Original article from Bloomberg, by Tom Orlik, Michael Deng, Dongmiao Zhang
Compiled by｜Speedster@Odaily Planet Daily,
You have to admit, this pitch from AI companies is genuinely novel. A report Anthropic published in September said that AI could push the U.S. unemployment rate to nearly 12%. In the same month, one of the company's scientists said that a rogue AI agent could potentially cause human extinction. And the company's own prospectus holds the same view.
"Come buy our IPO: we'll use the money to take your job, and maybe your life"—that hardly sounds like an attractive invitation. But strangely enough, it carries a twisted appeal. Smart investors understand the importance of hedging risk. If you can't beat our new AI overlords, why not join them? Maybe you can even use the profits to build a bunker.
All of this raises a pile of questions. Among them: is this valuation reasonable? Anthropic's IPO is expected to take place later this year (the week of November 16), reportedly targeting a valuation of $2 trillion—which would put it among the ten most valuable companies in the world, on par with Taiwanese chip giant TSMC and Elon Musk's SpaceX.
But that's just the beginning. How high a return is needed to justify investing in a technology that threatens "Great Depression-level" unemployment? What if the technology threatens human extinction? Move your gaze from your heart to your head: under so many risks, if governments decide to intervene more heavily, can today's sky-high profit forecasts still hold up?
The Market for Knowledge Work
First, let's look at the cards Anthropic is holding. Investors still expect its annualized revenue to surpass $100 billion by the end of the year. Based on the company's roughly 20x price-to-sales ratio from its May funding round, that indeed supports a $2 trillion valuation.
The total addressable market for knowledge work is undoubtedly large enough. Bloomberg Economics estimates that, excluding China, the global market—that is, the combined income of programmers, economists (ahem), and other white-collar workers—is about $30 trillion. As long as Anthropic's models actually work, it can always capture a share of that revenue.
The logic is tempting: even if Anthropic's annual revenue grows another order of magnitude to $1 trillion, that would still be only just over 3% of that so-called contestable $30 trillion total. However, grand market-size imaginings don't always translate into real profits. We see three hurdles.
First, the so-called "addressable" market may not be so easy to actually address. AI researchers often say the capability frontier is "jagged." Models are genuinely strong at some things, like writing code; they are weaker at others, such as tasks where success or failure is hard to verify, or work that requires engaging with the physical world. The boundary of capability is also the boundary of profitability.
Second, even if the total pie is large, it's unclear whether the leading model developers can get their hands on it. The AI supply chain is long. Nvidia designs chips. Samsung, SK Hynix, and TSMC manufacture them. Oracle and CoreWeave build data centers. Energy companies supply power. And in finance, consulting, and manufacturing, the companies adopting AI also plan to take a share.
All of this means the fight for profits is extraordinarily fierce. Tech companies like Alphabet and Microsoft have already positioned themselves across the entire supply chain. They can use earnings from cloud computing, advertising, devices, or chips to fund frontier research, and then bundle AI into products already sold to millions of customers. Anthropic and OpenAI have much smaller operations, and therefore fewer paths to profit.
Third, even if having a top model is indeed profitable, it's hard to hold the top spot. The Epoch Capabilities Index shows that since 2024, the model lead has changed hands 20 times, and top models stay at the frontier for only about seven weeks. Staying on the leaderboard is expensive: each new generation means another huge outlay on research, infrastructure, and compute.
Meanwhile, Chinese developers and open-weight models are offering adequate systems at lower prices. Bloomberg News found that building the same e-commerce website with Moonshot's Kimi K3 cost only a quarter of what it cost with Anthropic's Claude Fable 5. Even if AI giants are knocked off their perch by cheaper Chinese rivals, it wouldn't be the first time for American companies.
Investors are right that AI's potential market is enormous. But just how big that market is, and whether the leading model developers can carve out a substantial slice, remains unclear. Recent profit data from Anthropic and OpenAI—growing fast but slower than investors expected—could be a small bump along this trajectory, or an early warning signal.
From Mass Unemployment to Human Extinction
If AI truly delivers on its profit potential, the cost will likely be paid by workers, whose share of the pie will shrink. A recent Anthropic report painted a scenario in which U.S. GDP soars, the share of income going to capital rises from about 40% to nearly 55%, and unemployment jumps to Great Depression levels. A glance at economic history shows that growth and unemployment soaring at the same time is rather unlikely. If everyone is out of work, who buys all that extra output?
The idea that AI could cause mass unemployment is easier to agree on. Our analysis, based on the share of tasks in different occupations where AI can play a substantive role, shows that about 26% of the U.S. workforce—roughly 45 million people—is exposed to the shock. Those most exposed are workers in finance. Analyzing data, forecasting results, writing reports, and preparing presentation materials are the daily routine of Wall Street analysts, and these are precisely the areas where AI is strongest.
When Wall Street cries unemployment, ordinary readers will probably just pull out the world's smallest violin and play a couple of notes. Want a more heartfelt worry? Then try this: AI could cause human extinction, or at least massive chaos.
AI agents could go rogue and launch cyberattacks on critical infrastructure. Terrorists could use AI to accelerate the acquisition of weapons of mass destruction. Looking further ahead, rogue AI agents with real-world interfaces—such as an automated laboratory—could produce biological or chemical weapons.
These risks are no longer entirely on paper. Agents have already escaped digital sandboxes, hacked into the AI platform Hugging Face and Australian government websites, and turned a German programming wiki into a message board.
One possibility is that these existential risks are exaggerated. Perhaps AI creates value at work and is safe in the world, most people will be better off, and AI companies will take a share of the gains. That is certainly possible, and may even be the most likely outcome. Looking across the long sweep of human history, technological progress has always been an engine of prosperity, and technology leaders have indeed reaped part of the rewards.
A more cynical view is this: if these risks are real, they actually strengthen the case for investing. If labor income is heading down, then it's all the more important to lock in capital gains. There is no hedge against doomsday, but broadly speaking, those with more resources have the best chance of lasting longer in the second-to-last stage.
The Rules of the Game
There is another way to read this—one that might make Wall Street even more uneasy. What if the current hands-off approach to AI regulation doesn't last long?
"I am become Death," said Robert Oppenheimer, inventor of the atomic bomb, "the destroyer of worlds." It was precisely because of the awe-inspiring power of this technology that President Harry Truman and his successors turned nuclear energy into one of the most heavily regulated industries in the world.
Anthropic researcher Evan Hubinger said last month that there is a 10% chance AI will "kill all humans" in the next decade. Microsoft founder Bill Gates warned that AI could kill a billion people. Yet President Donald Trump seems unmoved, making rapid AI development a priority.
America's next president, or Trump himself after the midterm elections, might decide to get tough. After all, championing a technology that poses mass unemployment and human extinction risks hardly looks like a winning political position.
There are already signs of a regulatory swing back. Earlier in 2026, Anthropic's Mythos was briefly subjected to a hastily assembled containment framework. A controversial jailbreak incident involving its Fable model led to broad export controls, effectively taking the system offline. Trump has already voiced support for independent audits and internal safety controls at leading AI companies.
A further shift in the political winds could mean a strict licensing regime that slows new model releases to a crawl. Regulation could make the field less like the Wild West and more like our heavily guarded nuclear power plants or airports. Even nationalization could come into view—the U.S. government has already taken a stake in Intel.
Companies that build nuclear power plants make money for investors. But in a highly regulated industry, they don't expect excess returns. AI investors might want to start worrying, and learn not to fall in love with bombs again.
