Two engineers joined the same Bangalore company in 2022. Same campus. Same batch. Same salary, eighteen lakhs.
One spent the next two years learning how to fine-tune LLMs. The other stayed in their existing stack, picking up some API integration work with the new models coming out.
In 2024, the first engineer switched jobs for thirty-two lakhs. The second switched for twenty-two.
That gap, fourteen lakhs for roughly the same total years of experience, is the story the market is telling right now. And it's not an anomaly. It's a structural divide.
I've been watching this pattern emerge through conversations this year. On one side, what I'll call AI-core roles: ML engineers, applied scientists, NLP specialists, anyone who builds, trains, or optimises models. On the other side, AI-support roles: backend engineers who integrate API calls, frontend engineers who build chat UIs, platform engineers who deploy inference endpoints. Everyone is working with AI. But the market treats them very differently.
The AI-core people I'm seeing are getting annual hikes in the 16 to 21 percent range. Offers are competitive. Counter-offers are aggressive. One candidate told me his current company matched an outside offer in forty-eight hours, and then bumped it again a week later without his asking. That kind of thing happens when the supply of people who can actually tune a loss function is thin.
The AI-support people are getting 7 to 8 percent. Sometimes less. I spoke to a backend engineer whose team just shipped a major GenAI feature. He had built the entire serving layer. His appraisal rating was "exceeds expectations." His hike was eight percent. The company posted a record quarter that same week.
I'm not making a moral point here. The market is not a fairness machine. What worries me is how many people I talk to don't see the divergence yet, or see it and assume it will correct itself, that the AI-support roles will catch up as the hype normalises.
That assumption has two problems. One, the hype is not normalising. Companies are still scrambling to build AI-native products, and the premium for people who can do the core work is still rising. Two, the gap is not a temporary imbalance. It's a separation of two career tracks that used to be one.
Three years ago, a "machine learning engineer" title covered everything from data cleaning to model deployment to monitoring. Now those functions are splitting. The building side and the integrating side require different skills, attract different compensation, and follow different career arcs. You can be very good at the integration side and still watch your peer from the same batch pull ahead by twelve to fifteen lakhs a year, simply because they trained a different muscle.
This is not advice to switch tracks. Not everyone wants to work on model internals, and not everyone should. The AI-support roles are not disappearing, they are growing. But they are growing at a different velocity, and pretending otherwise is expensive.
I told the engineer who got twenty-two lakhs to ask himself one question: in two years, will the thing you're good at be harder to replace, or easier? He changed his answer three times before he hung up.

