My counterintuitive take is that the bear case on Indian IT is only half right. What do I mean when I say this? I think the AI race is an overall net positive and a good thing for India. Most takes I've read or seen are that due to the heavy services-based economy that India has landed in after the 1990s IT boom, good LLM models that code better than humans would mean that this industry is in grave danger. There is probably some truth to this. An average developer in Bangalore, India, costs about $1,000 a month to hire on your payroll or $12k annually. If all your developer did the entire time was write programs based on your instructions and you tested it, then their job is in danger.
The case for being bearish on the IT sector
Kimi K3 is a frontier model today that can do most developer jobs at less than $200 a month, nearly instantly, without being tired and without being on your payroll. Instead of hiring 10 developers, now you can probably get away with 3 devs and give them a liberal token budget.
I know this bit is true because of running Medblocks and the increase in efficiencies we've seen during the past couple years. Our hiring strategy has changed from hiring a mix of excellent engineers and mediocre engineers to only hiring excellent engineers. We've also had to go remote-first since the best engineers with the greatest potential often don't want to move. All of this means the heavily entrenched large IT companies, often called WITCH companies (Wipro, Infosys, TCS, Cognizant, HCL), have something to lose since their only competitive advantage was coders for hire at cheap rates. We've already seen drawdowns of 22% this year, and almost all gains in the past 5 years have been wiped out. This is all while their revenue has stayed stable and profits intact.
Today is the worst the coding models will ever be. That was true a year ago and seems like it will be true a year from now. All of this is characterized as the end times for the Indian IT growth story.
What can't AI take away?
We often wonder this at work: if AI can code at the speed of light and write better code that leads to more maintainable systems, what will humans still be required for? Over the past 3 years we've gone from building greenfield projects to maintaining old codebases that interact with legacy devices that probably shouldn't be kept running. System maintenance involves updating the underlying dependencies and adding new features to make sure the product moves in lockstep with the user requirements. This sounds easy to do using AI at the outset. However, every update and feature addition is a tradeoff between time taken to develop and utility of the feature itself. This is assuming the user knows what they want, which, if you work in any product development, you know isn't true.
All of this to say that there's an unseen consequence of a concentrated services industry, which is the ability to judge the user's need and understand it deeply. This ability might not exist with providers who work across sectors and fields, but is true of providers who are deeply into one sector only. This will create winners in the long term: global companies born in India that operate in a specialized niche.
There might not be globally competitive AI companies in India just yet (don't bet against that too hard either), but just the scale of agglomeration in understanding user needs and adding value means that there is a whole different game that's ready to be played. We might not have super large giants that span sectors anymore, but democratization of AI access will mean we will have winners across the country.