Three numbers, no common denominator

What Indian IT's AI-revenue disclosure season actually tells us

Between July 9 and July 24, three of India's largest IT services companies each put a number on how much of their business is now AI. TCS said $2.6 billion, annualized. HCLTech said $171 million for the quarter, up 62% year-on-year. Infosys said 8.2% of a $5.08 billion quarter. None of the three definitions match, none is audited, and none was disclosed a year ago. What makes this window worth stopping on is not that the Indian services majors are talking about AI — everyone talks about AI — but that in the space of two weeks they converted the talk into a number their investors will now expect every quarter, before anyone has agreed on what the number measures. Underneath the disclosures sits a genuine, unresolved question: is AI revenue additive to the roughly $250 billion Indian IT services export business, or is it partly relabeling the same work at a lower price point, dressed in a growth metric that reads well on an earnings call? The three prints, read together with a services-delivery model that started changing shape in the same window, are the closest thing the industry has offered to an answer — and the honest reading is that the answer is still both.

The events

TCS opens the season, July 9. TCS reported Q1 FY27 revenue of ₹72,275 crore, up 13.9% year-on-year, and led its own release with the AI figures rather than burying them: an annualized AI revenue run rate of $2.6 billion, AI-deal revenue up more than 13% year-on-year, $9.5 billion in total contract value, and an $800 million AI-led transformation deal with SKF. CEO K Krithivasan called the company "generally optimistic on Q2." As the July 12 digest noted, net profit grew at roughly a third the pace of revenue that quarter — growth framed around AI-led transformation, arriving with margin pressure attached. TCS also named an expanding multi-lab partnership bench — Anthropic, Mistral, Google Cloud, ServiceNow — the capability stack a $2.6 billion run rate is built on top of.

HCLTech answers four days later, July 13. HCLTech reported $171 million in advanced-AI revenue for the quarter, up 62.1% year-on-year and 10.6% quarter-on-quarter, against record Q1 net-new bookings of $2.407 billion. As the July 15 digest framed it, this was "the answer to a question the sector left open last week": whether AI was repricing services deals or sitting as a bolt-on. HCLTech's number, growing double digits sequentially, read as production business rather than pilot revenue — and it landed the same week the company set GST-inclusive rupee pricing context elsewhere in the archive, and three months into a relationship with Sarvam AI that would deepen further by the end of the window (below).

Infosys completes the trio, July 23. Infosys reported Q1 FY27 revenue of $5,082 million, up 2.4% year-on-year in constant currency, with AI-related revenue at 8.2% of the total — roughly $417 million in the quarter — continuing a run of double-digit sequential growth in AI work. Large-deal wins totaled $3.6 billion in total contract value, 61% net new. Against that, Infosys narrowed its FY27 constant-currency revenue growth guidance to 1.5%–3.0%, down from the prior 1.5%–3.5% range, citing softer volumes, a client program termination worth roughly 50 basis points, and weaker-than-expected pricing improvement. As the July 21 digest put it, "two numbers in the same release are pulling in opposite directions, and that's the actual story."

The delivery model moves under the disclosures, July 24. The same week Infosys published its number, Analytics India Magazine reported that TCS is targeting 6,000 forward-deployed engineer (FDE) roles — AI experts embedded directly at client sites, the model AI-native companies including Microsoft and OpenAI have used to sell enterprise AI implementation. Infosys, Cognizant, LTIMindtree, and HCLTech are building equivalent programs. The July 25 digest's framing is precise: the FDE model is "a billing and engagement structure change, not just a headcount reallocation" — closer to a management-consulting retainer than the RFP-won, milestone-paid project scope that has defined Indian IT billing for two decades.

Four events, eleven days, one arc: three companies each decided, independently, that AI revenue needed its own line — and the delivery model underneath that line started changing shape in the same window.

The mechanism: what "AI revenue" actually counts

The three disclosures are not comparable because they are not measuring the same thing, and it is worth being precise about how each is constructed from what each company has published.

A run rate, a discrete revenue line, and a percentage of total revenue are three different units of measurement, constructed from three different internal classification systems, none of which any company has published in enough detail to let an outside analyst check the boundary cases. Does a traditional application-maintenance contract that now uses an AI coding assistant internally count as "AI revenue," or only a contract explicitly sold as an AI transformation? Does a chatbot deployed inside an existing BPO contract count, or does the classification require a new statement of work? None of the three companies' Q1 FY27 disclosures answer this. TCS's own coverage acknowledges as much — "there is no standard definition of what counts as AI revenue in an IT-services book" was true when TCS reported it on July 9, and it remained true through Infosys's print on July 23.

This is not necessarily deception. It is closer to how "digital revenue" was reported across the same sector a decade ago — a real shift in delivery mix, reported through a metric each company defined for itself before any standards body or SEC guidance caught up. The useful reading is directional rather than comparative: each company is telling investors AI-labeled work is a growing share of what it sells, and each is unwilling to let a rival's disclosure norm go unanswered. That competitive pressure, more than any external reporting standard, is what produced three metrics in eleven days after zero metrics a year earlier.

The billing-model shift compounds the measurement problem. An FDE embedded at a client site — recurring, retainer-priced, working across the client's full IT surface rather than a bounded project scope — does not map cleanly onto either "project revenue" or a discrete "AI revenue" line. If TCS's 6,000 FDE roles get priced and delivered as a blended service, the revenue they generate may or may not get classified as "AI revenue" in next year's disclosure, depending on how the company chooses to draw the line. The metric problem the sector has now created for itself is likely to get harder before it gets easier, not easier, as the underlying delivery model itself becomes less legible to outside observers.

The tension inside Infosys's own print is the sharpest version of the open question. An 8.2% AI revenue share, growing double digits sequentially, with 61% of large-deal TCV explicitly net new — sitting against a full-year revenue guidance range narrowed downward, citing softer volumes and weak pricing improvement. Two readings are consistent with that combination of facts, and Infosys's disclosure as published cannot distinguish between them: AI work is genuinely additive and the rest of the book is shrinking under demand softness that has nothing to do with AI; or AI-led engagements are partly substituting for — not adding to — the volume and pricing power of the traditional services model, which is the "AI deflation" pattern the sector has been bracing for since the beginning of the year. Infosys has not broken out how the 8.2% figure interacts with the guidance narrowing, and nothing in the public record forces the company to.

Comparables: two other places the same substitution question is playing out

Inside the same disclosure season, the FDE model is the AI-native answer to the same substitution logic Indian IT is now defending against. ChatGPT Work, which OpenAI launched to general availability on July 9 — the same day TCS reported its $2.6 billion run rate — assembles finished documents, spreadsheets, and presentations from workplace context, automating a slice of exactly the delivery work Indian services firms bill for. The July 12 digest framed the live question precisely: "whether ChatGPT Work becomes a tool the SIs deploy for clients or a product that routes around them." TCS, Infosys, Cognizant, LTIMindtree, and HCLTech building out FDE programs in the same window is the sector's structural answer — not competing with the AI-native product on model access, which the SIs cannot win, but competing on enterprise breadth and multi-client relationship depth, which the AI-native companies do not have. Whether that trade holds is, again, a question the current disclosures cannot settle: an FDE program is a bet that clients will pay retainer rates for embedded expertise regardless of which model sits behind it, and that bet has not yet been priced in public.

Outside India, the token-pricing side of the same substitution pressure is visible in the same window. The July 7 digest reported that Chinese-origin models had held at least 30% of weekly US enterprise token volume on OpenRouter every week since February, peaking at 46.4%, at pricing 60–90% below US flagship rates — and framed the India-relevant consequence directly: "the argument that Indian buyers should pay the US-flagship premium for trust reasons gets harder to make internally." That is a different market — model-layer token pricing rather than services-layer delivery pricing — but the mechanism rhymes. In both cases, a cost structure that used to carry a premium (US frontier tokens; Indian services-delivery labor) is under pressure from a cheaper substitute (Chinese open-weight tokens; AI-automated delivery), and the incumbent's response in both cases is the same: reprice down, or move up the value chain into work the substitute cannot yet do. TCS's FDE program and HCLTech's Sarvam equity position are both, read this way, bets on the second strategy rather than the first.

Where it lands

The near-term test is mechanical: Q2 FY27, reporting in October. All three companies disclosed Q1 FY27 AI figures for the first time in this form; whether TCS's run rate, HCLTech's advanced-AI line, and Infosys's revenue share repeat as standing quarterly disclosures — rather than one-off numbers offered once and quietly dropped — is the first thing to watch. A second print from each converts a one-time number into a series. It is likely, given the competitive pressure that produced the first disclosures within eleven days of each other, that all three repeat; it is less certain that any of them narrows the definitional gap between the three metrics.

HCLTech's own stated threshold is concrete and checkable. The July 15 digest recorded HCLTech's own framing of the level "at which AI would be repricing the deal mix rather than adding a line to it" — an advanced-AI revenue line crossing $250 million a quarter within a year of the July 2026 disclosure. That is a specific, dated, company-set marker against which the next several quarters of HCLTech prints can be checked without relying on any cross-company comparison.

Infosys's guidance trajectory is the more direct read on the substitution question. If AI revenue is genuinely additive, the expectation is that a rising AI share should eventually show up as guidance holding or improving even as legacy volumes soften — evidence that new AI-native demand is filling the gap the traditional book is losing. If Infosys's AI share keeps climbing while full-year guidance keeps narrowing across successive quarters, that is closer to evidence for the deflation reading: AI work growing inside a shrinking envelope, not expanding it. One data point cannot distinguish these; a second and third narrowing alongside a rising AI share would be a meaningfully different signal than a guidance range that stabilizes.

The FDE compensation structure is the labor-market tell. Whether TCS, Infosys, Cognizant, LTIMindtree, and HCLTech price embedded AI-expert roles at management-consulting day rates, standard IT-project rates, or something between the two — data none of the July reporting disclosed — will determine two things at once: whether the FDE model retains senior AI engineering talent inside Indian IT rather than losing it to AI-native companies paying lab-adjacent compensation, and whether the FDE line, if it becomes large enough to matter, drags the sector's overall margin profile toward consulting economics or keeps it anchored to project-delivery economics. That data point has not been published as of this window and is worth watching for specifically, rather than inferring from headcount targets alone.

What we actually know, and what we don't

What is not in question: production AI work moving through the Indian services channel has scaled from an emerging line to a disclosed metric at three of the sector's largest companies inside a single quarter, and the companies now consider that metric material enough to lead earnings calls with. What is genuinely unresolved, and will not be resolved by any single quarter's print: whether that AI revenue is net-new demand the traditional services model could never have captured, or a repricing of demand the traditional model is losing anyway — with the AI label attached to whichever share of the shrinking book still commands a premium. Infosys's own quarter contains both readings at once and does not adjudicate between them. Three companies, three metrics, and — eleven days into the season — still no way to tell, from the outside, whether Indian IT's AI transition is a growth story or a margin defense wearing growth's clothes. The FDE rollout and the October Q2 prints are the two concrete places that question gets tested next; until then, the honest position is that the archive has three real numbers and one open interpretation, not a settled one.

Sources

• 2026-07-09. TCS Q1 FY27 results press release, "TCS Financial Results Q1 FY 2027" →; Business Standard results coverage. Covered in India AI Digest, July 12, 2026 — item "TCS reports a $2.6B annualized AI revenue run rate in Q1 FY27" →.
• 2026-07-09. OpenAI, "GPT-5.6: Frontier intelligence that scales with your ambition" →; Bloomberg and Forbes coverage of the ChatGPT Work launch. Covered in India AI Digest, July 12, 2026 →.
• 2026-07-07. CNBC, "Chinese AI models are gaining ground with U.S. companies as OpenAI, Anthropic costs surge" →. Covered in India AI Digest, July 12, 2026 →.
• 2026-07-13. HCLTech Q1 FY27 results. Covered in India AI Digest, July 15, 2026 — item "HCLTech books $171M in advanced-AI revenue as Q1 bookings hit a record $2.4B" →.
• 2026-07-23. Infosys Q1 FY27 press release →; Analytics India Magazine coverage. Covered in India AI Digest, July 21 and July 24, 2026 — items "Infosys posts Q1 FY27: AI revenue climbs to 8.2%, but full-year guidance narrows" → and "Infosys puts AI at 8.2% of revenue as it trims FY27 guidance to 1.5–3%" →.
• 2026-07-24. Analytics India Magazine, reporting on TCS's forward-deployed engineer program. Covered in India AI Digest, July 25, 2026 — item "TCS and Indian IT build forward-deployed engineer programs as the services model adapts" →.