Tech Salary Convergence: Why Only Half the Market Moved

Everyone's having the wrong argument about tech salary convergence. They're arguing about whether to chase cheap geography at all, as though the market's one big block moving together. It isn't. There are four different markets stacked on top of each other, they behave completely differently, and most leaders are pointing their strategy at the wrong one.

Easy bit first. If you're trying to compete for the absolute top of the market, the handful of engineers who could walk into Google DeepMind or OpenAI tomorrow, don't bother. That talent has been priced globally for years. London packages at US-headquartered frontier labs regularly approach 70 to 85% of US bands, and inside FAANG-owned operations in India, senior and staff engineers are pulling total compensation that genuinely overlaps UK senior bands, not close to them, overlapping them. You are not winning a geography argument against that.

To where I think most hiring strategy is actually pointed at the wrong target. Below the absolute superstar tier, there's a slice of talent that's very good, genuinely scarce, but not world-leading. AI/ML engineers, cybersecurity specialists, cloud and niche infrastructure people. Not the researcher who trained the model, but the engineer who can actually ship it reliably. That slice is where the real convergence is happening right now, and it's happening because of a specific, deliberate choice employers are making, not because of some abstract market force trickling downward.

The mechanism is worth understanding properly, because it changes what you should do about it. Remote-hiring companies broadly split into two camps: global-rate employers who pay the same wherever the person sits, and location-adjusted employers who pay 40 to 70% less for identical remote work in a cheaper country. Most companies still run location-adjusted. But for this specific tier, more and more of them are switching to global-rate, because the roles are scarce enough to be worth fighting for and the cost of getting it wrong, a bad hire or a six-month vacancy, dwarfs the geographic discount they'd otherwise pocket. The numbers back it up. A senior developer working locally for a Romanian or Hungarian employer still earns roughly €2,800 to €3,500 a month. The same person, same skills, hired remotely into that specialist tier by a UK or US company, earns €5,000 to €10,000. Inside India's GCCs, the average salary increase this year sits around 9.8%, but AI/ML engineers are getting roughly 21%, cybersecurity specialists 20%, and top performers generally 1.8 times the average uplift. That's not the whole market moving. That's a specific tier getting bid up hard by employers who've decided it's worth paying global rate for, and it's the clearest tech salary convergence happening anywhere in the market right now.

The ordinary mid-market, solid generalist engineers doing solid generalist work, hasn't followed, much. It's still mostly geography-priced, and importantly, that's a choice rather than a constraint. Remote software engineers in India earn roughly $49,000 on average against $97,000 in the US for comparable remote-hireable work, and that gap isn't closing because most employers haven't extended the global-rate policy down to that level. There's no commercial pressure forcing them to. The scarcity that justifies paying global rate for a top engineer simply doesn't exist for a competent generalist, so the discount survives.

Will that change? Possibly, but slowly, and this bit is forecast rather than fact. There's a real structural mechanism that could pull the median up over time: demographic decline and emigration are already tightening labour supply across Central and Eastern Europe, and nominal median wages there are climbing 6 to 8% a year against 3 to 5% in Western Europe. That's already visible in the data, it's just a fifteen-year story, not a five-year one, and the region still hasn't reached parity despite roughly a decade and a half of this pressure. There's also a genuine counter-argument worth taking seriously. If AI tooling narrows the gap between what a good generalist and a specialist can each deliver, employers may have even less reason to pay up for the median tier, which would keep this exact divide open rather than close it. I don't think anyone has a confident answer to that yet, and I'd be suspicious of anyone who claims they do.

And the bottom of the market, graduate and junior hiring, hasn't moved at all. It's still almost entirely local, because that pool is large, replaceable, and never worth the overhead of hiring remotely for in the first place.

Don't compete for the absolute top, you weren't invited to that auction and you'd lose it anyway. Don't waste effort building a geography-based cost case for the strong-but-not-superstar specialist tier, that arbitrage is already gone and shrinking further, so go wherever the capability is and compete on the work rather than the discount. For ordinary mid-market generalist hiring and for junior pipelines, a co-located hub still earns its keep, but the reason to run one has quietly changed underneath you. It used to be defensible purely as a cost play. Pick one off a "cheapest emerging hub" ranking and you'll get burned the way one US firm did chasing a rising-hub list for a Kubernetes platform search, six weeks with a thin pipeline before they moved the identical search to a city with the actual infrastructure depth and filled it in eleven days, same budget.

Closer to home, Bristol's persistent premium over Cardiff at the senior end isn't really about Bristol being pricier for its own sake, it's that Bristol carries a genuine aerospace and fintech cluster Cardiff doesn't, and that's worth paying for if that expertise is what you actually need. The mistake isn't picking hybrid over hubs, or the other way round. It's choosing a hub for the discount instead of the fit, using a cost logic that's already stopped applying to the talent you're most likely trying to hire.

Does the median catch up over the next decade the way the specialist tier already has, or does AI adoption keep that gap propped open indefinitely? I don't think it's settled, and I'd rather argue it out properly than pretend I've already called it.