Open any earnings call transcript from the last two years and count the mentions of “AI.” The number has gone up everywhere — at software firms, at banks, at companies that make industrial fasteners. What has not gone up everywhere is revenue attributable to AI. That gap, between the word and the money, is where most retail investors get hurt — and it is why picking an AI stock on the strength of a press release is closer to gambling than investing.
This is not an argument that every AI stock is overvalued, or that the build-out is a bubble. It is an argument that “this company is an AI company” is not an investment thesis, and that you need a way to tell a genuine earnings story from a repositioned press release. Below is a framework you can apply to any listed company, Indian or global, in about an hour of reading.
First, work out which layer the AI stock actually sits in
Every AI stock sits somewhere in a supply chain with distinct layers, and they have wildly different economics. Before anything else, place the company:
- Silicon and hardware. Chip designers, memory makers, networking equipment, the firms building the physical machines. Cyclical, capital-hungry, and currently the layer where the most cash is visibly changing hands.
- Infrastructure. Data centres, power generation and transmission, cooling, land. Slow-moving, asset-heavy, long contracts. Often the least glamorous and most predictable layer.
- Model developers. The firms training frontier models. Enormous costs, uncertain competitive moats, and very few of them are separately listed.
- Application and services. Companies putting models to work — software products, and the IT services firms integrating all of this for enterprise clients. This is where most Indian listed exposure sits.
The layer determines what questions matter. For a hardware firm, ask about order books and customer concentration. For an infrastructure firm, ask about contracted capacity and cost of capital. For a services firm, ask whether AI work is expanding the deal pipeline or quietly cannibalising the headcount-based billing that the business was built on.
That last point deserves emphasis for anyone holding Indian IT. The traditional model prices work by person-hours. If a tool lets four engineers do what twelve did, the client eventually asks to pay for four. AI is genuinely an opportunity for these firms, but it is also a threat to the pricing model, and a company that only talks about the first half is telling you half the story.
Second, find the revenue — or find out that you cannot
This is the step most people skip. Go to the actual filing, not the investor presentation, and look for AI revenue disclosed as a separate line or a stated figure.
Usually you will not find one. That is not automatically damning — a genuinely new business line may be too small to segment, and accounting standards do not require companies to invent a segment because the market is excited. But note what you are being given instead. There is a large difference between:
- “AI-related bookings were $X billion this year, up from $Y” — a number, comparable across periods, that an auditor has some relationship with;
- “We have over 200 AI engagements across our client base” — a count of things, with no revenue attached and no definition of what counts as an engagement;
- “AI is embedded across everything we do” — a sentence that cannot be falsified, and therefore tells you nothing.
If a company has been talking about AI for six or eight quarters and still offers only the third kind of statement, that is information. Companies disclose numbers when the numbers help them.
Third, follow the capital expenditure — and the depreciation behind it
The AI build-out is being financed by enormous capital spending. For any company doing that spending, three questions follow.
How is it funded? Capex out of operating cash flow is a very different risk profile from capex funded by debt raised against expected future demand. Check the cash flow statement and the change in borrowings, not just the capex headline.
What is the assumed useful life? This one is quietly important. Data centre hardware is depreciated over an assumed number of years, and that assumption is a management estimate disclosed in the notes to the accounts. Extending the assumed life of an asset reduces the annual depreciation charge and increases reported profit, without a single additional rupee of revenue. That may be entirely justified — or it may be a company managing its earnings optics. Either way, if the useful-life assumption changed, you want to know, and you want to know why. It will be in the notes. Almost nobody reads the notes — filings for listed Indian companies are free on the BSE and NSE websites.
What return is the spending earning? Capex is an investment, and investments are judged by returns. Watch return on capital employed over several years as the spending ramps. If capital keeps going in and returns keep drifting down, the market is funding an expansion that is not yet paying for itself. That can persist for a long time. It does not persist forever.
Fourth, be precise about what you are paying for
A high multiple is not a reason to avoid a company, and a low one is not a reason to buy. A multiple is a statement about expectations. Your job is to work out what expectations are embedded in the current price, and then form your own view on whether they are reasonable.
A practical way to do this without building a model: take the current market capitalisation, and ask what revenue and margin the company would need to reach in five years for today’s price to look ordinary at a normal multiple. Then ask whether that revenue figure is plausible — is it larger than the entire current market for what the company sells? Does it require the company to take share from every competitor simultaneously? Sometimes the answer is genuinely yes and the price is fair. Often the exercise makes the assumption visible in a way the headline P/E does not.
Do this before you read anyone else’s price target. Once you have seen a number, it anchors you.
Fifth, check the concentration risk on both sides
Two kinds of concentration matter, and companies disclose both if you look.
Customer concentration. If a large share of revenue comes from a handful of buyers, the company’s fortunes are really those customers’ capex budgets. A supplier whose top few clients are all funding AI infrastructure has one bet, not several — and if those customers slow their spending in the same quarter, which they plausibly would, the revenue falls together.
Supplier concentration. The other direction is just as real. A company dependent on one chip vendor, one foundry, or one geography for a critical input carries that risk whether or not it is discussed on the earnings call.
Both figures are usually in the annual report. Neither is usually in the investor presentation.
Practical notes for investors in India
A few structural points specific to investing from here.
Your accessible exposure is uneven. Indian listed markets give you strong access to the IT services layer and to parts of the infrastructure and power story. Direct exposure to frontier chip designers or model developers largely means investing overseas, through the Liberalised Remittance Scheme, a GIFT City route, or an Indian mutual fund or ETF with international holdings. Each route has different costs, different friction, and different tax treatment.
That last point matters more than most people expect. The tax treatment of domestic equity, international equity, and funds holding foreign assets is not the same, holding periods for long-term treatment differ, and the rules have changed more than once in recent years. Verify current rates against the Income Tax Department’s own material or a qualified tax advisor before you commit capital — do not rely on a figure quoted in an article, including this one.
There is also a currency layer. Overseas holdings expose you to rupee movements, which can add to or erase your returns independently of whether you picked the right company. That is not a reason to avoid international exposure. It is a reason to size it deliberately rather than accidentally.
Warning signs in any AI stock
None of these is conclusive alone. Several together are worth pausing over.
- The company added AI language to its description without a corresponding change in what it sells.
- Announcements consistently arrive as partnerships, memoranda of understanding, and pilots — but the pilots never appear again as contracts.
- Management discusses total addressable market far more than its own revenue.
- Depreciation assumptions were extended in a year when reported profit needed help.
- Insiders are selling into strength while the company issues equity.
- The bull case you hold requires everyone else in the sector to also succeed. Genuine competitive advantage is not shared by an entire industry.
The part that is actually hard
None of this analysis is difficult, and it applies to any AI stock on any exchange. It is a few hours with filings that anyone can download for free. The hard part is doing it when a stock has already doubled and the fear of missing out is doing your thinking for you.
Technology transitions are real, and this one has produced genuine businesses with genuine earnings. Previous transitions also produced genuine businesses — alongside a much larger number of companies that borrowed the vocabulary and little else. Both things were true at the same time, and the only way to tell them apart was to look at the numbers rather than the narrative.
The framework does not tell you what to buy. It tells you what you are buying, which is the prerequisite for the first question and the one most people skip.
This article is for information and education only. It is not investment advice, and it does not account for your individual financial situation, goals, or risk tolerance. Consult a SEBI-registered investment adviser before making investment decisions. Tax rules referenced here change; verify current provisions against the Income Tax Department independently.
