Our Take
Two years into the AI boom, there is no measurable employment or wage lift in US data yet—which means either adoption is still too shallow, productivity gains are being captured upstream, or both.
Why it matters
If AI is truly the productivity inflection point vendors claim, labour markets should show it first. The ECB's finding suggests either the deployment phase is much earlier than headline adoption rates imply, or the economic model of AI-driven efficiency differs sharply from prior technology waves.
Do this week
Finance teams: audit actual headcount and compensation changes in your AI-adjacent roles month-over-month for the next quarter—the macro data lag means internal signals are your earliest warning of whether AI is replacing or augmenting your workforce.
ECB analysis finds employment and wage effects from AI still absent
The European Central Bank conducted a study of US labor market data through the lens of AI adoption and found no statistically significant gains in either employment levels or wage growth so far. The analysis examined the period during which AI tools have achieved rapid commercial deployment and user growth—particularly large language models and enterprise AI applications—yet detected no measurable downstream effect on job creation or worker compensation (per ECB research).
This finding contradicts the narrative arc that typically accompanies major technology transitions. Past infrastructure waves (electrification, computing, broadband) generated measurable employment spikes within two to three years of mainstream adoption. The AI cycle, by contrast, shows adoption climbing sharply while labour market signals remain flat.
Adoption speed is outpacing deployment depth
Three explanations fit the data. First, adoption is still concentrated in knowledge work roles at large tech and financial firms, where efficiency gains are being captured as margin rather than headcount. Second, the productivity multiplier from AI is real but still too small relative to the total US labour base to register. Third, deployment at operational scale (the kind that actually replaces or creates jobs) lags the 12–18 month hype cycle by years.
The ECB's timing is material. We are now at the end of Year Two of public LLM deployment. If the employment thesis is correct, Year Three should show visible movement in Bureau of Labor Statistics data. If it doesn't, the productivity gains story becomes a margin-capture story, which changes how vendors should position ROI to enterprise buyers and how workers should think about reskilling.
Watch the Q1 2025 jobs reports for the first real signal
The US labor market releases monthly employment data and wage data quarterly. By March 2025, if AI adoption has genuinely created net new roles or lifted wages in affected sectors, those numbers should begin to separate from baseline trends. They haven't yet. Watch specifically for wage growth in AI-exposed job categories (data engineering, software engineering, data science roles) relative to the broader market. If those buckets are not growing faster in pay, the labour capture is upstream only.