An AI bubble threatens Silicon Valley, and all of us

Original Article ↗ Hacker News Discussion ↗

Profitability, Moats, and Bubble Talk

  • Many see classic bubble signs: huge spend, little profit beyond Nvidia (“selling shovels in a gold rush”).
  • Foundational-model companies are viewed as especially fragile: open-source and domain-specific models (e.g., DeepSeek-style) can undercut them, so their long‑term moat is unclear.
  • Some compare AI to airlines or dot‑coms: very real and useful, but structurally low-margin and overfunded on unrealistic profit expectations.
  • Others push back that calling everything a bubble is lazy; AI clearly has real utility and is already creating value, so a hype cycle/correction is more likely than total collapse.

OpenAI, Anthropic, and Closed vs Open Models

  • There is worry about OpenAI’s viability: talent losses, rising prices for newer models, reliability issues, and strong open-weight competition.
  • Debate over whether high prices are justified by quality (especially reasoning models) or just branding; some liken the strategy to Apple, others to a fading search engine.
  • Closed providers are criticized for restrictive output terms that explicitly block competitive uses; some see this as anti‑competitive and fragile if regulators step in.
  • Disillusionment is strong about OpenAI’s shift from a loudly non‑profit, open, “for humanity” stance to a closed, profit‑driven posture; some argue that original ideals were always mostly PR.

China, DeepSeek, and Protectionism

  • Several expect US protectionist measures to shield domestic firms from Chinese models: bans on Chinese AI, app distribution, or even downloading models.
  • A proposed US bill with harsh penalties for “importing” Chinese AI is cited; its exact scope (especially for open weights run locally or on US clouds) is seen as ambiguous.
  • OpenAI is reported to be lobbying to restrict Chinese models on security grounds, while US hyperscalers already host some of them, complicating the narrative.
  • Others welcome Chinese open-source pushes as a geopolitical check on US firms’ ability to lock in closed, rent-extracting AI.

Developer Productivity, Deskilling, and Real-World Use

  • Experiences with AI coding tools are sharply mixed:
    • Some developers and data scientists report meaningful gains, especially for scaffolding, boilerplate, planning, and “rubber‑duck” style problem exploration.
    • Others see no net speedup or even regress: more subtle bugs, heavy verification overhead, and “cognitive rent” paid later when maintaining AI‑generated tangles.
  • Vendor-funded studies claim 20–50% productivity boosts; these are treated skeptically as marketing. Independent evidence is described as thin and methodologically unclear.
  • A recurring theme: AI makes generation cheap and verification expensive, worsening spammy text/code and making serious review harder.

Cognitive Offloading and Cultural Concerns

  • Analogies to GPS: tools that make navigation/knowledge work easier can quietly erode human skill and situational awareness; people can stop really learning their “terrain.”
  • Some see AI as part of a broader management agenda to deskill workers, reduce bargaining power, and end-run around “uppity” knowledge labor.
  • There is worry that AI will supercharge low‑value uses (spam, scams, ad/banner optimization) more than high‑value ones, and that monetization (ads, product placement in answers) will corrupt LLM outputs like SEO corrupted search.

How Big Is the Transformation?

  • Optimists argue recent breakthroughs (LLMs, image models, new reasoning methods) make superhuman general intelligence within years plausible, and point to likely disruption in huge sectors such as transportation and healthcare.
  • Skeptics counter that this rhetoric rhymes with previous manias (blockchain, VR): strong “this could be huge” vibes, but so far mostly incremental tools—better autocomplete, search, and content generation—rather than wholesale replacement of skilled labor.
  • A common middle view: AI is clearly useful and here to stay; there will be a correction as speculation outruns sustainable business models, but the underlying technology will persist and slowly diffuse, much like the post‑dot‑com internet.