The instinctive story about artificial intelligence and employment runs in one direction: automation strips out roles and drags wages down with them. A new PwC analysis of more than one billion job postings across 27 countries says the opposite is unfolding in exactly the sectors where the technology has embedded itself most deeply.
The PwC 2026 Global AI Jobs Barometer, drawing on data through 2025, found that wages in high-AI-exposure industries are climbing at roughly twice the rate of low-exposure sectors. Roles requiring AI skills now carry a 62% wage premium over comparable positions without them. Two years earlier, that premium sat at 25%.
The speed of that shift is the signal worth examining, and it is where most of the popular narrative gets the direction of travel wrong.
This piece separates what the data actually shows from what most people assume, identifies which sectors and role types are capturing the gains, and gives you a framework for reading AI adoption signals in company disclosures and hiring patterns.
The data that upends the automation-kills-jobs assumption
Scale is what makes this dataset worth taking seriously. The PwC barometer analysed over one billion job advertisements spanning 27 countries and six continents, a reach that sets it apart from the narrower firm-level or single-market studies that dominate the debate.
Start with the wage premium, and read it as a trajectory rather than a single figure. Two years ago, AI-skilled roles paid 25% more than equivalent positions without those skills. The following year, that gap widened to 57%. It now stands at 62%.
That escalation from 25% to 62% in two years tells you something the headline number alone does not: the value of AI skills is compounding, not settling. The cost of waiting to build those capabilities rises with each year the premium climbs.
The wage story sits alongside a productivity story that is equally striking.
The top fifth of the most AI-exposed companies achieved 163% productivity growth since the 2018 benchmark, according to PwC’s analysis.
Then there is the finding that most directly contradicts the displacement thesis. Companies using AI to amplify what their workers can do grew headcount by 52%, against 36% for their lower-exposure peers. More AI exposure, in this data, coincided with more hiring, not less.
The premium is accelerating, not stabilising
Put the three data streams together and a coherent picture emerges. Wages, productivity, and headcount are all rising fastest at the firms leaning hardest into AI.
For investors, that combination matters. It means AI adoption and workforce growth are not opposing forces in the current evidence; the firms extracting the most productivity are also the ones expanding their teams. A survey of opinion would not carry this weight. A billion job postings across six continents is a credible baseline for strategic thinking, and it points away from the substitution story that dominates most commentary.
When big ASX news breaks, our subscribers know first
Why the same technology produces winners and laggards
Two companies can license the identical AI tools and end up in completely different places. Understanding why is the analytical core of this whole debate.
The fork comes down to strategy. One path uses AI to amplify what people produce. The other uses it primarily to remove people from the payroll.
- Amplification: AI absorbs routine work, human roles rise in complexity, output per worker climbs, and headcount grows alongside revenue.
- Substitution: AI is deployed as a cost lever, headcount is cut, and efficiency gains stall because the underlying workflow was never redesigned.
The productivity gap between these paths is not subtle. Top-quintile amplification firms reached 163% productivity growth since 2018, while low-exposure firms saw growth stay flat or marginal over the same window.
Why does substitution so often fail to deliver? Because bolting an AI tool onto disorganised legacy systems produces very little. As Nils Henning, Senior Solutions Engineer at Ninox, has framed it, the technology works as an output multiplier rather than a headcount minimiser. The gains come from redesigning how work gets done, not from swapping a licence for a salary.
AI capital flows into physical infrastructure tell part of the story that the wage data alone does not: hyperscalers committed over $650 billion to data-centre build-out in 2026, yet economy-wide productivity gains remain near zero, confirming that the transition friction between investment and measurable output is real and unevenly distributed across sectors.
There is a second dynamic that laggards miss. When AI takes over the predictable, process-level tasks, the human work that remains becomes harder and higher-stakes.
When AI handles the routine execution, the quality of everything it produces depends on the human judgment directing it. That raises the stakes of human performance rather than removing it.
You can see this playing out at the role level. In customer service, AI manages standard, templated inquiries while human agents are reallocated to complex, escalated cases that need empathy and judgment. In software engineering, tools augment rather than replace: the 2025 Stanford HAI AI Index documented that AI coding tools increased developers’ exploration of new technologies by 21.8%, with an average potential salary uplift of roughly $1,683 per developer per year.
Here is the diagnostic that matters for reading a company. The question is not whether a firm has deployed AI. It is whether the jobs and the wage bill are growing alongside it, because that pattern is what separates durable productivity from short-term cost arbitrage.
| Adoption strategy | Headcount trend | Productivity outcome | Investor signal |
|---|---|---|---|
| Amplification | Growing (52% at top firms) | Compounding (up to 163%) | Durable value creation |
| Substitution | Falling or flat | Flat or marginal | Short-term margin only |
Where the wage premium is landing, and why it is not evenly distributed
The premium is real, but it is not spread evenly. In consumer markets it has reached as high as 118%. In the public sector it sits at 16%. That is not statistical noise.
Manufacturing lands in between at a 73% premium, and the split within that number is revealing: AI-user roles command 69% while AI-developer roles command 102%. The closer a role sits to building the systems rather than operating them, the steeper the pay curve.
The demand behind these numbers is intense. Specialised AI roles grew 69% in 2025, seven times faster than the 9% overall job market growth. Sector job growth followed a clear hierarchy: technology, media and telecommunications led at 11%, professional services at 6%, and healthcare under 1%.
| Sector | AI wage premium | Job growth (2025) | Key dynamic |
|---|---|---|---|
| Consumer markets | Up to 118% | Strong | High-complexity human roles elevated |
| Manufacturing | 73% | Moderate | Developer roles (102%) outpace user roles (69%) |
| TMT | High | 11% | Fastest overall hiring growth |
| Professional services | Elevated | 6% | Judgment-based roles complemented |
| Public sector | 16% | Low | Automation of existing baseline |
Why does the same technology lift wages in one sector and barely move them in another? The answer sits in two competing dynamics. Professionalisation happens where AI automates routine tasks and raises the value of human judgment, pushing wages up. Democratisation happens where AI makes complex tasks easy for non-experts, commoditising the role and pushing wages down.
The spread from 118% in consumer markets to 16% in the public sector is a structural signal. It tells you where AI adoption is translating into differentiated human capital value, and where it is simply automating the baseline that already existed.
HSBC’s mapping of nearly 2,000 business categories into sector-level disruption scores gives the geographic dimension of the same story: Taiwan and South Korea score net positive at 53% and 33% respectively under the moderate scenario, while Austria faces a 30% projected revenue loss, illustrating how the same technology creates mirror-image outcomes across index compositions.
The distributional risk the headline figures obscure
The aggregate optimism hides a harder reality. The IMF and OECD both warn that positive wage effects are concentrated among high-wage, cognitively intensive workers, not distributed broadly across the workforce.
ECB modelling suggests roughly 25% of skills in the lowest wage quartile face high AI-substitutability exposure, implying acute displacement risk for lower-paid workers.
The Dallas Fed adds a generational wrinkle. AI substitutes well for entry-level, book-learned skills that lack tacit knowledge, which compresses junior wages while complementing experienced workers who bring judgment the tools cannot replicate. The result is an experience premium that favours seniority.
There is also a timing caveat. A 2025 meta-analysis by Carbonara suggests some of the current premiums may reflect temporary first-mover advantages rather than permanent structural shifts, particularly in digital freelance markets where AI is already pushing piece-rates down.
What this shift means for how investors should read AI adoption
The evidence converges on a practical lens. Rather than favouring any company that mentions AI, you can track three observable signals in disclosures and hiring data.
- Wage bill and headcount growth in AI-complementary roles alongside revenue. Rising employment and pay in these roles, moving with top-line growth, validates durable productivity over cost-cutting.
- Job posting language shifting toward judgment. When postings emphasise decision-making, stakeholder management and problem-solving, it signals roles are being upgraded rather than deskilled.
- Capital allocation into training and workflow redesign, not just licensing. Durable gains require investment in organisational capital, not a software subscription alone.
Cost-saving language in an earnings call, absent any correlation with top-line growth, is a weak signal. It points to temporary margin improvement, not the compounding productivity that the data associates with genuine amplification.
Separating credible automation risk signals from susceptibility rankings is the practical challenge the evidence creates; corporate headcount disclosures, such as GEICO’s disclosed 25% reduction in entry-level claims adjuster roles, carry far more actionable weight than occupation-level computerisation scores that have historically predicted growth as often as decline.
The quantified case is compelling.
Makridis and Johnston (May 2026) found that a one standard-deviation increase in generative AI exposure was associated with 10% higher sector output, 3.9% higher employment, and a 4.8% rise in the aggregate wage bill, across data covering over 95% of US employers from 2017 to 2024.
Output, employment and wages rising together is the signature of amplification. Supporting figures reinforce it: hourly wages rose 1.0-1.1% on average from AI exposure, with 1.4% for college graduates. Lightcast found in July 2025 that AI-skilled postings pay around 28% more, rising to 43% for two or more AI skills, and an OECD Canada study put the within-firm premium at roughly 11%. PwC’s own data shows wages at the most AI-exposed firms rose 24% against 17% at least-exposed peers since 2018.
So the read for you is direct. A company reporting AI-driven efficiency while its wage bill and AI-adjacent headcount stay flat or fall is telling a cost arbitrage story, not a productivity one, and it should be priced accordingly.
The limits of the current evidence
A word of caution on the numbers. Most of the premium data comes from 2022 to 2025, a period of acute AI skill scarcity that may normalise as training supply expands.
The Carbonara caveat on temporary premia in freelance and gig markets applies here too. The responsible position is to weight the structural signals, role elevation and headcount growth alongside revenue, over the premium magnitude, which is the more volatile variable.
This article is for informational purposes only and should not be considered financial advice. Investors should conduct their own research and consult with financial professionals before making investment decisions. Past performance does not guarantee future results. Financial projections are subject to market conditions and various risk factors.
Assessing AI adoption in a world where the numbers are still moving
The PwC data and the corroborating institutional research point in the same direction. AI amplification, not substitution, is the strategy producing compounding productivity and wage growth, and the clearest summary of that gap is the headcount differential: 52% growth at high-amplification firms against 36% at lower-exposure peers. The wage signal tracks it closely, at 24% versus 17% since 2018.
The picture is not settled, and several variables could reshape it over the next 24 to 36 months:
- How quickly rising AI training supply normalises the wage premium, which the jump from 25% to 62% in two years suggests may already be narrowing as the skill base broadens.
- Whether distributional policy responses alter the inequality trajectory the IMF, OECD and ECB have flagged.
- How fast laggard firms close the adoption gap with the amplifiers.
The AI adoption gap between frontier and laggard economies carries implications that extend beyond wages and headcount: Bank of America, Goldman Sachs, and JP Morgan each project the divergence will push equilibrium interest rates 25-60 basis points higher in leading markets over the next decade, an effect that compounds alongside the productivity differential the PwC data captures.
The actionable frame is straightforward. The companies worth tracking are those where AI investment shows up in the jobs data, growing headcount and rising wage bills in complementary roles, not only in the cost line. The amplification-versus-substitution distinction is not a settled verdict, but it is a practical analytical variable you can apply today.
