Google’s Revenue Model Is Changing: What the Cash Flow Reveals About a Business Model

On October 29, 2026, Alphabet’s shares fell almost 13 percent, the worst day since Google went public in 2004, although revenue had grown 25 percent. This post argues that the market reacted to a change in Google’s business model, and that the change shows up first in the revenue model. By revenue model I mean the interplay of costs and revenues over time, which you read in the cash flow, not in the profit line. Google is the case: its production has turned from software into capacity, and its revenue sources now have to follow. Amazon is the second case, a business that was always capital-intensive and whose cash flow has nevertheless changed. The post ends with four questions to read the Economics of a Business in any industry, and five lessons.

Why the revenue model shows a business model change first

Customers buy a product, but what they want is the value it promises. The product is the form in which they pay for that value; the whole business model behind it is what delivers it. The revenue model shows what delivering that value costs, and who pays for it when. The two can drift apart for years before anyone notices, and you only see the gap if you follow the cash.

In my Four-Element Business Model, the revenue model has two building blocks: the cost structure and the revenue sources. My book Das Richtige gründen asks a small set of plain questions about them. What costs us money, and how flexible are those costs relative to revenue? When do they occur? What do we earn money with, and when does the money flow? The book also contains the sentence this post is built on: the cost structure is defined by the value creation architecture. Change how you produce, and you change your costs. Change your costs, and sooner or later your revenue sources have to change too.

Most analyses read business models from the customer side. That is where the story is told. The Economics of a Business, the way cash flows in and out over time, is where the story is tested. In the data-centre post I argued that you should read a business model from its revenue model, not from its value proposition. Google shows how to do it.

Google’s production has changed: from software to capacity

For two decades, Google ran one of the most elegant revenue models ever built. Its production, by which I mean all the activities from building the service to the moment a customer benefits from it, was mostly software. An additional search cost almost nothing. Advertisers paid quickly through an auction. Users paid with their attention. Growth produced cash, and Google returned much of it to shareholders.

The third quarter of 2026 shows a different company. Revenue rose 25 percent to USD 127.6 billion. Operating cash flow fell from USD 48.4 billion to 42.3 billion. Capital expenditure more than doubled to USD 51.4 billion, and free cash flow turned negative at minus USD 9.1 billion (Alphabet 10-Q, Q3 2026; earnings release). The next day the share price fell almost 13 percent, its worst day since 2004, and Alphabet lost about USD 500 billion in market value (CNBC). Investors were pricing what kind of business Google is becoming.

Growth used to produce cash at Google. Now growth consumes it.

The reason sits in production. AI does not run on software alone. It runs on servers, chips and data centres that have to be built before the first customer uses them. Three things changed in Google’s cost structure at once.

The money now flows years before the revenue. Assets not yet in service grew from USD 78.6 billion at the end of 2025 to USD 145.3 billion nine months later (10-Q, property and equipment). In the first nine months of 2026, Google spent six dollars on new equipment for every dollar of depreciation on existing equipment, up from about four a year earlier (own calculation from the cash flow statement).

Costs now come in steps of capacity. In the old model, the cost of serving one more user was close to zero. In the new one, capacity is added in large blocks, and utilisation decides the margin. Inside Google Cloud, employee costs grew 27 percent in a year while all other costs, mostly infrastructure, grew 109 percent (10-Q, segment information).

The capital that carries the advantage ages fast. About 60 percent of Google’s technical infrastructure consists of servers and network equipment; the rest is land and buildings (10-Q, Q3 2026). Google depreciates the buildings over seven to 40 years and the servers over about six (10-K 2025). A data-centre building can outlast six generations of the servers inside it.

Google seems to draw its own conclusion from this. It has signed leases for data centres and energy infrastructure that have not yet started, with future payments of USD 130.5 billion between 2026 and 2031 (10-Q, Q3 2026). On my reading, Google rents much of the hall and buys the servers. It treats the building as interchangeable and keeps what is scarce, which fits what I argued about data centres two weeks ago.

How Google pays for it: the capital market becomes a partner

A change in production does not only change costs. It changes who finances the business.

In the first nine months of 2025, Google bought back USD 39.8 billion of its own shares. In the first nine months of 2026 it bought back none. Instead it raised USD 42.9 billion in new common stock, USD 19.1 billion in mandatory convertible preferred stock and USD 76.3 billion in net new debt (10-Q, cash flow statement). The flow of money between Google and its investors has reversed direction.

In the Four-Element Business Model, partners are part of the value creation architecture. The capital market has just become one of Google’s most important partners. Partners negotiate. The loss of about USD 500 billion in market value on October 29 was such a negotiation.

Google’s revenue sources must follow, and they do so only where customers pay

If the cost structure is defined by production, the revenue sources cannot stay what they were. An auction for attention was the perfect match for near-zero marginal costs. It is a poor match for capacity that has to be paid for years in advance.

In Google Cloud, the revenue side has followed. Cloud revenue more than doubled to USD 30.7 billion in the quarter and now accounts for about a quarter of Alphabet’s revenue and 62 percent of its growth (own calculation from 10-Q, revenue by segment). Here the customer pays directly, by consumption and through multi-year commitments. Google reports USD 567.7 billion of revenue already contracted but not yet recognised, almost all of it in Cloud, about 55 percent due within two years (10-Q, revenues). Cloud has even added a revenue source that would have been unthinkable for Google ten years ago: it now earns product revenue “primarily from the sale of TPU systems”, its own AI chips (10-Q, segment information).

Contracted revenue is the natural counterpart to capital spent in advance. The customer commits before the capacity exists, and the commitment makes the investment bankable. This is what it looks like when the revenue sources follow the cost structure.

It only works because the value proposition is strong enough. Companies sign multi-year commitments because they cannot get the benefit otherwise. Willingness to pay in advance is the clearest test of a value proposition I know.

For consumers, the revenue side has not followed yet. Search and YouTube, where most consumers meet Google’s AI, still run largely on advertising. The user does not pay; the advertiser does. Search revenue grew 12 percent in the quarter (10-Q, revenue by segment). Whether advertising can carry the cost of AI capacity for billions of users is an open question. Alphabet does not report the cost of an AI-supported search, so nobody outside the company can answer it with numbers.

The question behind it is a business model question, and it applies far beyond Google. Will users find the benefit of AI so compelling that they pay for it directly? If they do, the consumer revenue model will change, as Cloud’s has. If they do not, Google carries the cost of the new production with the revenue sources of the old one.

Amazon: always capital-intensive, and still a different cash flow today

Someone could object that capital intensity is nothing new. Amazon has always been capital-intensive and still built one of the most defensible companies in the world. That is true, and it is exactly why Amazon is a useful second case. Its cash flow shows that the change is not in the amount of capital but in its kind.

Until 2019, Amazon’s capital expenditure stayed at around five percent of sales, and it generated USD 25.8 billion in free cash flow that year while paying down debt (10-K 2019). In that year alone, higher payables to suppliers added USD 8.2 billion to operating cash flow. On my reading, customers who paid immediately and suppliers who were paid later financed part of the growth. The investment went into warehouses and logistics with lifetimes of decades.

Between 2019 and 2021, Amazon’s net capital expenditure more than quadrupled. In 2021 and 2022 capital expenditure exceeded operating cash flow, and free cash flow was negative. In 2023 Amazon braked, and free cash flow returned to USD 36.8 billion (10-K 2022; 10-K 2025). A capital cycle can reverse.

The current cycle looks different. Over the twelve months to September 2026, capital expenditure reached 26 percent of sales, more than twice the peak of 2021. Free cash flow fell to minus USD 42.4 billion, and Amazon raised USD 130 billion in new long-term debt. Amazon itself states that the increase in capital expenditure “primarily reflects investments in artificial intelligence” (Amazon Q3 2026 results, SEC filing). In 2025, AWS received 68 percent of all additions to property and equipment, up from 51 percent in 2023. Servers and networking became the largest asset class on the balance sheet, depreciated over five to six years. In 2025, Amazon shortened the useful life of a subset of its servers from six years back to five (10-K 2025).

Amazon used to grow on capital that its customers and suppliers financed and that lasted for decades. Today it grows on capital that bond investors finance and that lasts five or six years.

On the revenue side, AWS shows the same pattern as Google Cloud: business customers paying for capacity, with revenue growing 39 percent in the quarter. Oracle went down the same road earlier and with less cushion, from asset-light software to infrastructure; I described it in Oracle’s business model and the hard truths about the future.

Two companies with very different origins, one from retail, one from advertising, now show the same cash flow signature: capital expenditure above operating cash flow, financing from outside, short-lived assets. Once they enter the same layer of the AI stack, the economics of that layer override the economics of where they came from. This is the logic of a Dominant Business Model Design: an industry is a set of similar business models, and the layer decides which one.

Four questions to read the revenue model: the Economics of a Business

The numbers above will be outdated in three months. The questions behind them will not. They come straight from the revenue model in Das Richtige gründen, and they work for a machine builder, a hospital or a software company as well as for Google.

If two or more answers change, your business model has changed, whether or not your value proposition looks the same.

Five business model lessons for every industry

1. Read the business model from its revenue model, and read the revenue model over time. Profit is a snapshot. The interplay of operating, investing and financing cash flows shows what a company really is. Google still looks like a search company. Its cash flow says it is becoming an infrastructure business.

2. Production defines the cost structure. When the way you produce changes, your costs change, whether you planned a new business model or not. AI changed Google’s production first. The business model followed.

3. The revenue sources have to follow the cost structure, and they only follow where customers are willing to pay. Capacity paid for in advance needs revenue committed in advance. That works when the value proposition is strong enough that customers commit before they see the result. Where it is not, the old revenue model has to carry the new costs.

4. Who finances your growth is part of your business model. Customers, suppliers, your own cash or the capital market: each source is a partner with its own conditions. A change of financier changes who negotiates over your business.

5. Know how long the capital lives that carries your advantage. Own what is scarce, rent what is interchangeable, and price in how fast the scarce part ages. An advantage that sits in equipment with a life of five years has to be earned again every five years.

The revenue model has been the quiet element of business model thinking. It is where the numbers live, and numbers seem to belong to the finance department. Google’s quarter shows why that is a mistake. The change in its business model was visible in the cash flow long before it was visible in its products. Follow the money, ask when it flows, and you will see what a business is becoming. Understand. Imagine bigger. Act.

Annotated References

Alphabet Inc. (2026, October 28). Alphabet Announces Third Quarter 2026 Results. Earnings release with revenue, segment results, cash flow and capital expenditure guidance of USD 195 to 205 billion for 2026 and a significant increase in 2027. Link

Alphabet Inc. (2026). Form 10-Q for the quarter ended September 30, 2026. SEC. Primary source for the cash flow statement, segment costs, contracted revenue not yet recognised (USD 567.7 billion), the share of servers in technical infrastructure (about 60 percent) and USD 130.5 billion in data-centre and energy leases not yet commenced. Link

Alphabet Inc. (2026). Form 10-Q for the quarter ended June 30, 2026. SEC. Half-year cash flow statement, used to derive the third-quarter figures. Link

Alphabet Inc. (2026). Form 10-K for the year ended December 31, 2025. SEC. Accounting policies with the useful lives of servers (about six years) and data-centre buildings (seven to 40 years). Link

Amazon.com, Inc. (2026, October 29). Amazon.com Announces Third Quarter Results (Exhibit 99.1 to Form 8-K). SEC. Three-, nine- and twelve-month cash flow statements; twelve-month free cash flow of minus USD 42.4 billion, which Amazon attributes primarily to investments in AI. Link

Amazon.com, Inc. (2020, 2023, 2026). Form 10-K for the years 2019, 2022 and 2025. SEC. Annual cash flow statements from 2017 to 2025, property and equipment by category, additions by segment and the change in server useful lives. 2019 · 2022 · 2025

CNBC. (2026, October 29). Alphabet tanks nearly 13% for worst day ever, losing $500 billion in value. Market reaction to the third-quarter results: the largest one-day percentage drop since Google’s IPO in 2004, ahead of the previous record of 11.63 percent in March 2020. Link

Stähler, P. (2021). Das Richtige gründen: Werkzeugkasten für Unternehmer (5th ed.). Murmann. Source of the Four-Element Business Model and of the questions on cost structure and revenue sources used in this post, including the principle that the cost structure is defined by the value creation architecture.