Artificial intelligence has become the storyline of the decade, but much of the public debate misses what is already happening on the ground.
The effects are not hidden in far-off labs or theoretical forecasts. You can find them in metrics such as equipment orders, corporate budgets, power grids, and even in stock-driven consumer spending.
AI shows up in small ways that add up to something larger. Some firms are working faster, some jobs are changing shape, and entire sectors are reorganising around software that did not exist three years ago.
The idea that AI has yet to make an impact is no longer accurate. But the key is to understand where the impact is visible and why the official productivity numbers have only begun to move.
How far has AI changed the numbers so far
The cleanest data comes from the Penn Wharton Budget Model, which breaks AI’s influence into tasks rather than job titles. Their findings are a useful foundation because they use detailed US occupational data.
They estimate that around 40% of today’s labour income is linked to work that could be reshaped by generative AI. Not every task is profitable to automate, but a sizeable share is.
Their model suggests that around 10% of today’s GDP is likely to be affected and that the share could climb to about 15% in the next two decades as sectors with high exposure grow faster than the rest of the economy.
Studies of real deployments paint a clear picture of what happens inside firms. Customer support teams that use AI assistants complete more cases. Professional writers complete drafts roughly 40% faster.
Software engineers finish tasks faster when using Copilot-style tools.
Penn Wharton uses these studies to assume around 25% labour cost savings when AI is used today, rising toward 40% as systems improve further.
Source: Penn Wharton Budget Model
When these cost savings are weighted by the share of tasks affected and adjusted for adoption rates, the macro effect becomes clearer.
The boost to US total factor productivity is small today, roughly 0.01 percentage points. But it is expected to rise through the decade and peak in the early 2030s near 0.2 percentage points before fading.
The long-term result is a larger economy, not a permanently faster growth rate. Their central estimate is that AI lifts GDP by about 3% by 2055.
The takeaway from such research is that AI’s early productivity effects are meaningful but uneven and concentrated in certain tasks and companies.
Aggregate numbers reflect an economy that has not yet reorganised itself around the technology.
Why business investment is doing the heavy lifting
The clearest signs of AI’s influence appear on the investment side of the US economy. Recent research by Bloomberg showed that spending on information-processing equipment and software has surged this year.
The contribution to GDP growth from these categories is the largest in decades.
Data center construction has reached an annual rate of around $41 billion and is one of the few growing segments in private construction.
Three companies alone, Meta, Microsoft, and Google, spent $78 billion on capital equipment in the third quarter of this year, almost double their spending a year earlier. This is the physical footprint of the AI boom.
Bloomberg estimates that AI-related capital spending contributed around 1% to US GDP growth in the first half of 2025. That means AI accounted for more than half of the 1.6% growth rate over that period.
It is rare for a single technological wave to play such an outsized role in national accounts.
Source: Bloomberg
Some analysts expect the investment impulse to strengthen next year. Others think the peak has already passed. Either way, the contribution is measurable today.
There is a complication that often goes unnoticed. Much of the hardware needed for data centres is imported. That widens the trade deficit and removes some of the gross growth contribution.
Yet the federal government has allowed tariff exemptions for servers and circuit boards despite trade tensions elsewhere.
Economists note that the boom would have struggled if the hardware faced the same duties paid by industries such as autos or construction.
There is also a strain on the power system. Data centers demand large amounts of electricity. US power demand could rise by about 16% by 2029 if current trends continue.
Grid upgrades take years and face higher costs because of tariffs on equipment such as transformers. Rising electricity prices could slow AI adoption or compress margins for firms trying to scale.
Inside the companies already feeling the change
Surveys offer a different perspective. They show what happens once AI tools enter daily workflows. IBM’s 2025 EMEA study is one of the largest of its kind, based on 3,500 executives across ten countries.
Two-thirds said that AI has already produced significant productivity gains.
One in five say they have already reached their return-on-investment targets. Another 42% expect returns within a year, often through faster execution, lower costs, and improved service quality.
The report highlights that large firms are ahead. 72% of enterprises with more than a thousand employees report notable gains. Only 55% of small and medium-sized firms say the same.
Public sector organisations show similar patterns to smaller firms. This mirrors earlier technology cycles where larger organisations had the capital and technical capacity to move first.
IBM’s data also shows how work is changing. Executives say employees spend more time on planning, creative work, and idea development when AI handles repetitive tasks.
The pattern aligns with academic studies that show AI helps less experienced workers close performance gaps and allows experienced workers to focus on higher-value output.
What stands out in both the IBM and Penn Wharton findings is how uneven the gains are across occupations. The most exposed tasks are in office support roles, business operations, IT, sales, and middle management.
Exposure peaks around the eightieth to ninetieth wage percentiles, then drops for the highest earners who tend to perform tasks that require judgment, negotiation, or rare expertise.
The least exposed groups include construction, transport, food services, and in-person care work. This means the near-term labour impact is concentrated in mid- to high-wage white collar roles, not in manual jobs.
What CEOs are actually saying inside boardrooms
Corporate leaders describe a change that is faster than previous cycles. Goldman Sachs chief executive David Solomon said he cannot find a CEO who is not trying to redesign processes around automation.
He said companies want to raise output without raising headcount and that AI is now central to those efforts.
Microsoft’s Satya Nadella made similar comments by calling this moment an “AI platform shift.” Nvidia’s Jensen Huang, who has likened it to a new industrial revolution.
Their views describe what the current AI leaders see inside their own operations and among their customers.
The data support this sentiment. Growing numbers of firms are not simply adding AI tools to existing workflows. They are rebuilding workflows around the tools.
Some design value chains from scratch with AI in mind. Others switch from periodic planning cycles to continuous decision-making guided by AI systems.
These changes take time to filter through to official productivity numbers, which helps explain the gap between what firms report internally and what shows up in national statistics.
Another important point shows up in the surveys. Companies want open and interoperable AI systems. Around 85% of IBM’s respondents said transparency, interoperability, and provider flexibility are essential.
What does all this tell us about the real AI productivity effect
AI today appears in three places in the economy. It appears that rising capital spending by firms is racing to build computing capacity. It appears in the daily work of early adopters who report faster execution and higher output.
And it also appears in asset markets, where AI-led companies have generated trillions of dollars in new equity wealth that has fuelled higher consumption among affluent households, sparking concerns of an “AI bubble.”
What does not appear yet is a broad, economy-wide rise in productivity. This is not unusual.
Previous general-purpose technologies, such as electrification and the internet, showed up in the data only after firms reorganised production.
The same pattern is playing out again. AI is still in the investment and experimentation phase for most companies. Once the reorganisation is complete, the gains are likely to show up more clearly.
The most careful estimates suggest AI lifts productivity growth by a few tenths of a percentage point at its peak and leaves the economy permanently larger by a few percent.
The more ambitious forecasts suggest higher gains if AI accelerates innovation itself.
The gap between these views depends on how quickly firms restructure work, how widely AI tools spread to smaller companies and the public sector, and whether infrastructure such as the power grid can scale to match demand.
The post What is the real effect of AI on productivity? appeared first on Invezz
