This is how the AI bubble pops | Cory Doctorow

Cory Doctorow’s Core Claim

Cory Doctorow argues that today’s AI boom is best understood not as a steady march toward humanlike “true AI,” but as the next phase of capitalist “squeezing” after companies exhaust growth in existing markets.

He frames this as “inshtification”—optimizing profits by degrading products and replacing real labor with unreliable systems—and he suggests AI is serving that role. At the same time, he emphasizes AI also carries a built-in risk of collapse.

---

AI’s Role in the “Bubble”: From Squeezing to Inventing New Markets

Doctorow distinguishes between:

• AI the narrative: inevitable and transformative, replacing labor
• AI the technology: statistical systems that generate outputs

He argues the bubble exists because investors and wealthy elites view AI as a path to a “world without people”, where workers are replaced and the “power” employees gain from process knowledge is reduced.

He also cites Ed’s theory (as referenced during the discussion):

“OpenAI is giving away inference and calling it marketing.”

Doctorow’s key point is that even if expenses are relabeled, inference costs don’t disappear—accounting and marketing can mask ongoing unsustainable burn.

---

Why the Narrative Took Off (and Why It Persists)

Doctorow claims the narrative is effective due to:

• Financial incentives: billionaires and capital markets reward growth stories. When firms can’t grow anymore, they must create new “imaginary markets” to justify valuations.
• Two billionaire mindsets:
  1. A cynically dehumanizing view where people are treated as replaceable statistical abstractions.
  2. A more sales-driven cynicism: replacing expert workers with chatbots reduces friction from human expertise and customer confrontation.

He connects the current boom to earlier manias (metaverse, Web3, NFTs), arguing that capital keeps funding new stories whenever older growth runs out.

---

Why AI Will Not Deliver Its Grand Promises

Doctorow argues the tech broadly labeled “AI” today is mainly:

• statistical correlation / prediction
• not understanding

He says this “lucky guess” approach can yield impressive outputs, but it has hard limits when scenarios demand grounded understanding rather than repeating known patterns. He illustrates this with an analogy about predicting a spouse’s speech:

• Humans respond because of understanding
• LLMs often succeed because people repeat patterns

He also predicts progress will continue with diminishing returns: newer model generations deliver smaller gains relative to their cost.

---

Safety, “God” Rhetoric, and Weak Safeguarding

On claims that AI leaders are building something godlike or beyond nuclear weapons, Doctorow argues:

• the lack of credible safety/testing strongly suggests leaders aren’t telling the full truth, or
• they aren’t behaving like people facing an imminent existential threat

He compares it to building a nuke while claiming you have a plan—except, in his view, the build proceeds like an unconstrained general-purpose weapon.

---

The Mechanism That Could Pop the Bubble

Doctorow argues bubbles burst due to brittleness and offers several interacting failure modes:

• Financial structure and related-party loops Major AI players rely on massive borrowing, and suppliers—especially chipmaker ecosystems—are also deeply entangled financially.

• Operational bottlenecks Data center construction and energy availability depend on financing and supply chains. Contingency failures could cascade across many sites at once.

• Company-internal fragility He describes AI companies as hot-houses and points to internal instability—referencing OpenAI’s chaos around leadership after Sam Altman’s firing—as another risk.

• “Trigger” via revealed weakness He compares likely collapse dynamics to past bubble detonations where the true weak link is revealed only after the fact (a “casualty bias” style effect).

---

What Happens to People and “Process Knowledge” After the Dust Settles

Doctorow’s key social risk is:

• AI-driven layoffs replace workers with chatbots
• but when the bubble collapses, rehiring doesn’t restore original expertise

He emphasizes process knowledge—the tacit, on-the-job know-how that can’t be fully documented. Destroying it can impose multi-year productivity and quality losses across industries.

Even if statistical inference remains useful, he predicts the rollout pattern from the boom—pushing unreliable systems into broad production—will likely unwind.

---

Demand, Infrastructure Oversupply, and the “Death Spiral” Risk

Doctorow argues demand for basic LLM-like outputs will exist, but the question is whether firms can afford their fixed costs.

If users leave as prices rise or budgets tighten:

• remaining users absorb higher costs
• that can create a death spiral

He also expects data center overbuild and eventual underutilization, with some infrastructure possibly stranded or reallocated.

---

The “Reverse Centaur” and Labor Resistance

He discusses his “reverse centaur” framing:

• human labor does the grunt work under an algorithmic “top”

He advises workers to resist messaging that frames AI as inevitable job theft. Instead, he emphasizes resistance should target the real issue:

the boss replacing jobs with systems that can’t do the job reliably

He points to Hollywood screenwriters and unions as examples where labor power meaningfully constrained AI substitution.

---

What Survives: Incremental, Efficient AI—not Humanless Utopia

Doctorow predicts:

• Incremental components will remain useful (e.g., programmers and limited “world-model” systems that use LLMs for specific tasks)
• Open-source improvements will continue
• the hype-driven “swarm of thousands of LLMs” approach is likely to be inefficient and fade

Ultimately, he expects AI to behave like other technologies: gradually improving rather than delivering a once-and-for-all transformation—while the economic harm from premature labor substitution and process knowledge destruction may persist long after the bubble ends.

---

Presenters/Contributors

• Cory Doctorow
• Isaac (host / “Tech Report” presenter)