The AI Paradox: Why Traditional IT Services May Thrive, Not Die

[Avin]

With all the problems and challenges seen in Indian IT Sector, I also felt it necessary to think and have a perspective like the X/Twitter public that has a view on everything.

Indian IT services companies largely revolve around providing software services like developing applications for custom needs, providing support for products like SAP and others and managing full IT stack of large conglomerates especially in US and Europe as labour arbitrage plays a big role for servicing clients in those countries.

Real on-field work done by IT companies is writing codes to help manage software, make custom made software, dashboards, webpages etc. With the evolution of AI and notwithstanding other capabilities and pitfalls of AI, it seems that writing code - the fundamental work done by IT companies can be largely done by AI/LLMs. Almost all LLM based companies like Open AI, Anthropic etc claim that most of their coding work is now handled by AI and almost no human input is required.

Let me bring in a real life scenario. At my workplace as Finance professional, I am using AI, especially Claude, to get some amount of work done like writing a note and interpreting long texts (which I have no intention of reading and yet it is important to find out what the text actually says).

But perhaps most valuable help from Claude is saving time in designing a framework to put excel data, ppts and dashboards for management consumption. Trick is simple - tell Claude the objective and what exactly you want to portray in the excel/ppt/dashboard and bingo! You get a readymade framework even fully developed excel workbooks with multiple working sheets embedded with perfect excel formulas.

So what has AI done for me as a corporate white-collar executive? It seems that it has cut my job time significantly as I used to spend anything between a day or two and many times much more than that in designing a framework on excel to work with. Actual inputs/work anyways took lesser time earlier, thinking mostly went into how to arrange and present the data so that someone who reads can make sense - this is where AI comes now.

This personal productivity shift has resulted in organizational consequence. My demands from my IT department is to buy Enterprise Version of Claude, set up Python on every computer so that repetitive task can be taken care of by individuals through Python/coding etc with the help of AI. The IT department also have to buy GPUs, good grade hardware computers etc to help sail through this tsunami of AI coming to the system infrastructure.

So what the above change has brought for traditional IT services? Apart from purchasing the AI stack, the IT services team is now deployed more vigorously to develop advance dashboards, build an in-house LLM for the proprietary data of the company (seen across peer companies also), help identify which all tasks can be automated through AI or other means, how to combine data from various sources and present it in a uniform way to utilize for better decision making.

With the above additions to the demands on the IT department have resulted in serious pressure on IT department to deliver results fast. It is a matter for another day that so much data results in analysis paralysis!

By extension, demand for hardware rises exponentially, demand for “high end” analysis and custom designed software, dashboards etc. also rises significantly. The demand for managing the standard products like SAP is same as earlier and demand for really low end engagements like writing a small program goes away.

Demand for writing a small program anyways would have been insignificant for the big IT corporations. The other points above make a case to actually increase the IT services demand leaps and bounds rather than the other way around.

Further, I will like to add two learning from two great thinkers I follow closely.

First, I was reading Michael Maubossin’s one of the latest papers where he directly argues that AI and tech companies’ announcements of huge capex running in billions of dollars is bound to have time and cost over runs. The paper is based on the huge capex projects undertaken in the past where more than 90% of the projects underestimated the cost and time to build big things and overestimated the potential gains almost every single time.

Second, Taleb has also warn through his writings so many times that the bigger you are the more fragile you become. The bigger the project is, chances are it will miss the cost and time initially decided and overestimate the benefits of the same.

With the above context, I believe that traditional IT acts as a baseload to develop and increase the demand for AI. Further, potential delays in data centre construction and other related hardware will result in optimized management of full AI stacks, optimization of available sources and significant inflation in AI related costs.

With demand already rising for elite requirements courtesy AI as explained above, companies will return to traditional IT with labour arbitrage on cost to deliver.

This is a nice loop where hardware, new AI based software and old software everybody benefits, and the end consumer actually gets killed due to inflation all around caused by energy and commodity hungry AI, wars and build-ups post wars. Discretionary corporate consumption shifts to AI and IT spending, utility bills now include at least one AI subscription of $20/month. Even if roughly 60% of population of Asia, America and Europe subscribe to the $20/month, the annual subscription charge towards AI will cross $1 trillion per annum. This $1 trillion per annum will in all probabilities reduce the discretionary spend on other items.

But AI needs physical infrastructure, energy and training and what not handle copious amounts of data processing capacity. World may not simply have so many resources to start with. May be that’s why Elon Musk talks about Kardashev Scales so often as solution to this problem lies in Kardashev Scale utilization only.

So till the Kardashev Scale breakthrough is reached, traditional IT may even hit growth spurts from the current depressed growth status and energy and materials may keep delivering loads of cash flows upfront.

[vikas_chandrawat]

I subscribe to your thought process and agree that instead of disrupting, AI will in fact aid the large traditional IT companies grow. I believe AI is replacing jobs that are computation heavy but judgement light. For example, coding is computation heavy, but deciding what to build in the context of a business outcome is judgement heavy. AI can replace the coders required for one such solution, but the business analysts and product managers don’t go away — they in fact get empowered because their scope of work just increased without a similar increase in the number of hours. This is ground-breaking for traditional IT.

My wife used to work at Tata Croma and the biggest problem she faced during her tenure was the collaboration gap from the TCS tech guys — you were entirely at their mercy to deliver what you needed as solutions for your brand. TCS guys were just coding and their presence in meetings was just to provide a feasibility go-ahead. I think AI making coding, testing, and designing so lightweight will unlock the potential of so many talented product managers, business analysts and team managers who are part of these large IT firms but are currently blocked by low-quality engineering delivery.

This does require IT companies to navigate this new era and adapt accordingly. If they keep playing on labour arbitrage instead of utilizing new tech to improve internal operational efficiencies, the path can be harder. Wipro seems to be heading this way — they just announced a ₹15,000 crore buyback with their promoters, who hold ~73% of the company, set to participate, on top of an 88% dividend payout ratio in FY26. These are not signs of a company confident in its ability to adapt and invest in the next wave. TCS and HCL Tech, on the other hand, seem to be taking this battle more head on.

I am really excited about how this will play out. It will be the story of the decade irrespective of the outcome, and we are lucky to see it unfold in front of our eyes. I took a small tracking position in TCS to have some skin in the game and keep a close eye on developments in this sector.

[Avin]

Moving on from making a guess about what I know of AI to what I actually know –

1. $ 700 billion committed/announced capex by big AI players and hyperscalers like Microsoft, Alphabet, Amazon, Meta, and Oracle.

2. Amazon spending $200 billion in 2026 beating even very bullish estimates.

3. The above cost is bound to rise as bigger projects = bigger cost overruns and bigger time overruns.

4. Benefit will at least be delayed, I don’t want to argue on the quantum as it will be anybody’s guess.

5. AI becomes the interface layer for all the software – why, how and what to operate – AI will throw you instant solutions

6. Lower end/beginners’ software profiles shrink drastically

7. More noise, mediocre outputs all over social media platforms. Harder to recognise wheat from chaff.

8. Winner takes all… more pronounced in every area and every field.

9. AI becomes a utility like electricity eventually.

[rpattabi]

The edge our IT services used to have pre-AI era was large pool of people.

Now that AI reduces the need for such head count, our IT services now have serious competition, particularly GCCs (Global Capability Centers).

• GCCs allow businesses to have dedicated service provider exclusively.
• Businesses in the process of taking advantage of AI need not expose their critical data to third party service provider thanks to their GCCs.
• GCCs allow businesses to continue to tap into cheaper labour pool.

Also even smaller IT service companies can compete with larger ones (our assumption being head count is no more a constraint). This will impact the margins significantly.

---

On the other hand, the following are also to be considered:

• Biggest cost is head count for our IT services. Now that with smaller head count is needed to achieve similar productivity, it might help with the margins, even if clients push for some of the productivity gains through AI to be passed onto them.
• Since writing code becomes cheaper, there might be explotion of software. Software could be considered in areas where earlier it was not feasible due to cost. This could bring in much more work.

---

It is just very hard to predict how things will pan out for our IT services. It is better to bet on this with decent margin of safety and track how the business is progressing quarter by quarter.

Disclosure: Tracking position

[vikas_chandrawat]

The GCC angle is something I didn’t consider so far. It is a genuine risk. But like you said there are just too many ways things might play out. Very important to be vigilant about playing this narrative. Best way forward is tracking the developments quarter on quarter.

Disclosure: tracking position in TCS

[Capt_Cool]

A few thoughts on how tech is changing and its impact on India:

1. The world is moving from Software to Hardware, think SaaS/product to Chips/Energy/Power. India is behind, and the government push and private sector risk capital is the only way forward for growth.

2. While Indian IT services are shrinking, GCCs are booming, I believe they are a bigger threat to these services companies than AI is.

3. Expecting them to build large-scale AI models or infrastructure using their capital is just wishful thinking. Their culture, business model, and people are not made to take moonshots or risk shareholders’ capital.

4. But what they do is still very important and provides value to the world at large. No Bank or Retailer or even a Government, large or small, would ever be confident to build, implement, and maintain their tech fully in-house.

5. IT services will eventually catch up as they are very good at implementing tech at a cheaper rate, think Y2K, cloud, and any other tech transformation in the past.

6. They might have a different billing strategy, but they will still be here even 20 years from now and will keep hiring cheap labour to implement the latest AI agent for the Fortune 500 clients.

7. Some things never change, they just have a bigger cycle time, and this cycle shall also pass.

[pkeday]

Subscribing to most of what’s been said here, but want to add three pieces of scaffolding I think the bullish case is missing.

Two kinds of disruption, and AI is doing both at once.

Most disruptions in history have been one of two flavors. The first is de-skilling: power looms didn’t kill weaving, they collapsed a master craft into something a factory worker could do. Total textile employment actually grew because cloth got cheap and demand exploded. The Singer sewing machine: same story, more optimistic arc. Disruptive, but the work didn’t disappear. The second is elimination: telephone operators, bank tellers. Role gone, no demand-elasticity rescue.

AI is doing both, but on different layers. The bottom-most “implement this spec, write this function” work is being eliminated, not de-skilled. There is no version of that role that comes back. The architecture, the system design, the integration with messy enterprise data, the talking-to-the-business layer is being de-skilled, which historically grows employment in absolute terms because the cost collapse expands the pie. The thread is right that the high end grows. It’s worth being honest that the bottom does not.

The China-automation parallel is the cleanest historical analogue.

For years the fear was that as robots got cheaper, China would lose its manufacturing edge. Why pay a Chinese worker when a robot in Ohio can do the same thing? The opposite happened. China embraced automation faster and harder than anyone, over two million factory robots today, more than the rest of the world combined, industrial robot output growing nearly 30% YoY. The incumbent with the deepest operational knowledge didn’t get displaced by the new technology. They adopted it first and pulled further ahead.

Indian IT has the same setup. Thirty-five years of enterprise process knowledge, edge cases, and client trust that nobody else has bothered to accumulate. The labour arbitrage at the new skill level still holds. And these are exactly the players with the most to gain from aggressive AI adoption inside their own delivery, because their margin structure is most exposed to it.

The industry survives, but it probably looks quite different.

The shape of the IT services business in 2030 is unlikely to be the shape we know today. The pyramid likely flattens. The wide base of fresh engineering hires that built every Indian IT company we know probably contracts meaningfully. Headcount-as-moat weakens. Billing could shift away from time-and-materials toward outcome-based contracts, since clients won’t keep paying for hours when the hours are AI doing the work. Revenue per employee likely rises sharply, total employee count comes down, and total revenue can still grow on volume. A different business, not necessarily a smaller one, and the companies that survive in good shape are probably the ones that recognise this early enough to repaint themselves before the market repaints them.

But industry survival is not a buy signal for individual companies.

The bullish thesis on this thread is mostly a thesis about the industry. All of that can be true and individual stocks can still deliver poor returns. The transition will expose huge variance between companies that adapt and companies that defend; both will report decent numbers for a while because existing contracts are sticky, and the divergence will only show up when the next generation of contracts gets awarded. And the price you pay matters: if the “Indian IT thrives on AI” narrative is already in the multiple, forward returns get decided by multiple compression more than by earnings growth. The question to track isn’t whether the industry survives. It’s which specific companies are repositioning fast enough, and at what valuation the market lets you buy that repositioning.

[vikas_chandrawat]

Very interesting points all around in this thread. Either AI disrupts the traditional ITs or rebuild them. Those of you tracking these traditional ITs, what specific KPIs or signals are you looking at to anticipate which one of them is adapting AI to transform as a business and which ones are just sitting ducks?

[Avin]

I am not an IT guy, no in-depth knowledge about IT KPIs etc, but what i will like to track is :

1. Hirings slowdown by IT companies, esp Persistent etc as they are growing companies.
2. AI adoption by companies - can some services companies convert themselves to product instead of services (far fetched by if at all somebody can do, will create a lot of value)?
3. IT product companies can be good bet. I will study the likes of OFSS although I hold no position in the same. Need to see the impact on them due to AI.

[StonePitbull]

My Two cents - the metrics one should track are margins and deal pipeline / sizes. Advancements in AI have made IT Services deflationary as the playing field gets leveled and competition intensifies. Players start to undercut each other on pricing to fill their deal pipelines, which will lead to margin squeeze. The players who refuse to take price cuts and wait for big deals may not see much growth in deal pipeline.

TL;DR price wars to fill capacity → margin squeeze, or waiting for premium “transformation/AI-led” deals → potential growth slowdown in the interim.

[Avin]

AI is just a word prediction tool backed by statistics and tech, currently used to mechanize/automate repeatable tasks and those tasks where we need to copy paste from the web/other sources. Earlier, search - copy - paste then apply permutations combinations from different sources for complicated tasks of same nature. Now you just type a good prompt, correct it in LLM itself multiple times till you hit what exactly you want. Essentially some sort of labour replaced by machine.

When an enterprise require a real solution - AI is not seen delivering end to end things till now at least. What is actually rising is IT spending including cyber security, software, hardware, build-ups on standard products, custom build-ups like Small Language Models on corporate data etc. For me, this looks like IT may become value accretive. Market as always will focus on what is ahead and till the margins and deal sizes change, prices will factor in those variables before hand.

[Investor_No_1]

I will refrain from blabbering a long answer and give a 20 page gyan on AI or tech.

Short answer is - do you consider these companies as some specific tech or real tech companies? For AI is just another tech, a real impactful one & certainly detrimental if you ignore it, but after all, its just another tech.

That would answer the survival part.

Coming to the journey - This is highly volatile at moment. You might get some clues if you think as a business owner by putting yourself in promoter’s shoes….from strategic decisions at enterprise levels rather than QoQ details….

Disc: Invested in IT companies as a basket hence biased & critical. Transactions recently. Not a buy/sell recommendation. Not eligibles for any advice. Certainly not a registered advisor. Post only for learning purposes. I can be wrong in all my assessments.

[Deven]

Great discussion. Thanks all for your active participation and educating us. I have query on AI capex.

Currently, it appears that leading AI companies are collectively planning to invest around $800 billion this year alone, with expectations of continued high levels of investment going forward.

Baseline ROI (Non-AI Business)
What kind of ROI are these companies currently generating from their traditional (non-AI) businesses, such as cloud services?

Required Revenue / profit for Justified Returns
Given the scale of AI investments, what level of incremental revenue or margins would be required to generate a reasonable return on capital?

[Divyanshu_Taneja]

[gaurav]

Code is now a commodity. This is the real pain point of all the IT companies and workers. A very high value service/product has become a commodity. What happens now?

It has become cheaper to build products (time and money both). This should increase customer base. But will these customers go to Infosys or any other place is yet to be seen.

Meanwhile AI party should continue till openAI/anthropic brings an ipo. Till then swim!

Disc: chart created using AI :p

[Avin]

Just to play a devils advocate, ok - code is commodity just like oil. To be a commodity I need a market where I can purchase a standard code just like exact specification diesel, petrol you get from so many places in world. Where is that code market? Sold in tokens/ labour hours etc etc just like oil in barrels? And if at all such a code standardised globally available in a market traded freely, how can I use it just like that in my applications as I fill my car with petrol and press the accelerator…

[Vivek_Shetty]

I found this video talking about how LLMs are a dead end and we’re very far from AGI and actual intelligence. I would be grateful if anyone who understands this could share their opinion. As a non tech person these wildly conflicting opinions coming out of the tech world are very confusing.

[vikas_chandrawat]

This is a great perspective Avin. I have my two cents:

• First of all I don’t think coding as a standalone value was something IT companies were getting paid for. It’s the complete packaging of the solutions right from “understanding the business case and selling the solution before even writing the code and hence the long contracts” , “actually coding it and maintaining the coding operations which goes beyond just coding, designign and testing is a major-major part” & then the “delivery and maintenance”.
• “Coding getting commoditised” does’t destroy the value of traditional ITs. It empowers them. and like Avin mentioned where do you go out to buy that commodity? It’s either the IT giants or building your own GCCs. I don’t see any Banks, Insurance cos. etc. going out of their way to trust some lean-advanced AI solutions delivery cos. or establish GCCs when you can get the same solution from a long-time partner with credibility.
• There is no doubt that the existing business model will need to change. But that change is going to be detrimental to the ITs is a possibility yes, but not a certainty.

The metrics I am looking for are

• Increase in large clients (100M+ USD )
• “Stable or growing margins” coupled with “increase in USD revenue / employee”

[Avin]

Thanks Vikas,

I may be wrong and stupid, but to me it looks like this -

1. Traditional IT provides a solutions package.
   1. Earlier IT cos were helped through labour arbitrage hence good margins and bigger deals delivered at low costs to developed countries clients.
   2. Now labour arbitrage competes with AI for coding or in future providing full end to end solutions linked to software services. If AI is cheaper - cos prefer AI assuming AI is able to handle end to end services and support and customizations and endless improvements etc etc.
   3. If point 2 - AI is not able to handle the said assumption - no business is lost by IT cos, some very low value business may be lost.
   4. Now you spend north of $ 800 billion to set up AI and infra. IT cos - no hardware requirement. AI by logic of it - will search for returns very very soon; IT cos return ratios will remain high due to the nature of the business - inherently negligible capex.
   5. AI required huge infra - takes time; power generation - transmission - grid infra etc etc; then chips, GPUs etc; then local hardware to be updated - greater power CPUs etc.
   6. the above huge infra is set-up by cos like Google, Microsoft, Amazon, Meta, Open AI, Anthropic like names - many of them already have billion of dollars in cash coming from existing business.
   7. the cash required for spending hence is available.
   8. BUT, returns required/ anticipated will always be anchored by their current software based business which are astronomical.
   9. What it may lead to is - a rat race in AI to commoditize the sector - as always been seen in tech revolutions of the past - petchem revolution, railways, electricity, telephone, textiles power looms etc. All were very revolutionary technologies and changed the world for good. But the true tech product/service providers turned into commodity players due to competition eating away all the margins.
   10. AI already has so many competitors/ peers. It is bound to be commoditized going forward.
   11. the commodity called AI will be used as a commodity like - you purchase tokens, establish AI based workflows and run the business processes as required.
   12. After token purchase, AI will run on the software products available like SAP or any other ERP and provides data in nice, polished forms and will be required for analysis paralysis by corporate executives.
   13. But will AI go and meet customers? will AI go and build relations with employees? will a “predicting the next word” tool be enough to provide real end to end solutions?
2. So to me - IT remains IT with may be some reduction in lower/middle management layers of jobs, AI through tokens trade becomes a commodity and will be taken advantage in how you run and utilize the existing software products and services.
3. Critical thinking in such a commoditized and tokenized world becomes super important (it already is but increases its power leaps and bounds).
4. If IT delivers solutions through thinking like a human and coding is just a way to deliver the outcome, IT survives with no commoditization versus full commoditization may actually happen in AI.
5. And on the jobs part, to paraphrase Taleb : AI may be good for all the jobs created in 20th century and make more relevant all the jobs prior to 20th Century.

[gaurav]

Interesting anology with oil. But when did commercial oil drilling began? And when ford started its assembly plants (transportation is the biggest consumer of oil today)? There is a gap of 50 years between these two events. I don’t think we are asking the right questions.

The problem for Infosys, as I see it, even small player with <1000 mcap can compete with them on big projects now. Their advantage of higher headcounts is no longer the case. This is a problem they must solve if they wish to justify their secular growth story.

Disc: not invested in IT but looking for opportunity. I am looking for a player whose growth rate in revenue and cash sustains even with flat or degrowth in headcount.