Nvidia collected $75.2 billion from its data center business in one quarter. Amazon plans to spend about $200 billion in capital this year. Palantir produced $1.94 billion of quarterly revenue. All three numbers belong to the AI market, but they describe three very different businesses.
The stock market often groups them under one label: AI stocks. That label hides who sells the equipment, who pays to build the infrastructure, and who still needs customers to turn new software into durable profit. As of August 2026, AI is a real source of revenue. It is also one of the largest spending programs public companies have ever attempted.
Chips collect the first dollar
Every model needs computing power before it can answer a prompt. That puts chip and networking suppliers at the first paid layer of the market. In its latest reported quarter, Nvidia generated $81.6 billion of total revenue, up 85% from a year earlier. Data center revenue represented $75.2 billion, or about 92 cents of every revenue dollar.
That is not a projection. Cloud companies and model builders have already bought the systems. Nvidia also reported a 74.9% gross margin — the share of revenue left after the direct cost of producing what it sold. A high gross margin shows why the infrastructure layer has captured so much of the early economics.
Nvidia is not alone. AMD reported $5.8 billion of data center revenue in the first quarter of 2026, up 57% year over year, as Instinct GPU shipments increased. The gap remains large, but the result shows that AI demand can support more than one supplier. It also shows the risk: chip companies depend on a relatively small group of customers continuing to place very large orders.
The cloud companies are buyers and sellers
Microsoft, Amazon and Alphabet buy chips, build data centers, and then rent the computing power to customers. Their AI position sits on both sides of the transaction. They are Nvidia's largest kind of customer and its route to thousands of smaller businesses.
Amazon's latest results show both sides clearly. AWS revenue grew 37% to $42.2 billion in the second quarter of 2026, while AWS operating income reached $16.6 billion. Amazon said the annual revenue run rate of its AWS AI business had passed $25 billion. Yet trailing 12-month free cash flow fell to negative $7.6 billion, mainly because property and equipment purchases increased by $66.1 billion, with AI investment driving the increase.
Free cash flow is the cash left after operating costs and capital spending. The AWS figures say customers are paying for AI now. The cash-flow figure says Amazon is building capacity even faster than that demand converts into cash.
Microsoft tells a similar story. Azure grew 39% in constant currency in its fiscal 2026 fourth quarter, while management said demand still exceeded available supply. The company expects roughly $190 billion of capital expenditure during calendar 2026. Alphabet has also said its spending is concentrated in servers and data centers. These are profitable companies funding AI with cash from cloud, advertising and subscriptions, but the size of the build means investors must watch the return on each new dollar of equipment.
The chipmaker records a sale when the server ships. The cloud company earns its return one rented hour at a time.
Platforms can earn without selling an AI subscription
Meta offers a third model. Most of its revenue still comes from advertising, not from charging users for an AI assistant. AI can improve which ad a user sees, how advertisers create campaigns, and how long people stay inside its apps. The return appears inside the existing business rather than on a new line named AI.
In the second quarter of 2026, Meta's revenue rose 28% to $60.8 billion. Ad impressions increased 14%, and the average price per ad increased 12%. At the same time, quarterly capital expenditure reached $31.1 billion and free cash flow narrowed to $784 million. Revenue is responding, but so is the cost base. That makes Meta different from a chip supplier and different from a cloud rental business, even though all three can benefit from the same AI workload.
Software has the widest range of outcomes
The software layer includes companies that sell AI features, agents, analytics and automation. It requires less physical capital than building a data center, but competition is easier to enter and customers can test several tools before committing.
Palantir is one example of clear commercial acceleration. It reported second-quarter 2026 revenue of $1.94 billion, up 93% from a year earlier, with U.S. commercial revenue up 149%. Those numbers show that some businesses will pay for software tied to operational decisions. They do not prove that every company adding an AI button has the same demand, retention or pricing power.
This is where the AI label becomes least useful. A software company can report rapid AI adoption while the new product remains a small part of total revenue. Another can grow quickly from a small base while its stock price already assumes years of expansion. Revenue growth and a good product do not answer what a share is worth.
Read the financial statements by layer
Start with four numbers: AI-linked revenue, its growth rate, capital expenditure, and free cash flow. Then ask where the company sits in the chain.
For a chipmaker, watch data center revenue, gross margin and customer concentration. For a cloud provider, compare infrastructure spending with cloud growth and future contracted revenue. For a platform, look for improvement in the existing engine — advertising, search or subscriptions. For software, look for paid customers, contract value and evidence that growth survives after early trials.
Keep valuation separate from business quality. A company can post strong results and still be a risky purchase if the share price assumes even faster growth. This is the same discipline that keeps a historical pattern from becoming a prediction in the Bitcoin cycle: observed numbers describe what happened, not what the market owes you next.
Open your portfolio and place each AI-related holding into one of four columns: chips, cloud, platform or software. Beside each name, write the latest revenue-growth figure and free cash flow. If you cannot tell which business is paying for the AI story, read its latest earnings release before adding another share.