“AI stocks” is not one business. The chain includes accelerator designers, memory suppliers, networking companies, cloud operators, power developers, cooling firms, and data-center landlords. Each layer earns money differently and carries different risks.

The short answer

Separate the AI infrastructure chain before comparing companies. Chip growth can be strong while cloud margins weaken. Data-center demand can rise while a power developer waits years for grid approval. A good theme does not make every company in the theme equally attractive.

The main layers

Chip companies sell silicon, software, and systems. Their results can be sensitive to product cycles, customer concentration, export rules, and competition. Cloud operators sell access to compute and carry utilization, power, financing, and support costs.

Power and cooling companies benefit from the physical demand of data centers, but their projects may have long approval and construction timelines. Real-estate operators may receive leases while carrying development and tenant-concentration risk.

The same AI project can appear in several companies’ narratives. That does not mean the same dollar becomes profit for all of them. Trace who pays whom, what asset is delivered, and when revenue is recognized.

Questions for each layer

For chips, ask about customers, refresh cycles, software dependence, and inventory. For clouds, ask about utilization, contracts, pricing, power cost, and depreciation. For power, ask about permits, grid connection, project funding, and delivery. For real estate, ask about lease duration, tenant concentration, and construction obligations.

Use a 2-hour research session to build a simple chain diagram for one company. Mark confirmed revenue separately from projected demand. That exercise is often more revealing than a list of popular tickers.

Risk is not the same everywhere

A chip supplier can face a rapid product transition. A data-center landlord may face slower but longer commitments. A utility may have regulated returns and different exposure to demand. Compare the risk to the business model rather than using one AI valuation rule for all of them.

Sources and limits

NVIDIA’s annual report discusses indirect customers and infrastructure expense. Its quarterly filing discusses cloud commitments. OpenAI’s compute infrastructure overview shows the land, power, and workforce dependencies behind the theme. This is educational analysis, not personalized investment advice.

The useful conclusion is to invest in a business you understand, not a label that happens to contain the letters AI.

Build a comparable snapshot

For each company, record revenue source, major customers, asset ownership, capital intensity, geographic exposure, and the date when the next capacity is expected. Keep projections separate from reported figures. A company with a smaller headline may have a better balance sheet or a more durable contract.

Look for the mismatch between a story and its operating evidence. If demand is described as enormous but receivables, utilization, or delivered capacity do not move, ask why. If capex rises faster than revenue, ask when the new assets should become productive and how they are financed.

Spend 2 hours building this snapshot before making a comparison. The goal is not to predict a stock price. It is to understand which assumption would have to be true for the business case to work.

That assumption may be utilization, pricing power, delivery speed, or access to capital. Name it explicitly before comparing companies.

Do not double-count the story

One data-center project can support the revenue story of a chip supplier, a cloud operator, a power company, and a landlord. Their exposure is not identical. Map the cash flow between them before treating several headlines as independent confirmation.

An investor should also separate business analysis from personal risk tolerance. A company can be excellent and still be too volatile for a particular plan. Use primary filings, understand the downside, and avoid letting an exciting theme replace a diversified decision.

Diversification matters more than a fashionable category.

It protects the analysis too.

Keep the categories separate.

Use primary filings.

Always compare dates.

Carefully.

Always.