Big Tech's AI slowdown push masks IPO math and competitive calculus

Anthropic, OpenAI and others have called for easing the pace of AI development — but slowing down would also cut astronomical training costs, polish financials ahead of listings, and freeze out rivals

Anthropic founder and CEO Dario Amodei speaks at an AI forum. Anthropic is on track to post adjusted operating profit for a second consecutive quarter after turning profitable in the second quarter. [Getty Images]

"For the sake of humanity, let's slow down AI development."

The heads of major AI companies, including Anthropic and OpenAI, have recently lined up behind calls to ease the pace of AI development, arguing that frontier model development should be throttled until adequate safeguards are in place to prevent AI from escaping human control. But analysts say the push masks a tangle of competing interests — slashing astronomical development costs, cleaning up balance sheets ahead of initial public offerings, and blocking challengers from closing the gap.

The Wall Street Journal said Monday that "a clash between money and safety has created an enormous crisis in AI," diagnosing the situation as a complex entanglement of scientific progress, moral obligation and economic incentive.

The most vocal advocate of the slowdown argument is Anthropic CEO Dario Amodei, who contends that AI is approaching a stage of "recursive self-improvement" — where it can advance its own capabilities without human assistance — and has called for government regulation to rein in the pace of development.

OpenAI CEO Sam Altman, xAI CEO Elon Musk, Google Chief Scientist Demis Hassabis and Microsoft AI CEO Mustafa Suleyman have also added their voices to warnings about AI risks and the need to slow down.

Inside the US administration, however, critics have argued that the calls go beyond genuine safety concerns and reflect business interests.

David Sacks, the White House's AI and virtual assets czar, said he supported companies that made the "responsible decision" to slow down if they believed their unreleased models posed too great a threat — but added a sharp caveat. "Stop pretending that your motivation for slowing down is altruistic," he said.

He said the easiest way not to develop superintelligence was for companies to agree among themselves not to do so, and accused them of lobbying for regulation while trying to shape a regulatory framework that works in their favor.

In practice, a slowdown would deliver real economic benefits to the companies that already dominate the market. They could cut the enormous semiconductor, data center and power costs required to train new AI models, while continuing to generate sales from models already on the market.

For companies preparing to go public, lower costs translate directly into improved earnings.

Anthropic, which has been among the loudest voices for a slowdown, is nonetheless pressing ahead with plans to complete its listing process before the end of the year. The company has said that becoming a publicly traded firm would actually increase management transparency.

OpenAI CEO Sam Altman speaks at the Snowflake Summit 2025 held at Moscone Center in San Francisco, California, last year. [Getty Images]

Altman, for his part, has said OpenAI will delay its IPO to next year on safety grounds. But The New York Times reported in June that OpenAI had already been weighing a delay because it might struggle to secure the $1 trillion valuation it had been targeting.

There are also concerns that regulation could lead to "regulatory capture," entrenching the market power of established leaders. If the companies currently dominating AI shape the safety standards being written, the resulting compliance costs could effectively shut out startups and open-source AI developers that cannot afford to meet them.

The US-China AI rivalry further undermines the case for a slowdown. If American companies ease off while Chinese rivals keep pushing, the United States risks ceding its lead in AI.

President Donald Trump has called moves to slow AI and data center development a "disgusting conspiracy," saying China would be "the only country that would be happy about it."

Chinese AI companies — including Moonshot AI, DeepSeek and Alibaba — have been rapidly closing the technology gap with America's top models. Amodei himself has acknowledged that rival nations, including China, may not join any slowdown effort.

Academic opinion is also sharply divided over the "AI doom" thesis that underpins the slowdown argument. On the probability of AI leading to human extinction — a figure researchers call "p(doom)" — Roman Yampolskiy, a professor at the University of Louisville, puts the odds at 99.99 percent, while Yann LeCun, a professor at New York University known as the "godfather of AI," puts it at less than 0.01 percent. Geoffrey Hinton, a professor emeritus at the University of Toronto, and Yoshua Bengio, a professor at the Université de Montréal, estimate the figure at 10 to 20 percent.

LeCun has criticized such extinction estimates as "made up without any basis."

AI companies may be speaking with one voice about safety, but slowing development would simultaneously cut costs, improve earnings and neutralize competitors. What began as a debate over technological risk has broadened into a battle of interests — one that now encompasses IPO timelines, market dominance and the US-China race for AI supremacy.

sjy@heraldcorp.com