There’s something odd about the way Anthropic is wooing investors

The prospectus of a company seeking to woo shareholders to invest their cash is normally filled with predictions of tremendous financial rewards and bright futures.

But reading the prospectus released this week by Anthropic, an AI company that is seeking to raise $US100 billion ($142 billion), something odd sticks out. The firm behind Claude warns potential investors they may never see their money again – because the world will have come to an end.

“Sufficiently advanced AI systems could … pose catastrophic or existential risks to humanity, particularly if such systems were misused, replicated, weaponised, or allowed to operate with a high degree of autonomy,” it noted.

This is a company seeking to build advanced AI systems that proponents hope will transform how we work, save us from terrible diseases, resolve our deep environmental problems and improve the quality of our lives.

It just happens to come with a great big asterisk that boils down to this: in saving the world, we may accidentally destroy it.

The world is in the midst of an AI building boom that dwarfs anything it has seen over the past two centuries. This year alone, an estimated $1 trillion will be spent on data centres and the computer servers to fill them.

Australia is riding that boom. Analysts from PwC estimate spending in this country on data centres could reach $157 billion by the turn of the decade and $342 billion by 2036.

If the boom actually accelerates, spending here could reach $428 billion or more than 10 per cent of GDP. It would easily eclipse the mining construction boom of the early 2010s.

The surge in AI-related construction has caught both the Reserve Bank and the federal Treasury by surprise.

In last year’s budget, Treasury had forecast non-mining investment to grow by just 1 per cent through 2025-26. It actually grew by 14 per cent thanks to the AI boom.

The scale of spending by AI-related companies is difficult to grasp, leaving many analysts to compare it with previous booms.

Over the past two centuries, the largest construction booms in the United States have been the canals of 1836 and 1841, the railroads built between 1870 and 1890, the rollout of electricity between 1905 and 1925, the federal highway system established between 1956 and 1973 and the dotcom rush of the late 1990s and early noughties.

None of them come close to the AI building surge.

Columbia University economist Stijn Van Nieuwerburgh last month estimated spending on AI infrastructure between 2025 and 2032 would amount to 3.63 per cent of US GDP. The next in size was the railroads of the late 19th century at 2.24 per cent of GDP.

Behind all this construction are corporate behemoths.

The largest company by value in the world is currently Nvidia, which produces graphics processing units, computer chips and other hardware vital to AI. It is worth $US5.2 trillion ($7.4 trillion), or roughly three times Australia’s annual GDP.

If Anthropic is successful in raising money from its public listing it will become the 20th business to be worth more than $US1 trillion. Of the existing 19, 13 are directly plugged into the AI boom.

AI companies are spending so much money because they believe this new technology will transform the economy and our lives.

As Anthropic argued in its prospectus, the new technology may revolutionise “every area of human endeavour”, from physics and biology to economics and robotics. Diseases such as Alzheimer’s and cancer may finally be cured, it said, while economic prosperity could accelerate and spread around the world.

“We believe that AI is one of the most consequential technologies in human history,” it said. “If managed well, it could rapidly accelerate scientific innovation, greatly increase economic growth rates, bring cures for major diseases within reach and solve problems that have resisted generations of human effort.”

Promised by AI, the future promised is, as the 1980s band Timbuk 3 famously noted in their one and only hit, so bright you’ve got to wear shades.

But with every technological breakthrough, there are doubts and fears. When people first saw Stephenson’s Rocket steam locomotive, some warned it would fail as the human body could not withstand travelling faster than 20 miles an hour.

The issues around AI are much more grounded in reality.

There’s the immediate impact of the construction boom. Spending by IT companies on equipment – that is, computer servers and the like – has soared 220 per cent over the past 12 months.

That type of spending is dragging resources away from other parts of the economy, be it workers or electricity.

Australia’s inflation figures this week provided an insight into how that fight for resources is playing out.

Prices for games and hobbies soared by 15.5 per cent over the year to September, the fastest rate since the bureau started to separately track this sector in 1998.

There hasn’t been an explosion in the prices of jigsaws or stuffed toys. The surge is due to the world’s computer gaming companies fighting cashed-up AI firms for increasingly expensive computer chips.

Between 2000 and 2025, the prices of computers have fallen by 91 per cent while their quality has improved. But over the past 12 months, prices for this equipment have climbed by 6.3 per cent.

That’s the impact of AI right now. Then there is the future.

Investors have pushed up the value of AI-related companies in the hope that these firms will make enormous profits.

Reserve Bank assistant governor Brad Jones this week noted that the “exuberance” around tech companies was predicated on something that would be unparalleled in economic history.

“The earnings expectations for the US IT sector into perpetuity, beyond the next three years, up until recently, was running at about 45 per cent per annum forever,” he said.

“Far be it for me to opine on whether that’s irrational or not, but clearly there was a lot of good news in the price.”

This week Alphabet, which owns Google and YouTube, suffered a 10 per cent drop in its share price despite revealing a 108 per cent jump in revenues from its AI cloud business. Investors, however, were spooked by its plan to spend up to $US200 billion on infrastructure next year and whether it could make sufficiently large profits to cover that sort of expenditure.

The debt powering all this spending is of increasing interest and concern to people like Jones.

The Bank for International Settlements (BIS) is the central bank to the world’s central banks. One of its jobs is to take a step back and consider the globe’s financial and economic risks. The way the AI sector is financing its construction is one of those risks.

The bank’s chief economist, Phurichai Rungcharoenkitkul, earlier this year released a paper examining the size of the boom, how it compares with previous ones and the unusual way the AI surge is being financed.

AI companies are engaged in what they call “circular financing” – a large tech company or chipmaker invests or lends money to an AI start-up.

That cash gets the start-up going, and it must buy chips or cloud services from the company that gave it the money in the first place. By some accounts, up to 55 per cent of the investment in AI firms is coming from other AI businesses.

This arrangement, given the huge scale of money involved and the way the global economy is growing because of the AI boom, is making organisations such as the BIS more than a little nervous.

As Rungcharoenkitkul noted, booms by their very nature boost an economy until they end in a bust.

“This boom-bust pattern recurs across history, from the US canal mania in the 1830s and the British railway mania in the 1840s, to the roaring 20s and the dotcom boom in the late 90s. These episodes all ended in sharp corrections, with wider economic fallout,” he found.

Rungcharoenkitkul said while the demand for AI services was vast and could justify a huge investment boom, the size of what was taking place and the interlinkages between companies was a substantial economic risk.

The bigger the boom, the bigger the bust. “[The] AI boom creates fragility that undermines itself,” he concluded. “The more capacity the sector builds, the higher the productivity bar it must clear to sustain the boom, so a larger boom is both more likely to disappoint and more damaging when it does.”

Ratings agency Fitch this week put some numbers on the scale of economic trouble caused by an AI boom-bust.

It found a 35 per cent drop in share prices over a six-month period, due to investors losing their love for tech companies, would drive the US into recession exceptionally quickly. The drop in share prices would also force tech companies to ditch their AI construction plans, which would further hurt the economy. All up, the American economy would be 3.2 per cent smaller in a year’s time.

ANZ chief economist Richard Yetsenga said ultimately AI and the construction boom underpinning it would deliver long-term economic and productivity improvements that could dwarf previous technological advances.

He noted that, like past infrastructure booms, what is being built now would last long into the future, providing ongoing economic benefits if not huge profits to firms that financed the data centres and computer servers.

The questions over AI, particularly those facing governments and even companies such as Anthropic, such as whether the technology leads to the end of humanity, will have to be resolved at a political level.

“We’ve never had anything similar to the use of computer chips and the level of technological change that AI is offering,” he said. “Every new form of technology has prompted new policy imperatives that need to be resolved and that seems to be the same with AI.”

This week, the Australian Signals Directorate (ASD) released its annual report on the cyber threats facing the country.

ASD Director-General Abigail Bradshaw said AI technologies had changed the cybersecurity landscape. “[They] are creating enormous opportunities, but they are also enabling malicious cyber actors to identify vulnerabilities and conduct attacks faster, at greater scale and with greater precisions than ever before,” she said.

The agency believes AI will help it prevent the avalanche of attacks on the nation’s electronic infrastructure. But that avalanche is being driven by AI.

As Anthropic argued, if AI is directed at humanity’s hardest problems and risks managed, “the potential is extraordinary”. That’s if it doesn’t wipe out humanity in the process.

The infrastructure for AI is facing strong pushback from affected communities. Rampant construction is putting pressure on water and electricity systems with examples in the United States of residents facing higher utility bills for their new noisy data centre neighbours.

While supporters such as Elon Musk argue AI could lead to extraordinary levels of economic growth, critics warn it could lead to mass levels of unemployment as the technology replaces workers.

These are complaints that have accompanied almost every transformative technology-driven boom since the dawn of the industrial revolution.

The outlines of past booms are still evident in this country. In the 1880s, almost every colony went through a railway building boom. Victoria, then the nation’s economic centre, had the largest. Its network of rail lines more than doubled in less than a decade, from 1800 kilometres to more than 4600 kilometres.

This delivered Melbourne’s inner circle line and its outer circle line. The inner circle lasted until the 1940s and is now a cycle path, while the outer circle, which opened in 1891, started being closed down in 1893 although parts of it survive or were reused.

The railway boom created a rail network that still services much of Victoria. But the huge losses that accompanied it were key factors in the 1890s depression that crippled Melbourne for decades.

Railways, the data centres of their day, helped end horse-drawn carriages as an industry. But they didn’t threaten an end to humanity.

Start the day with a summary of the day’s most important and interesting stories, analysis and insights. Sign up for our Morning Edition newsletter.