Hi, I’m Kashish and I work on Plotlines.
It builds on Chatter, but with a more structured lens. Instead of looking at management commentary from earning concalls in isolation, we track a single theme across companies and over time to connect the dots. The goal is to piece together how narratives evolve, and surface the deeper structural shifts shaping industries.
Today, we cover how AI is changing the rules of Indian IT services industry. This edition couldn’t have happened without the help of my colleague, and in-house IT sector expert, Pranav Manie, so a huge shoutout to him.
For nearly three decades, the Indian IT services pitch was simple to draw on a whiteboard. Take work that costs $150 an hour in New York, route it to engineers who cost a fraction of that in Bengaluru, keep the spread, and scale by adding more engineers. The model was so durable that the listed Indian IT industry — TCS, Infosys, Wipro, HCL, Tech Mahindra and the next tier — grew into a $250-billion-plus export machine on the back of it.
FY26 was the year that loop got publicly disowned by the people running the companies. Not because clients stopped buying services — bookings, by and large, held up — but because management itself began naming the model’s mortality on earnings calls. Some called it deflation. Some called it accretive. Some called it disruption they were inflicting on themselves before someone else did it to them. The vocabulary varied. The shape underneath did not. What follows is FY26 in the words of the operators, across five fractures in the old way of doing business.
[1] Revenue and headcount come apart
The first fracture is the load-bearing one. For three decades, headcount was a leading indicator for revenue at every Indian IT major. In FY26, the link visibly broke. Here is HCL’s CEO C. Vijayakumar, in plain numbers:
“If you see our revenue in the last couple of years, we have grown 4%-5% and our headcount has not grown. So, that gives you a sense there is some non-linearity playing out. Even this quarter, in revenue growth and headcount, there is at least 1.5% or 1% difference.”
— C. Vijayakumar, CEO & MD, HCL Technologies | Q2 FY26
Two years of growth without bodies. The word the industry has started using is “non-linearity” — a clean piece of jargon that translates to: the old equation, where one more engineer billed one more set of hours, no longer governs how revenue is made. Mphasis says the same thing in its own language:
“This naturally means that there is certain amount of de-linkage between revenue growth and headcount growth, which, again, we’ve been seeing for the last few quarters.”
— Nitin Rakesh, CEO, Mphasis | Q2 FY26
The most uncomfortable version of the same point came earlier in HCL’s year, when Vijayakumar was asked what happens to the people whose work is now being automated:
“Of course, we have had a good amount of people released due to the productivity improvements. Now, not all of them are readily redeployable because the requirements for some of the entry level or lower end skills are being addressed through Automation and other elements.”
— C. Vijayakumar, CEO & MD, HCL Technologies | Q1 FY26
“Released” is the operative word. So is “not readily redeployable.” The Indian IT majors employ a combined two million people. Even small percentage shifts in redeployability translate to large absolute numbers of careers that have to find a new direction.
Infosys is the outlier — but the way it’s framed is the more interesting part. CEO Salil Parekh confirmed Infosys had added 13,000 people through the first three quarters and would keep adding:
“For the year, we have added ~13,000 net headcount for the first three quarters. My sense is, we will continue to add headcount as we go through. And it sort of comes back a little bit to an earlier discussion we were having, which is there is a macro element and there is an AI element.”
— Salil Parekh, CEO & MD, Infosys | Q3 FY26
Read carefully, that’s not a defence of the old model — it’s a hedge that the macro is still hungrier for engineers than AI is yet capable of replacing them. The decoupling is happening at HCL and Mphasis already. At Infosys it is, for the moment, being absorbed.
The cleanest data point on the decoupling, though, comes from a company that didn’t appear in this set of calls. Coforge — a mid-cap that has had the strongest growth print in the listed Indian IT universe this year — grew revenue roughly 30% in FY26 while its employee cost base grew only around 20%. The gap between the two lines is the entire thesis of this section, expressed in a single income statement. Coforge management hasn’t framed this on a call the way HCL’s Vijayakumar has, but the numbers describe the same phenomenon: the unit of revenue is detaching from the unit of labour.
[2] The margin paradox — same phenomenon, two stories
The strangest thing about FY26 is that the same productivity gain shows up on one CFO’s slide as a discount the company is forced to give clients, and on another’s as a price premium the company is finally able to charge. The cleanest illustration of the dichotomy is that Infosys’s own CFO articulated both views on the same earnings call.
On the accretive side, Jayesh Sanghrajka pointed out that Infosys’s pricing has actually firmed up — and credited AI for it:
“In terms of pricing, I think pricing environment for us has remained stable. On the contrary, actually, most of our growth this year has been pricing-led because the volumes have been softer. And that in a way corroborates with the fact that the AI revenue are coming at a better pricing.”
— Jayesh Sanghrajka, CFO, Infosys | Q4 FY26
And then, a few questions later on the same call, on the deflationary side:
“Market is competitive. As I said, the competitive intensity in the market has gone up and the productivity will get passed back to the client largely.”
— Jayesh Sanghrajka, CFO, Infosys | Q4 FY26
Both statements are true. New AI-led work commands premium pricing because clients are still figuring out what it’s worth. Existing services, where the productivity savings are quantifiable and the competition can match the offer, give the savings back. The phrase “AI is accretive” and “AI is deflationary” describe the same underlying economics applied to two different parts of the book.
HCL has taken the most explicit deflationary view in the industry — to the point of pre-announcing the revenue it expects to lose:
“I think we are being very transparent. We are telling the clients, if you allow us to use AI Force and use all the recipes that we’ve created, we will showcase to you the optimization that is possible. And it will mean some reduction in revenue for us. And we are okay with that.”
— C. Vijayakumar, CEO & MD, HCL Technologies | Q1 FY26
By the end of the year, Vijayakumar had put numbers to it. A $100-million deal in the old shape, he said, was now closing for closer to $80 million:
“I mean, $100 million deal would be much lesser today - maybe 80 million, just on a rough ballpark. So, deal TCV is flat. But technically, it does require at least 25%, 30% more effort to convert and get to the same number.”
— C. Vijayakumar, CEO & MD, HCL Technologies | Q4 FY26
And, taking a longer view, he sized the structural drag on HCL’s own portfolio:
“If we look at the industry today and categorize it, 40% of the industry runs the risk of being disrupted by AI and can shrink 3% to 5% CAGR for a few years and can eventually be 25% of the enterprise spend... The 3% to 5% deflation that I mentioned in the AI disrupted services, based on the mix of services that we have, it would translate to 2% to 3% for our portfolio.”
— C. Vijayakumar, CEO & MD, HCL Technologies | Q4 FY26
To make room for that pivot, HCL ran a restructuring program through the year. The cost showed up plainly on the Q4 P&L: a reported operating margin of 16.5%, against an underlying 17.7% — a 120-basis-point drag the company chose to absorb in a single year.
Mphasis sits in the accretive camp, and Nitin Rakesh has been the most forceful about why:
“The good news though is that because we have this approach, we are at least not playing the pricing game alone when it comes to winning business. We are playing the Savings-Led Transformation game, and while we win business, we don’t have to sacrifice profitability of those deals because we are using this as a leverage.”
— Nitin Rakesh, CEO, Mphasis | Q2 FY26
TCS, predictably more measured, made the same point on revenue productivity but flagged the timing wrinkle — early-stage AI delivery still carries investment costs that distort margin comparisons:
“On the AI and data part, the revenue productivity is definitely much better than the TCS average or the traditional business, both at onsite and offshore. Margins, I will not call out because there would be investments which would be temporary or in the initial phase, so it wouldn’t be like-to-like for comparison.”
— Samir Seksaria, CFO, TCS | Q4 FY26
The two camps are not actually contradicting each other. They are describing different ends of the same portfolio: new AI work prices well because it’s scarce and unmeasured; old work that has been re-priced under AI savings clauses gives margin back. The companies whose CFOs say “accretive” are the ones whose mix is tilted, today, toward the new end. The ones who say “deflation” are sizing the headwind from the old end. FY27 will tell us which dominates as the new work scales and the old work gets re-papered.
There is, however, a related move that two of the larger players have begun making — and it is the most concrete signal yet that they are willing to defend pricing with the one lever services firms rarely pull: walking away from deals. HCL’s Vijayakumar disclosed in Q4 that the company had voluntarily declined to chase a meaningful share of available pipeline:
“We have lost some deals which are voluntary losses. We have walked away from some deals which will not make sense and that would have easily contributed at least $1 billion more to this number. It’s only prudent to be a little bit more careful about this...”
— C. Vijayakumar, CEO & MD, HCL Technologies | Q4 FY26
A billion dollars of foregone TCV is a serious admission for an industry that has measured its quarterly press releases in TCV growth for two decades. Tech Mahindra has been making the same call, and saying so explicitly:
“We have stayed extremely disciplined in large deals. So we have stayed disciplined so that large deals don’t end up creating a problem for us in the future, or doing large deals that don’t make business sense. Clients don’t want us to do deals that don’t make sense either, right?”
— Mohit Joshi, MD & CEO, Tech Mahindra | Q4 FY26
Tech Mahindra’s CFO Rohit Anand confirmed the same posture, in CFO language:
“We’re extremely conscious on what margins and the risk profile we sign it up... [we’ve] been very selective not just on the two large deals that we’ve announced but even on the deals that we’ve been ramping up for the last 3-4 quarters. Our as-sold margins on each of these deals from a portfolio perspective are accretive.”
— Rohit Anand, CFO, Tech Mahindra | Q4 FY26
Notably absent from this chorus is Infosys, which spoke of “financial discipline” and a “margin protection programme” through FY26 but did not, on its Q4 call, name any deals it had walked away from. The question this raises is not whether HCL and Tech Mahindra are bluffing — both have given up real TCV to make the point — but whether either of them can keep doing it through FY27. Walking away from a billion dollars of deals is a luxury in a year where reported growth was already supported by macro tailwinds and currency. If FY27 starts with softer demand and the competitive set begins matching whatever pricing HCL refused, the discipline will be tested in a way it wasn’t this year. Investors should watch quarterly TCV disclosures with that filter on: is the deflation in deal sizes coming because AI is shrinking the work, or because HCL and Tech Mahindra are choosing not to take it? Through FY26, both stories have been true. They may not stay true together.
[3] The pricing reframe
If the labor-hours model is in retreat, what replaces it? The honest answer from FY26 is: no one knows yet, but everyone is auditioning a candidate. The sharpest reframe came from Tech Mahindra’s CEO Mohit Joshi — the most transparent any large-cap Indian IT services chief executive has been about the pricing logic of the AI era:
“The way of thinking about running an AP function for our client, for instance, in the age of AI, is thinking about the work overall in terms of the number of service tokens that you will deliver to a client. So a service token in the context of an AP could be a sub-process of AP that you need to deliver for a telco. And as the combination of human labour and digital labour changes over time, and as the pricing for digital labour changes over time, the result is very transparent to the client.”
— Mohit Joshi, MD & CEO, Tech Mahindra | Q4 FY26
The “token” framing is borrowed, deliberately, from how large language models are priced. A unit of work, abstracted from how it gets done. Whether the work is performed by a human in Pune, an agent on a GPU, or some blend of both, the client buys outcomes by the token. Notice what it strips away: the offshoring rate-card. There is no longer a “billing rate” in the traditional sense.
Mphasis has taken a different route to the same destination — selling the savings, not the hours:
“Clients are also asking us to not just show me on a PPT or tell me but actually show me in a live sandbox environment in many cases. So, think of this as ‘RFPs are turning into hackathons’, and that’s their yardstick of who can deliver on what they’re promising versus not. So, it has in a way become a lot more about the ability to showcase through execution.”
— Nitin Rakesh, CEO, Mphasis | Q2 FY26
That sentence is worth pausing on. RFPs — request for proposals — have been the industry’s procurement language for thirty years. A multi-hundred-page document goes out, vendors respond with a multi-hundred-page document back, the decision is made on the document. If RFPs are being replaced by live sandbox demonstrations, the selling motion itself has changed: showing the working AI agent in the room beats describing it. The salesforce that wins is the one with engineering in the field, not slideware in the back office.
Infosys, more cautiously, hinted that pricing is in transition without committing to where it lands:
“Over a longer period of time, on the back of AI, etc., we may expect some part of newer pricing models emerging. It could be outcome-based pricing model. It could be pod-based or studio-based pricing model, etc. So there are various new pricing models that are emerging as we speak. I do not think over the next year or so the entire model is going to change.”
— Jayesh Sanghrajka, CFO, Infosys | Q1 FY26
“Pod-based” and “studio-based” are worth unpacking briefly. A pod is a small dedicated team — sometimes ten people, sometimes three plus an agent stack — billed as a unit rather than per head. A studio is a longer-running engagement priced on capacity and outcomes. Both share a common property: the client never sees an hourly rate.
[4] From labour to platforms and IP
If the work is no longer priced by the hour, where is the leverage? Every CEO in the brief converged on roughly the same answer: stop being a pure services firm, start owning intellectual property and platforms that get embedded into client environments. HCL’s Vijayakumar said it most directly:
“We believe this industry will have to evolve from being a pure labor-based service provider to people plus IP and platform-based service provider. When you have the platform as a third-party platform, there is very little leverage, very little stickiness that we can build. And we can really deliver very good quality vertical IP solutions, which can be replicated across customers.”
— C. Vijayakumar, CEO & MD, HCL Technologies | Q2 FY26
The argument is structural. A consultancy that uses someone else’s platform — Microsoft’s, Salesforce’s, ServiceNow’s, OpenAI’s — is renting leverage. A consultancy that builds its own platform, even a narrow vertical one, captures the leverage. HCL’s AI Force is the lead exhibit; by the end of FY26, the company reported $155 million in quarterly Advanced AI revenue, up from a $100-million annual milestone just two quarters earlier.
Wipro, in the most explicit organizational signal of the year, set up a separate business unit around the same thesis:
“As intelligence becomes industrialized and widely accessible, we are making a deliberate strategic pivot to stay ahead. We have launched a dedicated AI-native business and platforms unit to expand beyond a services-only model to a services-as-a-software approach. This unit will operate with dedicated leadership, focused investments and a distinct operating model to accelerate enterprise-grade agentic AI solutions.”
— Srini Pallia, CEO & MD, Wipro | Q4 FY26
“Services-as-a-software” is the phrase to watch. It is the inverse of “software-as-a-service” — instead of subscription software that requires services around it, the service itself is delivered as software that runs continuously. The CEO described the resulting structure as a “dual engine”: traditional services on one side, AI-native platforms on the other.
Tech Mahindra is making the same move under a different label — rebranding entire service lines rather than spinning up a new unit:
“We are repurposing our application development and maintenance services to agentic development and modernization services and it’s not just a name change, right. It’s not just a name change because it’s about how are we driving value to the customers. Now, how are we bringing that experience in the agentic form to our customers for building application agentic development is something that we are very, very focused on...”
— Atul Soneja, COO, Tech Mahindra | Q4 FY26
TCS has stated the destination in the largest terms:
“With AI, TCS aspires to be the World’s largest AI-led Tech Services company. This aspiration is powered by capitalizing on AI-led renewals, vendor consolidation and cost optimization deals resulting in market share gains; using new-age services and adjacencies that enable enterprises to ‘Get ready for AI’; becoming a full-stack AI services player — Infrastructure to Intelligence — thereby delivering maximum ROI to clients on their AI investments; and building new revenue streams such as building AI infrastructure.”
— K. Krithivasan, CEO & MD, TCS | Q4 FY26
The most provocative version of the labour-to-IP thesis, though, came from Infosys’s co-founder and chairman Nandan Nilekani, in a line that pointed at something even further out — the threat to the SaaS industry the IT services companies have spent twenty years serving:
“As AI becomes a bigger part of the spend, the balance of advantage is moving towards ‘build’ rather than ‘buy’. If you see some of the concerns about what will happen to SaaS companies and all that, it is because of this, that building applications has become so simple that very often you may just build, or you may replace something that you have, which you bought, with something to be built.”
— Nandan Nilekani, Chairman, Infosys | Q3 FY26
If Nilekani is right, the next chapter is not just IT services repricing labour. It is enterprise software being eaten by the same wave — and the consultancies that ride it being the ones that own the new application stack, not the old one.
[5] Where the new money is
The closing fracture is the most underappreciated one in the brief. The growth pocket nobody had on their bingo card eighteen months ago has turned out to be the building, running, and feeding of AI itself — what HCL calls “Day-1 services”:
“The real acceleration in what we are seeing is not necessarily in deploying AI within enterprises, but really ‘Day-1’ services, which are foundational for enabling AI, like a lot of work in our engineering services. Like I mentioned about custom silicon for edge inferencing, it is a big area with a lot of companies across multiple industry verticals. This is not restricted to semiconductor industry.”
— C. Vijayakumar, CEO & MD, HCL Technologies | Q3 FY26
Edge inferencing is the act of running AI models on the device — a car, a sensor, a factory floor controller — rather than in a centralized data center. Custom silicon is the chip designed to do that efficiently. Both used to be niche semiconductor specialties. In HCL’s telling, they have become mainstream IT services work because every industrial company building AI capability now needs both. By Q4, HCL had a concrete deal to point to:
“A global technology major selected HCLTech for another AI Factory program worth over $100 million. The HCLTech solution will fast-track the client’s requirements of building and operating next-generation AI data centers to support cutting-edge AI workloads using the latest GPU technologies. A global semiconductor major selected HCLTech’s AI engineering services to support ASIC development across multiple advanced node chips, strengthening its position in Physical AI.”
— C. Vijayakumar, CEO & MD, HCL Technologies | Q4 FY26
TCS made the most striking infrastructure announcement of the year — a move into the data centre business itself:
“The promise we see in our HyperVault Business — which has made significant progress this quarter on its journey to build out 1 GW of capacity. This includes winning customer commitments, land parcel finalizations and partnering agreements.”
— K. Krithivasan, CEO & MD, TCS | Q4 FY26
One gigawatt of data centre capacity, for context, is a serious number. Indian data center capacity in early 2026 totals roughly 1.4 GW. TCS, a services company, has signalled it intends to build something approaching the size of the entire existing Indian colocation industry — and intends to sell that capacity to AI workloads.
The more interesting move underneath HyperVault, though, is what TCS is positioning around it. The company’s stated aspiration of being a “full-stack AI services player — Infrastructure to Intelligence” reads, at first, like investor-day boilerplate. Read against the HyperVault build-out, it describes a sequence no other Indian IT firm is set up to deliver: an AI lab buys compute from TCS, then turns to TCS again for GPU operations, then cloud management, then the model fine-tuning and red-teaming work that Wipro and Tech Mahindra are doing on a project basis. Each successive layer is more services-intensive and more margin-rich than the layer below it. Owning the data centre at the bottom of the stack gives TCS a contractual reason to be in the conversation at every layer above it. That is structurally accretive in a way that running someone else’s GPUs cannot be — and, for now, no other Indian IT firm has the balance sheet, the customer book, and the infrastructure muscle to attempt the full stack the way TCS is.
Wipro and Tech Mahindra are working a different end of the same picture — getting paid to run models for the model-makers themselves:
“In my first example, a leading global technology company has engaged Wipro to help run and improve its frontier AI models. Wipro will manage the end-to-end operation of these AI models from training, governance and evaluation to domain-specific validation. In fact, this engagement will be done through a specialized global delivery platform.”
— Srini Pallia, CEO & MD, Wipro | Q4 FY26
“When we look at our BPS services, for example, close to one-tenth of our current business in BPS is working with technology players, high-tech players creating their AI models, continuously fine-tuning them and managing them over a period of time. And more and more work we are doing on the BPS side is actually towards that now.”
— Atul Soneja, COO, Tech Mahindra | Q4 FY26
For a layer of context here: BPS — Business Process Services — used to be the lowest-margin, most automatable layer of the Indian IT stack. The fact that one-tenth of Tech Mahindra’s BPS book is now training and fine-tuning AI models for hyperscalers is a sentence that would have read as nonsense three years ago. The hyperscalers, ironically, have become the buyers that need the most armies of humans — to label data, evaluate outputs, and red-team models.
HCL has put the cleanest framework on what comes next — a three-bucket split of its own portfolio:
“We will see differential growth rates in all the three different categories: AI disrupted, AI amplified, and AI native or Advanced AI services. We really look forward to growing our AI-native services in the 25% to 30% range. And that will truly be the validation of how we are evolving as a company.”
— C. Vijayakumar, CEO & MD, HCL Technologies | Q4 FY26
Three buckets, three growth rates. AI disrupted shrinks. AI amplified grows in line with the business. AI native compounds at 25-30%. The whole question for FY27 is whether the third bucket gets big enough, fast enough, to outrun the drag from the first.
What it adds up to
The Indian IT services industry spent FY26 publicly admitting that its core business is changing shape — and putting numbers, language, and organisational decisions behind that admission. The vocabulary fractured into deflation and accretion, tokens and pods, disrupted and native — but the underlying movement was a single one. The companies that used to sell engineering hours by the thousand are now trying to sell platforms by the licence, savings by the contract, and infrastructure by the gigawatt. None of them have completed the pivot. All of them have started it. FY27 is the year the math becomes legible — when the new revenue mix is large enough to either validate the bet or expose the gap.
What to Watch
HCL’s 2-3% portfolio deflation forecast: Vijayakumar has put a number on the headwind; track whether reported FY27 revenue mix bears it out, and whether AI-native services growth (targeted at 25-30%) is fast enough to offset it.
TCS HyperVault execution: 1 GW is a number large enough to reshape the Indian data centre market. Watch for customer commitments converting to live capacity, partnership announcements with GPU vendors, and the first revenue disclosure.
Tech Mahindra’s “service tokens” pricing: No client has yet been named signing a token-priced contract. The first concrete deal at a stated token rate will be the signal that the reframe is real.
Infosys headcount additions in FY27: Salil Parekh hedged “we will continue to add” — but the gap between Infosys’s hiring posture and HCL/Mphasis’s de-linkage narrows with every quarter of productivity gain. Watch the net adds in Q1 and Q2.
Whether any IP/platform business crosses 10% of group revenue: HCL’s Advanced AI sits around 4-5% on a $13-billion run-rate; Infosys put its AI Hexagon at 5.5% in Q3. The 10% line is when “we’re pivoting” becomes “we have pivoted.”
That’s it for now! Your feedback will really help shape how The Chatter evolves. Drop it down in the comments below!
Disclaimer: We’ve used AI tools in filtering and cleaning up these quotes & narratives so there maybe some mistakes. Now, if you are thinking why we are using AI, please remember that we are just a small team of 5 people running everything you see on Zerodha Markets 😬 So, all the good stuff is human and mistakes are AI.
The content is great. It just reads a little AI-ish, distracting/diminishing interest in reading. Also the language is just a tad technical, maybe help us with examples? Otherwise lovely read!
This is a great job Kashish & team. Very minutely read and analysed with some crystal gazing. Fantastic effort !!
I would less than honest if I did not admit i did not get it all but yes I have a broad framework (very broad 😅) with which to look at circa 2027 and observe where US hypersclaers actions & india policy on energy / chips progress intersect. From the macro...this will be a very very interesting time indeed.