Follow AI Revenue to the End: Who Is Paying?

Follow an AI company's revenue upstream and you find customers; follow its spending downstream and you find model developers, cloud providers, and chipmakers. Revenue can grow at every company. Yet adding all that revenue can produce a figure very different from the amount newly spent on AI by end customers.

To understand the difference, distinguish three roles: the customer who pays for the service, the party funding that customer, and the final payer buying AI's utility. One entity may fill all three roles, or they may be spread across businesses and investors.

1. Spending KRW 100 can generate KRW 142 of AI revenue

Combined revenue is KRW 142, while the manufacturer spent KRW 100 on the service. The KRW 42 of intermediate revenue includes money received at one stage and spent again at the next. Each company's revenue is a valid operating result. However, the sum of company revenues measures a different scope from final demand across the supply chain.

Treating the document service's remaining KRW 70 as profit would create another error. Payroll, selling expenses, storage, and other service costs still need to be paid. Check which costs each company includes in gross profit as well.

Adding the price of GPUs bought by the cloud provider introduces another timing issue. Subscriptions and API fees pay for services in the relevant period; GPU purchases pay for assets used across several periods. The chipmaker records revenue when the sale is recognized, while the buyer records an asset and subsequent depreciation. Simply adding one quarter's service revenue to purchases of equipment used for years can overstate market size.

2. Final payers fall into four groups

The final payer can differ from the person who last interacts with AI. Advertisers stand behind consumers using free services, and company purchasing budgets stand behind tools used by employees. Following the purpose of each payment clarifies the structure.

A company can recover AI spending by producing at a lower cost at the same selling price, retaining customers longer, or increasing sales volume. Funding then comes from its existing business revenue and profit. AI's economic contribution can be substantial even when a separately reported ‘AI revenue’ line is small.

Advertising illustrates this. In Q2 2026, Meta reported approximately $60.8 billion in total revenue and $59.4 billion in advertising revenue, about 97.6% of the total.^[1] At a company using AI in recommendations and advertising tools, advertisers are the key paying customers. Distinguish total existing advertising revenue from incremental revenue caused by AI adoption. The latter requires comparing revenue and costs with what would have happened without AI.

Enterprise subscriptions make the payer more visible. In its fiscal Q4 results for the quarter ended June 2026, Microsoft said Microsoft 365 Copilot had exceeded 30 million paid seats.^[2] That demonstrates paid contracts. Converting seat counts into actual revenue using public list prices would also require contract discounts, bundling, and billing periods. How actively purchased seats are used is a separate indicator for renewal.

3. Follow the customer's money one step further

A cloud provider generates revenue when it provides paid services to an AI startup. But demand's durability depends on whether that startup pays with revenue from users or with invested capital.

Early-stage businesses use investment capital while acquiring customers and developing products. The question is whether the relationship between revenue per customer and service delivery costs improves before that capital runs out. If losses grow proportionally with usage, conditions for the next funding round can affect the entire service supply chain.

Funding channels carry different obligations. Equity issuance dilutes ownership, borrowing creates interest and principal repayments, and customer prepayments create future service obligations. Large companies using cash from existing operations forgo other investment opportunities. All these sources enable AI investment, but pressure to recover capital emerges at different times and in different forms.

4. What matters when investors, customers, and suppliers overlap?

An AI industry's infrastructure provider can also be an investor. Capital may then travel through the supply chain and return as the investor's operating revenue. Read equity investment agreements separately from service procurement contracts.

In February 2026, OpenAI announced $110 billion of new investment, including $30 billion from SoftBank, $30 billion from NVIDIA, and $50 billion from Amazon.^[3] The same day's Amazon partnership announcement described investment stages and AWS purchase commitments. Amazon's investment comprised an initial $15 billion and another $35 billion subject to conditions. The AWS agreement added $100 billion over eight years to an existing $38 billion agreement.^[4]

These figures have different meanings and time horizons. Investment commitments raise capital; cloud purchase commitments cover future services. Treating the full investment announcement as cash received that day, or adding the entire multiyear procurement commitment to current-year revenue, misreads the transactions.

The risk to examine is the link between demand and financing. If an AI company's customer revenue grows more slowly than expected, additional fundraising, fulfillment of infrastructure commitments, and suppliers' follow-on capital spending may come under pressure together. If services spread quickly, long contracts provide capacity secured in advance. The value of the same contract depends on how quickly final customer spending and actual usage catch up.

5. Backlog, revenue, and cash run on three clocks

After signing a contract, a provider may need to build a data center and secure power and equipment before delivering services. Revenue is recognized as services are provided; cash arrives according to prepayment or postpayment terms. Contract growth, accounting revenue growth, and investment payback therefore proceed at different speeds.

CoreWeave's Q2 2026 revenue was $2.575 billion, and revenue backlog was approximately $104 billion at the end of June. Its backlog definition includes remaining performance obligations and additional estimates of future committed revenue, subject to service delivery and availability conditions. Operating cash flow for the same quarter was $0.679 billion, while purchases of property and equipment, including capitalized software, were $6.422 billion.^[5]

Subtracting those two cash items gives approximately −$5.743 billion. This is narrower than all of the company's financing and cash transactions. It shows large equipment spending running ahead of cash generated by current operations. Assess it alongside future customer payments, additional financing capacity, delivery schedules, and contract terms.

The inference that many contracts imply sufficient cash skips service delivery and cash collection. Conversely, cash outflows from expansion spending alone do not establish the profitability of individual services. Separating cash generation at operating facilities from investment in new ones distinguishes the existing business from its expansion.

6. Assess revenue quality through five connections

Risk differs according to which connection breaks. Many users but weak paid conversion may indicate a pricing problem. Rising revenue with even faster inference-cost growth may indicate a delivery-cost problem. If customers are concentrated and depend on similar funding conditions, risk may be less diversified than the customer count suggests.

Public data currently do not support a precise allocation of all industry revenue by final payer. Private companies' costs and customer composition, and AI's contribution at large platforms, are not fully disclosed. In this setting, tracing observable company cash flows and contracts is more useful than presenting elaborate estimated percentages.

7. The trail ends with value delivered to the customer

Within the AI supply chain, one company's revenue is another company's cost. Sustained supplier growth requires customers downstream to keep buying. Business customers need actual work outcomes; consumers need recurring utility; advertisers need purchase conversion and customer value.

The money paying current costs may differ from the money that eventually recovers those costs. Capital markets bridge the interval. As that interval lengthens, the pace of final demand growth, falling service costs, and funding maturities becomes more important.

Following AI revenue to the end clarifies the test: start with who paid the invoice, then ask where the money came from and whether the customer received enough value to pay again. After reading supply-chain growth rates, read customer repurchases and the pace of capital recovery.

#AIRevenue #AIIndustry #FinalDemand #CashFlow #CloudComputing #BigTech #AIInvestment #CompanyAnalysis