IT Services: Caught in the narrative war - Kotak Institutional Equities report summary

IT stocks have taken a beating due to heightened fears of revenue deflation from GenAI. Fears are triggered by recent model releases from frontier AI labs and the growing use of agent-based workflows. Model improvements so far look incremental and broadly in line with our expectations, but the sharp stock move reflects deeper concerns about the long-term relevance and longevity of IT services. In our view, markets are discounting disruption far more than current evidence supports. Growth rates (US$ terms) to perpetuity implied at current stock prices for incumbents stand at 2.7-3.8% and ~6-21% during the high-growth phase for challengers (Exhibits 4-5). Continued stock price correction can provide an attractive entry into quality challengers (possible beneficiaries), while stocks such as TCS, TechM and Infosys now offer reasonable value.

Sharp stock price decline; once again fears are extrapolated

IT services stocks have declined 2-21% in February 2026 (Exhibit 1) due to concerns about sharp effort/revenue deflation in services owing to GenAI adoption. The sell-off is global and spilled into software and other service industries. The concerns are due to (1) the rapid pace of development of frontier GenAI models and agents, especially in software engineering and (2) the assumption that improvement in capabilities is generalizable to other IT services segments. The risks are understandable, though overstated. Deflation, in our view, can also co‑exist with healthy enterprise tech spending.

Recent LLMs from Anthropic and OpenAI are incremental, not revolutionary

Exhibits 2 and 6-11 compare Opus 4.6 and Codex GPT-5.3, the latest models released by Anthropic and OpenAI, respectively, based on company releases and benchmarking from external players. Models improved on the capabilities in software engineering and are state-of-the-art, but benchmarking results align more with incremental improvements rather than a large step-up in capabilities. There are standout cases, but the average experience remains mixed. Improvements appear to be larger in other areas such as computer use. It is difficult to justify expectations of a large disruption to the IT services industry from version upgrades with meaningful, but incremental improvements.

What is baked into current stock prices?

Exhibits 4-5 detail implied growth assumptions in current stock prices. Assuming 11-12% cost of equity, 2% Rupee depreciation and an RoE range of ~20-60%. The implied Rupee growth rate embedded in stock prices is particularly low for TCS at 5% (3% in US$). Infosys' growth rate embedded in stock prices is at 5.2%, while for HCLT it is 5.8%. These growth assumptions are fairly moderate related to the overall market size and opportunities. Challengers such as Persistent and Coforge have a higher implied growth rate of 21% and 16.4%, respectively, over the next 10 years, followed by 5% growth to perpetuity. Note implied growth is higher for challengers due to large use of cash for acquisitions, leading to lower RoE. TCS, Infosys and TechM screen well. Coforge and Hexaware are also attractive, in our view.

Our current estimates already bake in incremental improvements

Incremental GenAI model improvements are more in line with our estimate of a 2-3% downside risk to growth over ~3 years from GenAI adoption (Exhibit 3). Peak of headwinds will be in FY2027E. We believe FY2027E could also be the year of peak pessimism over IT stocks. Downside risks are already partially baked into our estimates. Applications, software engineering and customer service BPO have higher vulnerability. Productivity gains will be lower in brownfield deployments, IMS and vertical BPO. AI agents and agentic AI can have a larger impact, but low accuracy, reliability and security issues can constrain holistic adoption in the near term. Software engineering is more vulnerable to the disruption from agents and agentic AI though.

Incumbents versus challengers; IT services versus ERD and BPO

Stock price decline has been high across IT services, BPO and ERD stocks and across large and mid-tier firms. The impact of GenAI will be varied, though. Incumbents will bear the brunt of headwinds. Quality challengers can gain share and still benefit. Vertical BPO and ERD firms will face a lesser impact. ERD is less impacted but suffers from growth challenges due to the slowdown in the auto vertical.

Sharp stock price correction reflects sharp change in assumptions on future of IT services stocks

IT services stocks have corrected sharply and now the discount outlook for some that look overly pessimistic. Cognizant trades at 11.5X CY2026E earnings. For a US$ P&L, a cost of equity of 8.5% is a reasonable one. A no-growth multiple for this cost of equity stands at 12X, pretty much where CTSH is trading. This is despite management guiding to 2.5-5% organic revenue growth and 4-6.5%, including inorganic initiatives for CY2026. In effect, the market is assuming that current growth is either temporary or unsustainable and that earnings will struggle to compound beyond CY2026.

This is a strong conclusion for a company that remains a capable global IT services player, with scale, deep client relationships and relevance across multiple service lines. An ex-growth valuation is not simply a statement of near-term uncertainty; it is a statement on terminal value. The market is effectively pricing Cognizant as a business where service spending growth fades over time, or even worse, it reverses.

Taken to the sector level, this logic implies something even stronger. It assumes that incremental global tech spending will accrue almost entirely to chips and hardware and frontier model companies, with limited or no incremental spending flowing through to IT services. While value capture within services may shift and pricing pressure is real, it is difficult to argue that enterprise technology transformation can bypass services altogether. The assumption embedded in current valuations appears extreme for a few cases such as CTSH.

A similar framework applied to TCS is instructive. The stock trades at 18X December 2025 quarter annualized EPS. Assuming a cost of equity of 11% and RoIC moderating to 50% (well below the current 70%), the valuation implies perpetual revenue growth of 3%, mirrored at the EPS level, plus 2% annual benefit from rupee depreciation. These assumptions are not aggressive.

That said, valuation comfort alone is unlikely to drive a rerating. The current sell-off reflects fear around the durability of the business model in an AI world, not just near-term earnings risk. For sentiment to turn, companies will need to demonstrate visible and sustained growth acceleration. Moderate improvement at Cognizant and a minor improvement in Capgemini (1.4% organic growth in CY2025 and guidance of ~2-4% with a midpoint of ~3%) have indicated growth acceleration in a year of peak GenAI headwinds. Healthy guidance from the likes of EPAM and Globant will further help halt the narrative of terminal decline. The near-term challenge is that CY2026 will likely see higher AI-led deflation as pilots move into production. Whether growth can accelerate in a year of high AI-led deflation will determine the next phase for the sector.

Realization of existing downside risks to current assumptions may not help boost investor sentiment

We expect industry growth to accelerate a tad in FY2027E due to the lower impact of macro uncertainties and the GCC shift, with the impact from GenAI headwinds partially baked in. A possible downside scenario is that the incremental benefit from macro is limited, the GCC shift continues at a similar or only slightly lower pace and GenAI headwinds need to be fully baked in. These could lead to a scenario where the revenue growth is lower in FY2027E than FY2026, which may not help in the recovery of multiples. Note that this downside scenario does not build in any additional headwinds from GenAI adoption beyond the expected 2-3%. Cognizant's guidance is the opposite of this scenario. Healthy guidance or outlook from other firms will help quell investor fears.

Capability improvement in GenAI models may not be fully generalizable

GenAI LLMs have become significantly better compared with the models from 2-3 years ago. Improvements have been focused on a few areas-reasoning, math and coding. One of the elements fueling fear related to the impact of GenAI models on services is that the output of LLMs has gotten much better in a few task categories. It is only a matter of time before similar improvements will occur in other categories. For example, LLMs have become quite adept at greenfield software development in some cases where the human effort has reduced to a large extent. The fear is that a similar amount of effort deflation is possible in (1) all cases of greenfield software development, (2) allied areas such as brownfield development, (3) adjacent areas such as application maintenance and infrastructure management and (4) unrelated areas such as legal and finance.

We believe that sharp capability improvement in a few areas cannot be generalized to a larger superset of tasks. GenAI models tend to be trained extensively to master tasks in select areas. This may not provide similar capability improvement in other areas. In fact, capabilities can even erode. For example, OpenAI noted that a focus on coding in the GPT 5.2 model could have led to a deterioration in writing-related tasks. As such, effort deflation in a few areas of IT services is not generalizable to all segments.

A possible question is whether GenAI models can be trained specifically for each task category. This is possible and is currently being undertaken by frontier AI labs and various GenAI startups. We do not have a definite answer currently. We expect better clarity in the next 1-2 years as we get a better picture of the costs and benefits involved.

Things to watch out for-where we can go wrong in our thesis

Our estimate of a 2-3% downside risk to growth over 2-3 years from GenAI adoption could undergo a change in the case of a few factors, which we discuss as follows:

- Higher-than-expected productivity improvements from GenAI models and tools. We expect ~16% revenue deflation on existing services spends on a gross basis over three years. This is based on estimates of productivity benefits in various segments (Exhibit 3). A higher-than-expected benefit will lead to higher deflation, either from faster progress in models or an increase in the reliability of agents and agentic AI in IT services tasks.

- Lower reinvestment of GenAI savings. We have assumed that 50% of the productivity savings from GenAI gets reinvested back into services spends to drive cloud and data modernization and automation. Commentary from enterprises so far has been consistent with the assumption of significant re-investments.

- A combination of a higher spending squeeze on services and overeager vendors. Higher spending on hardware and resiliency in software spends is leading to a squeeze on services. Higher pressure on services spending through strict cost controls and requests for pricing cuts from vendors will create further downsides when combined with enough vendors who are willing to go along with client requests to maintain or gain market share. This might be happening with select clients in select portions of work currently. Expansion of the same to a large portion of the market will be a challenge.

- Ramping up productivity in IT services becomes a key medium-term priority for frontier AI labs. Our expectation is that frontier AI labs after 1-1.5 years will deprioritize software engineering and focus on other sectors such as healthcare and finance. This will lead to lower productivity increases in software engineering and an increase in AI for business use cases. On the contrary, a continued focus on automating software engineering and broader IT services beyond 1-1.5 years will present additional downside risks.

Source : Equity Bulls

Keywords

ITServices KotakInstitutionalEquities ReportSummary