AI is slowing down

Reaction to the article and tone

• Many readers find the piece emotionally charged, polemical, and “preaching to the choir,” which makes them distrust or tune out the arguments even when the numbers look interesting.
• Others defend the aggressive style as a necessary counterweight to relentless AI boosterism and corporate PR, though some think this has slid into ideologically driven “anti-AI content” rather than analysis.
• Several point out a history of dramatic, time‑bound collapse predictions that have not materialized, which for them undermines the author’s credibility, especially on anything forward‑looking.

Financial sustainability and bubble risk

• Core claim discussed: current capex, chip and data‑center commitments imply AI needs hundreds of billions to trillions in annual revenue by ~2030; many argue that’s implausible and signals a bubble.
• Some agree the math looks scary and see parallels to dot‑com, subprime, or Chinese real estate: circular deals, overbuild, and investor FOMO.
• Others counter that current ARR growth at major labs, strong enterprise demand, and historical precedent (railroads, internet) show that big bubbles can coexist with genuinely transformative tech.
• There’s debate over whether “AI is unsustainably expensive” vs “AI is valuable but currently sold below cost”; several note that even if the tech wins long‑term, today’s leading labs and some cloud/infra players could still get wiped out or bailed out.

Productivity, value, and organizational impact

• Heavy disagreement over “massive, undeniable productivity gains.”
  • Pro‑AI commenters report enormous speedups in coding and other knowledge work, widespread adoption inside companies, and say skepticism now ignores lived experience.
  • Skeptics cite studies of thousands of teams showing more code churn and bugs without throughput gains, personal experiences of no net speedup, and the fact that code is rarely the true bottleneck (coordination, product direction, integration are).
• A recurring theme: individual “I go faster” ≠ organization “we ship more valuable features.” Improved LOC/PRs can be offset by rework, tech debt, and loss of deep understanding.

Technical progress: slowing vs not

• One side argues progress is clearly slowing: recent frontier models feel like incremental upgrades; small/open and Chinese models are “good enough” and close behind; commoditization and shrinking moats loom.
• Others insist capability is still rising fast, especially with agentic coding and security tools (e.g., AI‑driven vulnerability discovery), and say we outsiders don’t see top models.
• Several distinguish “intelligence gain may be logarithmic” from “utility can jump when reliability crosses thresholds.”

Inference, infrastructure, and alternatives

• Strong debate over inference costs and margins:
  • Some lean on leaked or secondary financials to argue margins are thin and token‑spend is already “breaking budgets,” leading to per‑employee caps and tool reductions.
  • Others point to reports of high gross margins and the existence of profitable third‑party open‑weights hosts as evidence that inference can be economically viable as hardware and software improve.
• Concern that cloud build‑out may overshoot just as:
  • local models on consumer hardware get “good enough” for many tasks, and
  • cheaper open or Chinese models undercut frontier labs by large factors.
• Several predict a future where frontier labs are pressured, but AI itself persists via on‑device, open, or cheaper regional models.

Societal, labor, and environmental concerns

• Some commenters are explicitly “anti‑AI” on social grounds (culture, misinformation, democracy, junior‑talent hollowing‑out) even while conceding personal utility.
• Others argue that large‑scale layoffs and financial contagion from an AI crash could hurt many people whose retirement and jobs are indirectly tied to the bubble.
• Environmental worries center on data‑center power and water use, though some note AI queries can be more efficient than traditional search or offline alternatives; several draw analogies to cars and cheap oil: individually useful, systemically distorting.