For most of the current AI cycle, the central question has been how quickly demand would grow.
How many users would adopt AI?
How rapidly would enterprises deploy it?
How much compute would be required?
How many data centres would have to be built?
Those questions still matter.
I think another one now matters just as much:
Who is financing the growth?
Revenue growth and revenue quality are not the same thing.
A company can double its annualised revenue.
It can sign multi-billion-dollar contracts.
Its backlog can expand rapidly.
All of those figures can be real.
But if a meaningful share of the customers generating that revenue also depend on continuous external financing to maintain their spending, the economic profile of that revenue changes.
That is one of the more interesting features of today's AI ecosystem.
At one end are some of the largest and most cash-generative companies in the world.
At another are frontier model companies growing rapidly while making extraordinary infrastructure commitments.
Below them are cloud and data-centre operators that must borrow heavily to acquire GPUs, secure power and build capacity before the associated revenue arrives.
They are all participating in the same AI boom.
Their revenue is not economically equivalent.
What do I mean by revenue quality?
I am not using revenue quality here as an accounting term.
This is not a question about whether a company has recognised revenue correctly.
It is an investment and strategy question.
Who is the customer?
Is the customer actually using the product?
Does the revenue repeat?
Is the invoice being collected?
How concentrated is the customer base?
Is the customer paying from operating cash flow or from newly raised capital?
How much investment does the supplier need to make before it can earn that revenue?
Can today's revenue continue under similar economic conditions tomorrow?
For investors and CFOs, these questions are becoming almost as important as the headline growth rate.
Table 1. The same revenue can have very different quality
| Dimension | Higher-quality revenue | More fragile revenue |
|---|---|---|
| Customer base | Broad and diversified | Concentrated among a few clients |
| Source of customer funding | Operating cash flow | New equity or debt |
| Revenue type | Actual usage and recurring payment | Future capacity commitment |
| Collection | Regular and short-cycle | Dependent on long contracts |
| Capital required | Controlled relative to revenue | Heavy CAPEX before revenue |
| Gross margin | Improves with scale | Pressured by compute costs |
| Cash flow | Positive or improving | Negative for an extended period |
| Customer concentration | Low | High |
| Growth dependency | Product demand | Product demand plus capital markets |
This distinction matters more as the scale of the AI investment cycle increases.
The capital requirement is now measured in trillions
S&P Global estimates that the largest hyperscalers could spend more than $7 trillion on AI-related capital investment between 2025 and 2030.
The rating agency has also highlighted the increasingly complex structures being used to finance that expansion, including project finance, special-purpose vehicles, guarantees and long-term contractual arrangements. Reuters
Morgan Stanley estimates cited by Reuters suggest that AI infrastructure could require around $1.5 trillion of external financing by 2028. Reuters
There is an important distinction here.
A sector can be experiencing genuine demand and still require enormous amounts of external capital to satisfy it.
AI is currently doing both.
The demand is real.
Financing the infrastructure required to serve that demand is exceptionally capital intensive.
Microsoft shows what high-quality demand can look like
Microsoft provides a useful reference point.
Microsoft Cloud generated $214.4 billion of revenue in fiscal 2026.
Commercial remaining performance obligations reached $678 billion.
More importantly, management said that nearly 90% of Microsoft Cloud revenue came from customers outside frontier model companies. Microsoft
That matters.
Microsoft's AI exposure is not dependent on a single model company or a handful of counterparties.
Azure, Microsoft 365 and its broader cloud portfolio are supported by a large and diversified enterprise customer base.
The company also generates substantial cash.
Operating cash flow was $55.4 billion in the final quarter of fiscal 2026.
Free cash flow remained positive at $19.6 billion despite very high infrastructure investment. Microsoft
Microsoft expects roughly $190 billion of capital expenditure in calendar 2026. Microsoft
That is an extraordinary investment number.
But the number becomes more meaningful when considered alongside the quality of the revenue supporting it.
Microsoft has broad customer diversification, existing profitable products, strong collections and a substantial operating cash engine.
The relevant question is therefore not simply whether $190 billion of investment is high.
It is what kind of customer demand sits on the other side of that investment.
Amazon shows that genuine demand can still absorb enormous amounts of cash
Amazon presents a similar but slightly different picture.
AWS generated $42.2 billion of revenue in the second quarter of 2026, up 36.7% from the previous year.
Amazon also said its AI business had exceeded a $25 billion annual revenue run rate. Securities and Exchange Commission
This is substantial operating demand.
At the same time, Amazon's trailing twelve-month free cash flow moved from a positive $18.2 billion to an outflow of $7.6 billion.
The company attributes much of that change to the rapid increase in capital expenditure required for AI infrastructure. ir.aboutamazon.com
I would not describe AWS revenue as low quality.
It is a large, profitable business serving a broad base of real customers.
The lesson is different.
Even high-quality revenue can put pressure on free cash flow when the investment required to support future growth accelerates faster than current cash generation.
For AI businesses, investors therefore need to look beyond revenue growth and consider how much new capital is required to create each additional dollar of revenue.
Alphabet raises the same question
Alphabet's second-quarter results make the point again.
Google Cloud revenue increased 82% year over year to $24.8 billion.
Alphabet's total quarterly revenue reached $119.8 billion. Securities and Exchange Commission
The demand side looks exceptionally strong.
At the same time, Alphabet increased its 2026 capital expenditure outlook to between $195 billion and $205 billion.
Reuters reported that the company recorded negative free cash flow of about $5.9 billion during the quarter, the first negative quarter in its history. Reuters
Cloud revenue is growing at 82%.
The infrastructure required to produce future cloud revenue is also consuming capital at an extraordinary rate.
The next question for investors is therefore not only how quickly Google Cloud can grow.
It is how much incremental capital is required for every incremental dollar of growth.
Meta is financing AI through an existing cash engine
Meta has a different economic structure.
Its principal source of revenue remains advertising.
Second-quarter 2026 revenue increased 28% to $60.8 billion.
Costs and expenses rose 55%.
Operating margin fell from 43% to 31%.
Capital expenditure in the quarter reached $31.1 billion, while free cash flow fell to only $784 million.
Meta expects full-year 2026 capital expenditure of between $130 billion and $145 billion. Atmeta
In Meta's case, the current AI build-out is not primarily being funded by customers purchasing a standalone AI product.
It is being financed by the cash-generating capacity of its existing advertising business and balance sheet.
That is not necessarily a weakness.
Meta's core business remains an exceptional cash-generating platform.
But the valuation question is different.
Is AI creating value because it improves the economics of the advertising business today?
Or is today's advertising cash being invested in an entirely new revenue pool that is expected to emerge later?
Those two cases should not be valued in exactly the same way.
OpenAI shows why even the definition of revenue now matters
Recent discussion around OpenAI's revenue numbers is particularly relevant.
Reuters reported on October 8 that OpenAI had told investors its September annualised revenue was close to $50 billion.
Earlier comparisons had suggested a number approaching $70 billion.
The difference largely reflected the way revenues flowing through cloud partners were treated when comparisons were made with Anthropic. Investing.com
This is not an allegation of improper accounting.
The important issue is what the metric actually represents.
An annualised revenue run rate takes the current pace of sales, often over a relatively short period, and extrapolates it over twelve months.
That can be useful in a rapidly growing business.
It is not the same as twelve months of realised revenue.
It is not cash flow.
It does not tell us the quality of the customer.
And it does not guarantee that the same level of spending will continue next year.
For AI companies in particular, investors need to distinguish several concepts.
Table 2. Revenue metrics that should not be treated as interchangeable
| Metric | What it tells us | What it does not tell us |
|---|---|---|
| Realised revenue | Sales already recognised | Future growth |
| Annualised run rate | Current sales pace extrapolated | Realised annual sales |
| Bookings / backlog | Contracted future business | Timing and margin of revenue |
| RPO | Visibility into future contracted revenue | Collection and profitability |
| Usage growth | Product adoption | Customer profitability |
| ARR | Recurring revenue potential | Capital requirements and cash burn |
This distinction is becoming increasingly important because much of the sector's valuation rests on future revenue.
Future revenue and current cash are not the same thing.
CoreWeave is where the issue becomes most visible
CoreWeave is one of the clearest examples of the financial structure behind the AI infrastructure cycle.
The company has signed long-term contracts worth tens of billions of dollars with customers including Microsoft, OpenAI, Meta and Anthropic.
Reuters estimates its debt at roughly $50 billion.
Microsoft represented approximately 67% of CoreWeave's 2025 revenue. Reuters
Those contracts provide significant revenue visibility.
But the infrastructure has to be built before much of the revenue arrives.
GPUs must be purchased.
Data centres must be completed.
Power must be secured.
Interest must be paid.
Revenue is then recognised over time as capacity becomes available.
That creates a timing problem.
A company can have a very large backlog and still have weak free cash flow.
J.P. Morgan expects CoreWeave to remain free-cash-flow negative until 2030. Reuters
The lesson is important:
Contract quality and balance-sheet quality are not the same thing.
A business may have excellent customers under long-term agreements and still carry significant financial risk if fulfilling those agreements requires expensive external capital.
The next question is who finances the customer
This is where I think AI analysis will become more interesting.
In CoreWeave's case, Nvidia is not only a supplier of GPUs.
It is also an investor, and Reuters reports that Nvidia has agreed to purchase up to $6.3 billion of unsold computing capacity. Reuters
Elsewhere in the ecosystem, chip suppliers invest in infrastructure providers.
Hyperscalers invest in model companies.
Model companies then commit billions of dollars to cloud services.
Cloud companies use that demand to finance infrastructure that purchases more chips.
None of these relationships is inherently problematic.
There can be perfectly rational strategic reasons for each one.
But from an investor's perspective, it makes the source of economic demand increasingly important.
Table 3. The AI capital cycle
| Stage | Flow of capital | Financial question |
|---|---|---|
| Investor | Provides capital to AI company | What future cash flow will repay the investment? |
| Model company | Buys cloud and compute | Is spending funded by operating revenue or new capital? |
| Cloud / neocloud | Builds GPU and data-centre capacity | Do contracts cover CAPEX and financing costs? |
| Chip supplier | Sells GPUs | Is demand coming from end usage or infrastructure investment? |
| Strategic investor | Invests or provides guarantees | Is capital flowing back into the sales ecosystem? |
| End customer | Pays for AI products | Is the economic value strong enough to sustain spending? |
For the AI system to become financially self-sustaining, the final part of that chain needs to keep growing.
Businesses and consumers ultimately need to pay for AI because it creates enough economic value for them to continue doing so.
Capital can accelerate the ecosystem.
It cannot permanently replace end-customer economics.
Why the source of revenue affects valuation
Imagine two companies.
Both generate $1 billion of annual revenue.
Both are growing at 60%.
At first glance, they might appear to deserve similar revenue multiples.
Now look beneath the headline numbers.
Table 4. Same growth, different value
| Company A | Company B | |
|---|---|---|
| Revenue | $1bn | $1bn |
| Growth | 60% | 60% |
| Largest customer | 8% of revenue | 65% of revenue |
| Gross margin | 70% | 40% |
| Operating cash flow | Positive | Negative |
| CAPEX supporting revenue | $200m | $1.5bn |
| Growth funding | Internal cash | Debt and new equity |
| Customer funding | Operating budgets | Largely external capital |
| Revenue visibility | Diversified | Concentrated contracts |
| Capital-market dependency | Low | High |
The revenue numbers are identical.
The economic value is not.
Company A is moving toward financing growth through its own operations.
Company B needs capital markets to remain open in order to maintain the same rate of expansion.
This is why AI valuation needs to move beyond headline revenue multiples.
I would look at four layers together.
Revenue growth
How fast is the business growing?
Revenue quality
Who is paying, and how durable is the payment?
Capital intensity
How much capital is required to create that growth?
Cash conversion
How quickly does accounting revenue become free cash flow?
A company can be exceptional on the first measure and weak on the other three.
High growth alone may not be sufficient to justify a premium valuation.
A different dashboard for CFOs and investors
Traditional SaaS metrics still matter.
AI infrastructure introduces additional questions.
Table 5. Revenue-quality dashboard for AI investments
| Metric | Question to ask |
|---|---|
| Revenue growth | How much real growth is occurring? |
| Price / Volume / Usage | Is growth driven by pricing or actual usage? |
| Customer concentration | What share comes from the top 1, 5 and 10 customers? |
| ARR / Run rate | How different is it from realised revenue? |
| Backlog / RPO | When will it convert into actual revenue? |
| Gross margin after compute | What margin remains after compute costs? |
| Operating cash conversion | How quickly does revenue become cash? |
| CAPEX / incremental revenue | How much investment is required for another dollar of revenue? |
| External funding dependency | Can growth continue without new capital? |
| Customer funding quality | Is the customer paying from operations or financing? |
| Contract duration | How long is the contract and what are the cancellation terms? |
| Renewal / usage | Is real usage growing beyond contractual commitments? |
| ROIC | Will infrastructure earn above its cost of capital? |
I expect CAPEX per incremental dollar of revenue to become a particularly useful measure in AI infrastructure.
Traditional software can often generate significant incremental revenue with limited physical investment.
AI infrastructure is different.
New revenue may require GPUs, data centres, power and networking equipment.
The same revenue growth rate can therefore create very different economic outcomes.
Lower revenue quality does not mean AI demand is unreal
This distinction matters.
The data do not suggest that AI demand is fictional.
Quite the opposite.
Microsoft Cloud generated more than $214 billion in fiscal 2026.
AWS's AI business exceeded a $25 billion annualised revenue rate.
Google Cloud grew 82% in a single quarter. Microsoft
There is real demand.
There are real customers.
There is real revenue.
The question has moved on.
Will today's investment required to serve that demand generate an adequate economic return?
The answer will not be the same for every company.
Some firms will earn extraordinary returns on the infrastructure they are building today.
Some will build too much capacity.
Some model companies will become global platforms.
Others will struggle if external capital becomes more difficult or expensive to access.
That dispersion is normal in a major investment cycle.
What changes for valuation?
During the early phase of the AI cycle, growth was the dominant valuation variable.
I expect the composition of that growth to matter more from here.
Higher-quality revenue can support a higher valuation when:
the customer base is broad,
revenue repeats,
collections are strong,
customers are paying from their own operating economics,
gross margins remain attractive,
capital required for growth is manageable,
and the path to free cash flow is visible.
A greater valuation discount may be appropriate when:
revenue is concentrated,
customers themselves depend heavily on external capital,
large CAPEX is required before revenue arrives,
borrowing costs are rising,
contract economics take years to mature,
and continued growth depends on capital markets remaining open.
At that point, a simple revenue multiple stops telling us enough.
Conclusion
The first phase of the AI investment cycle asked whether the technology would work.
The second asked whether customers would adopt it.
I think we are now entering a third phase.
Demand has arrived.
The market is beginning to ask whether the economics behind that demand work.
Answering that question requires more than looking at revenue.
We need to understand who is paying for it.
We need to understand where that customer obtained the money.
We need to understand how much capital must be invested before the revenue can be earned.
And, most importantly, we need to understand when accounting revenue becomes genuine free cash flow.
The most valuable companies in the next phase of AI may not simply be the companies growing fastest.
The more important divide may be elsewhere.
Between businesses that must continually finance growth with new capital and those that increasingly finance growth with cash generated by their own customers.
Revenue tells us how large the business is becoming.
Revenue quality tells us how durable that growth may be.
Mehmet Taşdemir
Finance & Strategy Focus
21.10.2026
Microsoft Investor Relations, FY2026 Q4 Investor Metrics
Microsoft FY2026 Q4 Metrics
Microsoft, FY2026 Q4 Earnings Conference Call
Microsoft FY2026 Q4 Earnings Call
Microsoft, FY2026 Q3 Earnings Conference Call
Microsoft FY2026 Q3 Earnings Call
Amazon, Second Quarter 2026 Results
Amazon Q2 2026 Results
Amazon SEC Filing, Quarter Ended June 30, 2026
Amazon Q2 2026 SEC Filing
Alphabet, Second Quarter 2026 Results
Alphabet Q2 2026 Results
Reuters, Alphabet AI Investment and Cloud Growth
Reuters: Alphabet Raises AI Capex Forecast
Meta, Second Quarter 2026 Results
Meta Q2 2026 Results
Reuters, OpenAI September Annualized Revenue
Reuters: OpenAI Revenue Run Rate
Reuters, CoreWeave AI Contracts and Financing
Reuters: CoreWeave Contracts and Lenders
Reuters, CoreWeave Debt and Bank Exposure
Reuters: CoreWeave Debt Analysis
Reuters, S&P Global on AI-linked Debt and Hyperscaler Spending
Reuters: Hyperscaler AI Debt and Capital Spending
Reuters, AI Infrastructure External Financing
Reuters: Financing the AI Infrastructure Cycle