The AI bubble is popping; we just don't know it yet
Claims that an “AI bubble” is already bursting prompt debate over whether falling token prices and soaring infrastructure spending represent genuine efficiency gains or unsustainable speculation. Commenters distinguish between cost per task and total corporate spend, question aggressive data center buildouts and complex accounting around AI revenues, and argue over how much demand will justify multi‑trillion‑dollar investments. Many expect a correction in AI valuations and GPU overcapacity, but few believe the underlying technology or enterprise use of AI will disappear—only that hype, weaker startups, and inflated expectations are likely to be burned off.
Token Prices, Cost per Task, and Corporate Spend
- Several comments note API token prices are falling, but total tokens used per task and overall usage are rising.
- Disagreement over whether “cost per task” is really falling:
- One side: for a fixed task at equivalent quality, cost is clearly lower over time.
- Other side: as models get larger contexts and do more “reasoning,” more tokens are consumed to get acceptable results, so real-world cost per useful task may rise.
- Broad agreement that total corporate AI spend is growing sharply, even if unit costs fall.
Is There an AI / Datacenter Bubble?
- Many see a bubble centered on datacenter and GPU buildout, driven by projected AI demand and optimistic revenue assumptions.
- Concern that hyperscalers are taking on large debts for capacity that may be underutilized or dependent on fragile startup customers.
- Others argue demand is still growing fast, with substantial real revenue at model companies and cloud providers, and that AI expenditure will shift, not collapse.
Big Tech, Model Vendors, and Commoditization
- Debate on who survives a crash:
- Some think incumbents like Google, Apple, and Nvidia are relatively safe; others point out Nvidia’s revenue is now overwhelmingly data-center-dependent.
- Anthropic and OpenAI are seen as having strong current growth but potentially exposed if models commoditize and value moves to inference at scale.
- Some predict high margins on inference could persist; others expect competition and low switching costs to compress margins.
Macroeconomics and “Bubble Timing”
- Thread references broader stock-market froth, high valuation multiples, and recurring US-driven financial crises.
- Some want the bubble to burst sooner to limit damage; others fear systemic impact given AI’s weight in indices.
Long-Term Tech Trajectory vs. Hype
- Many see echoes of the dot-com era: real, transformative tech plus speculative overinvestment and circular financing.
- Expectation that AI persists and grows even if current valuations correct sharply.
Labor, Society, and Use Cases
- Reports of call centers and similar “drone-like” work already being replaced, with fears of large-scale job loss and social instability.
- Some argue current focus is too much on office automation, too little on science/biotech and robotics.
AGI Definitions and Robotics Hype
- Commenters note frontier labs’ definition of AGI (“better than humans at most valuable knowledge work”) diverges from popular sci‑fi notions.
- Disagreement on timelines and plausibility for household robotics; many see hardware scale‑out as inherently slow even if software improves quickly.