Every transformative technology shows up twice.
First as a promise, loud and unavoidable, the kind that floods your phone, your group chats, and your retirement account. Then as a line item, quiet and underwhelming, buried in productivity data almost nobody reads.
Right now, artificial intelligence (AI) is living loudly in the first version and almost silently in the second.
We have all seen the demos. We have all been told that AI will rewrite work, medicine, and the markets our savings ride on. The promise is everywhere.
Yet when economists go hunting for that revolution in the hard numbers, the figures that show up in gross domestic product (GDP) and paychecks, they keep finding something close to a rounding error.
The official tally for how much AI is adding to growth right now rounds to almost nothing. Not zero, but close enough that most people would never feel it in a paycheck or a portfolio statement.
That gap is where Bank of America (BAC) decided to plant its flag. In a late-May research note, the bank had a blunt message for anyone who has soured on the AI boom. You are thinking too small.
Why Bank of America says AI skeptics think too small
The bank’s case starts with a reframing. AI is not like electricity or even the internet, its economists wrote. It is more powerful than both, according to Fortune, and the productivity boom it eventually delivers could run ten times larger than anything the economy is showing today.
Then comes the awkward part. That payoff is currently running near 0.1%, which the bank itself called “a small aggregate effect relative to all the excitement around AI,” per Fortune. Set against global growth around 3.5%, a tenth of a percentage point barely registers.
More AI:
- Micron sits at the center of a red-hot chip rally
- IBM CEO sends blunt message on AI and quantum computing
- Anthropic CEO makes shocking admission about AI
So why so small? AI can already transform roughly 20% of workplace tasks, but only 23% of those are cheap enough to automate at today’s prices. The savings come to about 27% of labor costs, and labor is roughly half of all business costs.
Run that math and the theoretical ceiling is a 0.66% productivity gain, before friction, skills gaps, and slow adoption grind it toward 0.1%.
When I traced that tenfold claim back to its origin, it led to a 2024 paper from Nobel laureate Philippe Aghion and a co-author. Their model plugs in faster, cheaper AI and finds gains over the next decade that dwarf today’s numbers.
As inference costs fall, currently halving about every three months, the slice of tasks worth automating widens, and each expansion stacks on the last.
The task-level wins are not the part in dispute:
- Software developers ship 55% more work using AI coding tools, Bank of America found.
- Customer support agents resolve 14% more tickets.
- Professional writers finish projects 37% to 40% faster, the bank reported.
Source: Fortune
Bank of America’s sharpest point is about invention. Electricity automated physical labor and the internet moved information faster, but neither made inventing new things quicker. AI can, the bank argues, by speeding up the research that produces breakthroughs.
Related: BofA raises red flag on SpaceX, OpenAI IPOs
The AI bubble warning that lands first
The problem with a payoff a decade out is that a lot has to survive the wait.
Eight days before Bank of America’s note, Panmure Liberum strategist Joachim Klement made the opposite case, according to a report from Proactive Investors. The AI cycle, he argued, is not a productivity story waiting to unfold. It is a bubble waiting to pop.
His numbers are hard to wave away. The boom is already 60% bigger than the dot-com peak, with tech investment driving about 93% of all US GDP growth, far past the 56% high of the dot-com era.
Hyperscalers like Amazon (AMZN), Microsoft (MSFT), Alphabet (GOOGL), Meta (META), and Oracle (ORCL) are on track to spend $658 billion on capital projects in 2026 alone, growing about 20% a year through 2030.
To earn even a 10% return on that, Klement calculated, those companies would need to find $2 trillion to $5 trillion in new annual revenue, roughly quadrupling what they bring in now. He pegs Meta’s implied return on its planned spending at negative 28.8%, and Oracle’s at negative 35.6%.
The AI trade, he wrote, shows clear signs of “irrational exuberance.”
There is a structural worry under the valuations. Research from Tsinghua University suggests AI hallucinations are built into these models, not a bug to be patched out, which would keep them out of high-stakes work like accounting, legal filings, and compliance.
Smaller models running locally can already handle routine tasks at up to 1,000 times less cost than the cloud systems the boom is built to feed.
Klement is not calling for an imminent crash. Rate cuts, he figures, could stretch the bubble another one to two years. But in a true repeat of the dot-com unwind, he warned, technology stocks would fall more than 70%.
What the AI growth gap means for your money
Between the bull and the bear sits a number that should matter more to your household than either extreme.
Tyler Cowen, one of the most widely read economists in the country, put AI’s likely contribution to US growth at 2% to 2.5%, according to Fortune.
“Real,” he said, but short of what Silicon Valley keeps promising.
The reason is plumbing, not technology. Close to half of US GDP sits in slow-moving sectors like government, healthcare, higher education, and nonprofits, so the payoff arrives later and more unevenly than the boosters claim.
What stopped me, reading Cowen’s framing, was how personal that half-point becomes. The country is carrying about $39 trillion in debt. The difference between growing at 2% and 2.5% is the difference between that debt spiraling and that debt staying manageable.
Grow too slowly, in his telling, and the country drifts toward becoming the next Greece, with the bill arriving as higher taxes and cuts to Medicare, Medicaid, and Social Security. Grow a half-point faster and much of that pressure eases. “There is no plan B,” Cowen told an audience in New York, per Fortune.
That is why 0.1% is not a trivia question. It marks whether the benefits you are counting on later actually get paid for. The retirement and the safety net your kids will lean on ride, in part, on whether that number climbs.
Strip away the fight over valuations and the two camps agree on more than they let on. AI works. The task-level gains are real. The thing slowing the macro payoff is organizational friction, not the technology, according to Fortune.
I read both notes back to back, and the unsettling part was not the disagreement. It was the shared admission that a path from 0.1% to something real exists. What nobody can supply is the timeline.
So watch two things. Whether hyperscaler revenue starts catching up to that $658 billion in spending, and whether the productivity line finally moves off 0.1%.
Whichever way that number breaks, it will settle an argument your taxes, your benefits, and your portfolio are already invested in. For now, skeptics and believers are staring at the same tiny figure, waiting to see who blinks.
Related: Bank of America resets Apple stock price target on AI update
