A bubble that knows it's a bubble

Is AI a bubble and how big?

• Many commenters think we are in a financial bubble around LLMs, not a “capabilities bubble”: the tech is real, but valuations and spending are ahead of sustainable economics.
• Some argue even a modest, durable 2x software dev productivity gain would justify large valuations; others say such gains are unproven and mostly appear in toy/prototype work.
• There’s concern that markets have already priced in near‑ubiquitous AI automation; if progress plateaus, a sharp correction is expected.

Compute, GPUs, and “infrastructure”

• Strong disagreement on whether current GPU build‑out is analogous to railroads or fiber:
  • One side: GPUs become obsolete in ~3–5 years; this is not durable infrastructure.
  • Other side: data center buildings, power, cooling, fiber, improved logistics and fab capacity are long‑lived and will enable future uses, even if current GPUs are scrapped.
• Some note Moore’s Law is slowing, so current compute might stay “good enough” longer than past clusters. Others see this optimism as grasping for silver linings.

Robotics and humanoid hype

• Several expect real long‑term value in robotics, but view the current humanoid craze as over‑engineered, expensive, and bubble‑like.
• Wheels vs legs: wheels are cheaper, more reliable, and adequate in many ADA‑compliant environments; bipedal robots only make sense where humanlike mobility is essential.
• Discussion branches into disability tech (wheelchairs vs exoskeletons), with cost, simplicity, and safety cited as reasons wheelchairs still dominate.
• Privacy and cloud dependence are flagged as major barriers to household robots; luxury and business markets may appear first, deepening inequality.

Economics of AI companies

• Example: huge raises at multibillion valuations with large losses are seen as classic bubble markers; defenders invoke “grow at all costs” playbooks and VC power‑law returns.
• Skeptics stress that most such companies historically fail; high margins are uncertain given intense competition and compute costs.
• There is debate over whether regulatory action (especially in Europe) will constrain dominant AI platforms and their hoped‑for winner‑take‑most economics.

Historical analogies and creative destruction

• Comparisons span railways, fiber, dot‑com, housing, VR, 3D printing, crypto, and Japan’s long stagnation.
• Some endorse the “victims unknowingly funded the future” view (fiber after dot‑com); others note many bubbles (VR headsets, some hardware) leave little reusable infrastructure.
• A subthread clarifies that capital, money, and real productive capacity are distinct: bubbles can destroy useful time and misallocate resources even if money is later “recreated.”

Altman, incentives, and regulation

• Altman’s simultaneous “bubbly” rhetoric and talk of “trillions” in AI investment is seen by some as price‑talk to suppress startup valuations and intensify regulatory moat‑building.
• Others see self‑contradictory messaging on AI existential risk and regulation as evidence of self‑interested attempts at regulatory capture: “AI is dangerous, so only we should build it.”

Systemic risk and leverage

• Several note that unlike past manias, retail investors have limited access to early AI equity, which may reduce broad household devastation if/when it pops.
• Others worry less about immediate financial collapse and more about a lost decade of misdirected engineering talent and underfunded “real” public research.

Who survives / what remains?

• Speculation ranges from “big cloud and chip vendors will be fine” to extreme scenarios where a crash plus geopolitics and climate undermine major Western tech firms.
• More moderate voices expect:
  • Data centers and power/fiber build‑out to persist.
  • AI tools to remain as niche but valuable productivity aids (similar to 3D printing).
  • FOSS and local, non‑cloud software to be relatively resilient.