No new technology has ever attracted as much money as artificial intelligence. The amounts companies are pouring into it already exceed even what was once consumed by railroad construction or the internet boom of the late 1990s. Economists warn, however, that the bold plans and enormous corporate valuations rest on the assumption of sharp productivity growth and future profits, for which there is still neither evidence nor historical precedent. Technology companies are therefore racing against time. Artificial intelligence must start generating substantial profits before the money runs out and loan repayments come due.
Spending on a scale the world has never seen
Consulting firm PwC has calculated that the world could invest more than thirty trillion dollars in data centers alone by 2050. That is almost as much as the current value of all outstanding U.S. government bonds. According to PwC, this spending easily eclipses both the railroad and internet booms, even after adjusting for inflation.
The scale of the money at stake is also illustrated by a single company. According to an initial public offering prospectus reviewed by Reuters, Anthropic expects to spend five hundred and eighteen billion dollars over the coming years. That is more than one hundred times its revenue last year.
The center of the action is in the United States, which, according to some estimates, accounts for roughly three-quarters of global investment in AI. Economist Stijn Van Nieuwerburgh of Columbia Business School estimates that Americans will invest as much as roughly nine trillion dollars in it between 2025 and 2032. More than three percent of U.S. GDP would thus go toward artificial intelligence each year. For these investments to generate even a ten-percent return, the U.S. AI sector would need to earn approximately three and a half trillion dollars annually by 2032. Today, it takes in only a fraction of that amount.
Someone has to pay the bill
Consulting firm Bain & Company takes a similarly sober view of the situation. According to its study, operators of massive cloud platforms such as Google, Amazon and Microsoft, together with other companies, must find more than 4.2 trillion dollars in new revenue within five years simply to finance the construction. According to analysts, improved performance in existing industries will not be enough. Entirely new markets must emerge, such as AI-controlled robots or the development of new materials for batteries and chips. The study's authors question whether such applications will arrive soon enough to pay for all the construction.
In August, JP Morgan wrote that broad-based productivity growth had yet to materialize in the United States, casting doubt on whether current corporate valuations are sustainable. Using the example of Nvidia, whose chips power most of today's artificial intelligence, the bank calculated that U.S. productivity would need to grow by three to five percent annually over the next ten years for the company's valuation to make sense. The U.S. Congressional Budget Office, however, expects growth of around 1.75 percent over the same period. According to the bank, history shows that technology booms usually end when infrastructure development ceases to generate sufficient returns.
The method of financing also increases the risk. Van Nieuwerburgh points out that a large share of data centers is being funded with highly leveraged borrowed money. Even a modest decline in demand, a delay or a drop in asset values can therefore turn into much larger losses.
Executives talk of miracles
The staggering figures are not deterring the heads of U.S. companies developing artificial intelligence. Anthropic's Dario Amodei described a future with AI as something of transcendent beauty. OpenAI's Sam Altman predicts that new discoveries will come at an enormous pace once models learn to improve themselves and accelerate scientific progress.
Jasjeet Sekhon, chief strategy officer at Google DeepMind, considers artificial intelligence's ability to improve itself, technically known as recursive self-improvement, one of the pillars of the entire investment thesis. At an August gathering at the University of California, Berkeley, he said that if it can be mastered, it could deliver an unprecedented leap in productivity. But the same capability also raises fears that development could spiral out of control and threaten humanity itself.
Productivity is taking its time
Even the fastest progress may not keep pace with the timetable demanded by corporate accountants. Economist Diane Coyle of the University of Cambridge notes that with previous transformative technologies, the impact on productivity usually took ten to fifty years to emerge.
Anthropic's economists modeled several possibilities for how much AI could accelerate annual economic growth in 2030. If the economy would grow by two percent without artificial intelligence, growth would reach 2.4 percent with a moderate impact, 5.4 percent with a significant impact and as much as 15.4 percent in an extreme case. According to them, however, faster growth would also mean more jobs lost. The company did not estimate the probability of the individual scenarios.
Even if the promised transformation takes longer than the figures surrounding AI companies suggest, economists say the real benefits should not disappear. Trains continued to run after the crash of 1873, which ruined American railroad magnates, and the internet continued to function after the dot-com bubble burst. Coyle takes a similar view. If the boom leaves behind infrastructure on which future productivity growth can build, she says, it is no catastrophe.
Graduates are already feeling the change
Last year, Amodei warned that artificial intelligence could wipe out half of entry-level office jobs within five years. So far, according to some researchers, the situation does not appear quite so dramatic; rather, the chances of finding a first office job have worsened. Studies from the United States and Britain show that companies are hiring fewer entry-level workers for positions in which AI performs well, even though overall employment remains strong.
In August, researchers at Stanford University found that among people aged twenty-two to twenty-five working in fields that artificial intelligence can easily handle, such as accounting or legal assistance, employment is nineteen percent lower than in occupations that AI would find difficult to replace. These include cleaners and construction workers, for example.
Source: reuters.com
