A surreal gathering of angels before a dark city and tower
Image credit: @1dontknows

We typically think of high AI stock valuations as a reflection of optimism: expectations of explosive growth, dominance of AI tools and the long-term upside of technological leadership.

But what if that is only part of the story?

A provocative paper by economist Andrew Y. Chen at the US Federal Reserve Board introduces a striking alternative. AI stocks may also be priced as hedges against a worst-case AI future—a scenario in which AI progress decimates household welfare while AI assets thrive.

This idea changes how we might think about risk, equity pricing and artificial intelligence.

The concept: hedging the singularity

The paper, Hedging the AI Singularity, introduces a simple but powerful economic model. Its core thesis is that AI assets may gain value precisely when human livelihoods collapse because of AI disruption. That is what gives them hedging value.

It considers a potential negative AI singularity: an event in which AI development becomes recursive and self-improving, rapidly outpacing human capabilities. In such a scenario, labour markets collapse, human consumption falls and the average worker is economically displaced.

But while households suffer, AI-focused firms may surge and capture a larger share of the economy. Investors holding those stocks are partially shielded.

The model in a nutshell

The model contains two kinds of economic agent:

  • AI owners: entities that own and control AI infrastructure and are fully invested in AI assets.
  • The representative household: the marginal investor in the stock market, whose consumption determines asset prices.

Here is the crux of the model:

  • With probability p, a “disaster”—an AI breakthrough—occurs.
  • Household consumption falls dramatically by b.
  • The share of the economy owned by AI assets rises by h.

An AI stock’s dividend is tied to the size of the economy and the number of these disasters. As disasters accumulate, AI stocks’ earnings become more dominant even as the wider population’s consumption and wages decline.

Why it matters for asset prices

What emerges is counterintuitive: the worse the disaster for humans, the higher the price–dividend ratio of AI assets.

Investors—especially those with risk aversion greater than one—will pay a premium for assets that gain value when their own economic future deteriorates. AI stocks become a financial hedge against the collapse of labour-based income.

This cuts against the standard narrative that associates high stock valuations with bullish outlooks. It suggests that pessimism about AI’s effects on society might itself increase demand for AI equities.

Key insights from the paper

Using standard asset-pricing techniques, the model produces several results:

  • When risk aversion is high, AI stocks become more valuable in bad states of the world.
  • As disaster probability p and severity b rise, so does the price–dividend ratio.
  • Even mild increases in p or b produce disproportionate increases in valuations.
  • The paper’s Table 1 shows price–dividend ratios above 50× in some scenarios, even when disaster risk is moderate.

Financial markets may therefore be pricing AI stocks not only as growth bets, but also as insurance instruments.

Market incompleteness: the limitation

The hedge is imperfect, and the paper is careful about that. Most households cannot invest in the most advanced AI labs, including OpenAI, Anthropic and Cohere, because they remain privately held. Even if the economy booms because of AI, the average person does not participate fully in the upside.

This disconnect means AI-driven disasters remain net negative for most people even when partial hedging is possible. The model captures this through the disaster magnitude b: how much worse off the average household remains after accounting for the hedge.

Complementary to policy solutions

Much of the discussion of AI catastrophe risk focuses on regulation, safety and policies such as universal basic income.

The role financial markets could play in sharing risk is discussed less often. Chen’s paper suggests a complementary mechanism: even if we cannot prevent every form of disruption, the financial system may allow some people to hedge against it.

This is not a complete solution. Not everyone can access the relevant assets, and wealth and financial-literacy gaps limit participation. But the possibility adds another dimension to the debate.

Who should care?

  • Investors: AI exposure can be understood not just as a growth play, but as partial protection against labour collapse.
  • Policymakers: broader access to hedging tools might involve public investment vehicles or sovereign wealth funds.
  • Researchers: the argument opens the door to deeper modelling of rare disasters and AI’s asymmetric effects across the economy.

One more twist: the paper was written with AI

Chen produced the paper by prompting large language models, including ChatGPT and Claude, through the modelling, mathematics, literature review and writing. Appendix A contains a human-written README describing the experience. It is candid, reflective and—perhaps unsurprisingly—human.

Final thought

The paper does not argue that a negative AI singularity will happen. It argues that financial markets may already be pricing the possibility.

That is an interesting lens through which to view both public and private market valuations. If you are concerned about AI, economic inequality or the resilience of the financial system, the paper is worth reading.

Read the paper and its prompts: Hedging the AI Singularity.