CAPITAL ACCOUNT

By Greg Ip

Is AI’s 9% GDP Goal Plausible?

Believers in the artificial- intelligence boom need to take a close look at this number: 9% of GDP.

That is how much American businesses and consumers eventually have to spend per year on the services of companies like Anthropic and OpenAI to justify the sums being committed to the technology right now.

Is it plausible that Americans will spend as much of their income on this one technology as they do on food? Roughly twice what the nation pays for all forms of energy or all computers and software? Seven times what consumers spend on phone, streaming, and internet services combined?

You should be skeptical. Even the most transformative inventions run into the law of diminishing returns: each additional dollar a user spends yields less additional productivity than the last. That imposes a natural ceiling.

The question, of course, is where that ceiling is. Whether or not you think 9% of GDP is right, you have to care, because this figure isn’t some fever dream: it is implicit in the dollars that investors and companies are committing right now.

The figure is courtesy of Columbia University finance professor Stijn Van Nieuwerburgh. His paper, presented at the Brookings Institution, showed the AI build-out is bigger than any investment boom in American history. His more intriguing, and sobering, statistic is how much revenue AI would have to garner to justify that boom.

Van Nieuwerburgh started with the actual plans of today’s builders of data centers and the average cost of a data center, then assumed some data centers would be canceled, that each dollar of revenue produces 50 cents of cash flow, and that builders need an unleveraged return on invested capital of 10%.

He concludes from all this that AI revenue in 2032 would need to be $3.5 trillion, or 8.8% of GDP (which he assumes grows 4% per year, unadjusted for inflation).

A key assumption in this forecast is that the AI companies will be able to keep charging today’s prices for computing power. But Van Nieuwerburgh said the scarcity of computing capacity is holding up prices and profit margins now. The companies are “basically saying we’re going to keep charging scarcity pricing in 2032, when presumably there will be tons of competition and we’ve quadrupled our capacity.”

Plunging prices helped undo previous investment booms. Between 1997 and 2001, the price of bandwidth on fiber between London and New York plunged 96% as more fiber strands were laid and new technology expanded the capacity of each fiber. That price collapse accelerated the bankruptcy of numer--ous long-haul fiber companies.

In tech, falling prices are the flip side of rising efficiency. AI models double in capability roughly every four to five months. The effective price for a given level of capability has fallen 47% per quarter since 2023, according to Epoch AI, six times faster than for computing power.

The optimistic case for AI is that a falling price and rising capability spurs more than enough demand to offset the falling price, a relationship dubbed the “Jevons Paradox” after British economist William Stanley Jevons who noted the phenomenon with coal power in 19th-century England.

That certainly seems to be the story so far. A few months ago, an expert panel surveyed by a team led by Ezra Karger at the Federal Reserve Bank of Chicago, projected OpenAI and Anthropic would have combined annual revenue of $300 billion by 2030.

That looks way too conservative: the two companies’ combined run rate is already around $180 billion, and growing at double-digit rates each quarter.

AI evangelists say artificial intelligence will, in effect, become a third factor of production, alongside capital and labor. In that case, 9% of GDP doesn’t seem so much. (Labor now gets about 51% of GDP.) Of course, in such a scenario, employment would plummet and governments would have to tax AI to make up for lost income and payroll taxes.

The more prosaic risk is that current growth rates won’t be sustained, once adoption is widespread and the most productive applications deployed.

From the 1960s through the early 2000s, the density of microchips doubled roughly every two years, a phenomenon dubbed “Moore’s Law,” and the real price of computers fell roughly 15% per year. The personal computer and killer apps such as spreadsheets fueled an explosion of demand that far offset that falling price: between 1974 and 1984, business investment in computers and peripherals tripled to 0.8% of GDP.

Thereafter, computers continued to get better and cheaper, but didn’t spur the same new demand. Computers were now ubiquitous. The first versions of VisiCalc and Lotus 1-2-3 revolutionized business in ways that ever more powerful versions of Microsoft Excel couldn’t. So business spending on computers plateaued, except for secondary booms fueled by the internet in the late 1990s and AI now.

AI is improving and adoption is growing much more quickly than with computers, and productivity improvements have been impressive.

Harvard economics doctoral students Fiona Chen and James Stratton reviewed software projects through the platform Jellyfish, which firms use to analyze their engineering teams. AI assistants, they found, boosted lines of code by 12%, and “pull requests,” when new code is merged into an existing code base, by 5%. AI agents, which can operate autonomously from humans, boosted lines of code by 30% and pull requests by 23%.

And yet, the authors found, AI’s boost to completed projects was small and “statistically insignificant.” The reason was that users were spending much more time reviewing, commenting on, and changing pull requests.

The point is that AI will almost certainly raise productivity and economic growth, but not necessarily in ways that translate directly into revenue, or the fat returns investors are counting on.