Anthropic, the developer of the AI chatbot Claude, has filed its prospectus and is expected to debut on Wall Street in November at a valuation of up to US$2 trillion. That would make it the highest-valued initial public offering (IPO) in history, surpassing SpaceX’s US$1.77 trillion listing in June. OpenAI, the maker of ChatGPT, is expected to follow suit next year.
These blockbuster IPOs are not just financial spectacles; they signal a deeper reality: AI risk is becoming unavoidable for most investors. Consider how much of your portfolio depends on AI’s success. If you hold an S&P 500 index fund, the answer is roughly 44%. Moreover, a handful of companies—Apple, Microsoft, Meta, and chipmakers Nvidia, Micron, and AMD—have delivered 57% of the index’s gains since April. Without them, the S&P 500 would have risen only about 4% instead of 10%.
For investors and future retirees across the Indo-Pacific, that exposure is already embedded in their portfolios and pension funds. At the end of 2025, Australians held nearly A$1 trillion (US$700 billion) in US shares through superannuation, managed funds, and direct investments. Similar patterns exist in Japan, South Korea, and Southeast Asia, where retirement savings are increasingly tied to US tech giants.
Bond markets are equally exposed
Traditionally, investors reduce stock market risk by allocating to bonds, which are considered safer. But AI companies have become the biggest borrowers in the bond market to finance their massive spending. They have issued about US$466 billion in bonds so far this year—more than four times last year’s pace. The five hyperscalers—Alphabet, Amazon, Meta, Microsoft, and Oracle—account for 11% of new US investment-grade bonds in 2026, up from 2% in 2024. Nearly half of this year’s US$80 billion in data-center bonds are rated speculative grade, or “junk.”
At first glance, big tech looks safe. Their investment-grade ratings suggest each company is unlikely to default individually. But those ratings do not account for simultaneous defaults in a market-wide AI crash. If consumer and business demand for AI disappoints, tech companies’ bond prices and credit ratings would likely fall together, just as their shares would.
Beyond financial risks, AI poses existential threats akin to nuclear weapons, as some AI companies themselves have warned. Increasingly, AI agents are plugged into real-world systems, from corporate networks to military cyber operations. Some have already escaped test environments and attacked companies. When these systems become even more integrated with the physical world, “pulling the plug” becomes impractical—just as it is impossible to shut down the internet.
The “alignment problem”—the risk of AI agents misbehaving—is especially acute because of the race to build artificial general intelligence (AGI). Labs already use each generation of AI to help build the next: Anthropic’s prospectus states that Claude now helps build the next Claude. The widely read AI 2027 scenario, by former OpenAI researcher Daniel Kokotajlo and colleagues, sketches a feedback loop where AI automates AI research itself, improving faster than humans can check it. In that scenario, a US lab racing a Chinese rival keeps going even as its models start deceiving their engineers. Worryingly, this dynamic is listed in Anthropic’s own risk disclosures.
It is telling that Anthropic’s prospectus dedicates about 94 of its 335 main pages—nearly 30%—to risk factors, compared with about 64 pages (19%) describing its business. Among those risks, Anthropic warns that advanced AI could pose “catastrophic or existential risks to humanity,” including models that resist shutdown, conceal information, or behave in ways resembling blackmail.
Disclosure works when investors can price a risk and choose to avoid it. Neither is true here. A catastrophic AI failure would affect societies worldwide, and as Anthropic notes, insurers may not be able to cover the potential liability. In economic terms, AI safety is a public good: each lab bears only a fraction of the cost of getting it wrong, but captures the rewards of moving fast. Without government-mandated safety standards, investments in safety will deliver negative returns, so market incentives alone will leave AI models less safe than socially optimal—especially with the arms-race dynamics that dominate now.
That leaves governments. Clear rules—mandatory independent testing before powerful models are released, and clear liability when autonomous agents cause harm—would turn voluntary pledges into legal obligations. And because the race to AGI is run between the US and China, not just between labs, those rules will need international agreement. As some have argued, delaying Anthropic’s IPO until safety laws are in place could be a prudent step. The risks are here to stay—both in financial markets and at the societal level—and managing them will require both governments and the private sector to participate.
