed, see best cloud stocks. For the adjacent layer that shares three of these tickers, see best software stocks.

The cloud computing theme is six companies across three layers, and the layering is the whole idea. Microsoft, Amazon, and Alphabet own the rentable capacity and are the most diluted cloud exposure on the roster, because cloud is a segment inside a company that mostly does something else. Oracle and Snowflake are the layer where the customer's data and applications live, which is where switching costs actually accumulate. Cloudflare is the layer between users and everything else, paid for traffic regardless of where that traffic lands.

Understood as a flat list of six cloud stocks, the theme looks like three mega-caps plus three software names. Understood as three layers that supply and compete with each other, and whose capital intensity has changed fundamentally at the top while staying light at the bottom, it is a structure with an open question inside it: whether the spending now required to stay in the infrastructure business leaves the economics that made cloud worth owning in the first place. That question is the theme, and it is what you are deciding whether to own. Nothing here is a recommendation, and Walnut is not an investment adviser.