A Message from Mike
Is the AI Productivity Boom Beginning?
For the past three years, almost every conversation about artificial intelligence has eventually come back to the same extraordinary numbers. Hundreds of billions of dollars being spent on chips, data centres, power, networking and software. Nvidia selling more processors than it can make. Microsoft, Alphabet, Amazon and Meta committing sums to AI infrastructure that would have seemed almost unimaginable only a few years ago.
Naturally, investors have started asking the obvious question.
Where is the return?
I think we may finally be starting to see the beginnings of an answer.
Not in another chatbot, a new data centre or the latest generation of semiconductor. The number I have become increasingly interested in is far less exciting: output per hour worked.
That is productivity, and over long periods of time it is one of the most powerful forces in an economy.
The latest numbers from the US Bureau of Labor Statistics are worth paying attention to. In the second quarter of 2026, US nonfarm business productivity was 2.2% higher than a year earlier. Output increased 2.5%, while the number of hours worked increased by just 0.2%. In other words, American businesses produced substantially more without requiring substantially more labour. (Bureau of Labor Statistics)
One quarter doesn’t make a productivity boom. But this isn’t simply a one-quarter phenomenon.
Since the end of 2019, US labour productivity has grown at an annualised rate of 2.1%. During the previous economic cycle, from 2007 to 2019, it grew at just 1.5% a year. (Bureau of Labor Statistics)
Six-tenths of one percentage point doesn’t sound particularly exciting.
Compounded over a decade, it is.
Imagine an economy producing $100 of output per worker today. At 1.5% annual productivity growth, that worker would produce about $116 a decade from now. At 2.5%, they would produce roughly $128.
That is around 10% more economic output from essentially the same worker, simply because productivity grows one percentage point faster each year.
Multiply that across an economy the size of the United States and the numbers become enormous.
What if all this spending is actually working?
This is where the AI story becomes particularly interesting.
We have spent years discussing the inputs. How many GPUs are being sold? How much are the hyperscalers spending? How many gigawatts of electricity will data centres require?
But ultimately none of those things determines whether AI transforms the economy.
Productivity does.
If an accountant can complete in three hours what previously took four, a software developer can write and test code faster, a lawyer can review documents in a fraction of the time, or a financial adviser can spend less time preparing paperwork and more time speaking with clients, something important has happened.
The same number of people can produce more.
And we are beginning to get evidence of exactly that.
Earlier this year, researchers surveyed nearly 750 corporate executives about how AI was being deployed inside their businesses. More than half of their companies had already invested in AI, and the executives reported positive productivity improvements, with the gains expected to strengthen through 2026.
Interestingly, some of the largest effects were appearing in high-skill services and finance, precisely the sort of knowledge-intensive industries where generative AI has the potential to automate parts of existing workflows rather than simply replace physical labour. (National Bureau of Economic Research)
Even more importantly, the researchers found that the gains were not primarily the result of companies simply throwing more capital at the problem. They found evidence of improvements in what economists call total factor productivity: broadly speaking, businesses becoming better at turning their existing labour and capital into output. (National Bureau of Economic Research)
That is exactly what we should be looking for.
But let’s not declare victory yet
There is an important qualification.
We cannot look at America’s recent productivity improvement and simply declare that AI caused it.
Productivity moves for many reasons. Businesses invest in new equipment. Supply chains improve. Workers move between industries. Companies become more efficient after periods of disruption. Economic cycles matter.
Even the research into AI itself identifies what economists are calling a productivity paradox. Executives frequently report that AI is making their businesses more productive, but those improvements are not always showing up immediately in measured revenue and economic output. (National Bureau of Economic Research)
That doesn’t necessarily surprise me.
History suggests that transformative technologies often take time to appear in economy-wide productivity statistics. Buying a new technology is the easy part. Redesigning businesses around it is much harder.
Installing Microsoft Copilot does not suddenly make an organisation 20% more productive. Companies need to change workflows, retrain employees, integrate systems and, eventually, work out what to do with the hours that technology saves.
That last part may be the most important.
Saving an employee two hours a week doesn’t increase GDP if those two hours simply disappear. It becomes economically significant when those hours are redirected into producing more, serving more customers, developing new products or doing higher-value work.
That transition is still happening.
The corporate numbers are becoming interesting
There is another set of figures that caught my attention.
Within America’s nonfinancial corporate sector, productivity was 3.1% higher over the year to June. In the second quarter alone, output increased at a 3.9% annualised rate while hours worked increased 1.7%.
At the same time, unit labour costs actually fell at a 0.3% annualised rate during the quarter. (Bureau of Labor Statistics)
Think about what that combination means.
Companies are producing more. Productivity is increasing. And the labour cost associated with producing each unit of output is not rising with it.
For shareholders, that is a very attractive equation.
It is also one of the reasons productivity matters enormously beyond the technology sector.
The economic prize is much bigger than AI
Suppose, purely for illustration, that AI and other technological improvements eventually lift sustainable US productivity growth from around 1.5% to 2.5%.
That extra one percentage point doesn’t simply mean employees answer emails faster.
It potentially changes the speed at which the entire economy can grow.
If the labour force grows by roughly 0.5% a year and productivity grows by 1.5%, an economy can broadly produce around 2% more each year before considering other factors.
Increase productivity to 2.5%, and suddenly something closer to 3% becomes possible.
And importantly, that additional growth comes from producing more with the resources we already have.
That is fundamentally different from stimulating an economy by simply adding more money or creating more demand.
Greater productivity can allow companies to pay higher wages, earn higher profits and produce more goods and services without requiring an equivalent increase in costs.
That matters for inflation.
It matters for interest rates.
And it matters enormously for investors.
Follow the earnings
For shareholders, productivity eventually finds its way into one place: earnings.
A company that can grow revenue without increasing its workforce at the same rate becomes more profitable. A company that can automate routine work can redirect people towards higher-value activities. A business that can serve twice as many customers using roughly the same infrastructure becomes more valuable.
This is why I think the next stage of the AI story could ultimately be more important than the first.
The first stage was about building AI.
Semiconductors. Data centres. Networking equipment. Power generation. Cloud infrastructure.
The next stage is about using it.
If AI begins lifting productivity across banking, healthcare, accounting, insurance, engineering, manufacturing, logistics, law, software and thousands of other industries, the economic beneficiaries become far broader than the companies selling the infrastructure.
That is the part of this story I find most exciting.
We do not yet have enough evidence to say that an AI-driven productivity boom has arrived. Anyone making that claim today is getting ahead of the data.
But we do have enough evidence to start watching very closely.
US productivity growth since 2019 is running ahead of the previous economic cycle. Businesses are producing more while barely increasing aggregate hours worked. Corporate productivity has strengthened. Executives deploying AI are reporting measurable gains. And researchers are beginning to see improvements in the efficiency with which businesses combine labour and capital. (Bureau of Labor Statistics)
Perhaps all of that turns out to be coincidence.
But perhaps we are witnessing the very early stages of something considerably more important.
For the last few years, investors have understandably watched AI spending measured in billions of dollars, GPU shipments and data-centre capacity.
Over the next few years, I think there is another number worth watching.
Output per hour.
Because if that number keeps moving higher, the most important return on the enormous investment in artificial intelligence may not belong to any single technology company.
It may belong to the economy as a whole.
.png)
