The market separated
The defining change is that a single figure for technology pay stopped being meaningful.
Specialisations with genuine scarcity — machine learning, security, distributed systems, data infrastructure — have continued to see strong upward pressure. General application development, particularly at entry level, became considerably more competitive, with more applicants per role and longer hiring processes.
The useful question is therefore not whether technology pays well but whether your particular specialisation is one employers are struggling to fill. Those are now very different situations.
AI and machine learning
The clearest area of sustained demand, and the split within it is worth understanding.
Research roles remain small in number and require doctoral-level preparation. Machine learning engineering — building pipelines, deploying models, operating inference at scale — is much larger and reachable from a strong software engineering background.
The fastest-growing category is applied work: engineers building products on top of existing models, dealing with retrieval, evaluation and the substantial engineering needed to make probabilistic systems behave acceptably. Evaluation skill in particular is scarce and valued, because knowing whether a system actually works is harder than getting it to produce output.
Security
Security demand has been elevated for years and has not eased, driven by both incident frequency and regulatory pressure.
Application security sits at a particular premium because it requires genuine engineering ability alongside security knowledge, and few people have both. Cloud security, detection engineering and incident response are similarly short-staffed.
Certifications carry more weight here than elsewhere in technology, and demonstrable practical work — published research, bug bounty findings, competition results — is taken seriously in hiring.
Data and infrastructure
Data engineering has become one of the more reliably well-paid specialisations, partly because it is less fashionable than model work and therefore less crowded.
Every AI and analytics initiative depends on data infrastructure that mostly does not exist yet, which sustains demand independently of whatever is happening in the model layer. The skills are concrete and learnable.
Platform and site reliability engineering follow a similar pattern. As systems grow more complex and cost pressure increases, people who can make infrastructure both reliable and efficient are valued — cost optimisation has become a genuinely marketable specialisation in a way it was not a few years ago.
How compensation is structured now
Equity has become a larger share of packages again as valuations recovered from their trough, which changes how offers should be compared.
At public companies, restricted stock is close to cash with a vesting delay, though the value moves with the share price. At private companies, options require you to understand the strike price, the current valuation, the vesting schedule and the likely dilution before treating them as compensation at all.
The practical rule is to compare total packages rather than base salaries, and to satisfy yourself that the cash component alone is acceptable. Equity that may be worth a great deal is not the same as equity that is.
Where hiring is hardest
Entry-level is the most difficult part of the market at present, which is a real change from a few years ago.
More applicants, fewer junior openings, and some routine junior tasks now automated have all combined to raise the bar. Graduates report substantially longer searches than the previous cohort, and demonstrable work matters more than it used to.
Mid and senior hiring in scarce specialisations remains competitive in candidates' favour. The gap between those two experiences is wide enough that general advice about "the tech job market" is close to useless without specifying level.
What to do about it
A few practical conclusions.
Specialise deliberately rather than accumulating general experience. Depth in a scarce area is what commands a premium, and breadth is increasingly commoditised.
Build and show real work, particularly if you are early in your career. Employers can verify a deployed project; they cannot verify a claim on a resume.
And check your position against the external market annually. Internal raises are bounded by bands, external offers are priced at current market, and in a period where specialisations are diverging, that gap can grow quickly without you noticing.