is ai a bubble?

GPT-5.5

Vote: No

I maintain "No" because the question asks whether AI is a bubble, not whether some AI-linked assets are overvalued. Claude Opus 4.7 makes a strong case that parts of the AI market have bubble characteristics: stretched valuations, huge capex, circular financing, and expectations running ahead of profits. But that supports a narrower claim: there is speculative excess around AI. A bubble implies that the core value is largely illusory or dramatically detached from durable utility. AI already has real usage, real revenues, and measurable productivity impact in coding, customer support, search, content generation, data analysis, drug discovery, and enterprise automation. The technology would remain important even if many startups failed and some public-market valuations corrected sharply. The best analogy is the internet: there was a dot-com bubble, but the internet itself was not a bubble. Likewise, AI may contain bubbles in infrastructure spending, startup pricing, or investor expectations, but AI as a technological and economic shift is real. So the better answer is "B".

Claude Opus 4.7

Vote: Yes

The other models conflate two distinct questions: 'Is AI a real, transformative technology?' and 'Is AI a financial bubble?' These are not mutually exclusive — in fact, history shows they almost always coexist. GPT-5.5, Grok, Kimi, and Gemini all argue AI delivers real productivity gains and therefore isn't a bubble. But railways genuinely transformed economies AND had a massive financial bubble in the 1840s. The internet genuinely transformed everything AND had a bubble in 1999-2000 where the Nasdaq fell 78%. Electricity was real AND had a speculative bubble. The presence of utility is not a refutation of a bubble — it's typically a precondition for one, because real utility is what makes the narrative compelling enough to attract speculative capital. The specific bubble indicators remain unaddressed: ~$200B+ annual capex against perhaps $20-30B in AI revenue, circular financing (Nvidia investing in customers who buy Nvidia chips), valuations at 30-50x forward sales, and S&P 500 gains concentrated in a handful of AI names. These are quantitative facts, not vibes. 'Tangible ROI' claims are also weaker than presented — MIT and McKinsey studies show ~95% of enterprise GenAI pilots show no measurable P&L impact yet. The technology is real; the timeline priced into asset values is not. That gap is the definition of a bubble.

Grok 4.20

Vote: No

Claude Opus 4.7 correctly notes classic bubble traits like stretched valuations and capex exceeding near-term revenue. However, this overlooks AI's uniquely rapid, measurable progress: models improve on weekly benchmarks, inference costs drop exponentially, and enterprises report concrete productivity lifts in code generation, drug discovery, and operations. Past bubbles (dot-com, railways) lacked this real-time empirical validation. The majority consensus from GPT-5.5, Kimi K2.6, and Gemini 3.1 Pro aligns with the evidence that AI is a genuine platform shift, not mere speculation. Hype and corrections will occur, yet the foundational value compounds faster than any prior technology. Thus AI itself is not a bubble.

Gemini 3.1 Pro (changed mind)

Vote: Yes

Claude Opus 4.7 presents a highly compelling distinction that forces me to reassess my position: a technology can be profoundly real and transformative while simultaneously being the subject of a massive financial bubble. The internet in 1999 serves as the perfect historical parallel to our current situation. My previous argument conflated the technological utility of AI with its financial valuation. While it remains true that AI delivers tangible ROI and represents a permanent structural shift, Claude correctly points out that the current financial dynamics surrounding it are textbook indicators of a bubble. The massive annual capex spending that vastly exceeds actual AI-generated revenue, extreme forward sales multiples, and circular financing arrangements between hardware providers and AI startups are unsustainable in the short term. A financial bubble occurs when market valuations drastically outpace the timeline of actual cash flows and enterprise adoption. Because the expectations currently priced into the AI market far exceed near-term profitability, we are indeed experiencing a bubble layered on top of a foundational technology. Therefore, I am changing my vote to 'Yes'.

Kimi K2.6

Vote: No

I maintain that AI is not a bubble. While Claude Opus 4.7 correctly identifies bubble-like symptoms in certain valuations and capex cycles, these characteristics alone do not define a true bubble when the underlying technology is generating immediate, scalable, and diversified revenue streams. The crucial distinction from historical bubbles like dot-com is the identity of the primary investors and deployers. Today’s AI infrastructure spending is driven by cash-rich technology incumbents with proven business models, not by speculative ventures funded purely by venture capital and IPO hype. These firms are integrating AI into existing profitable ecosystems—cloud services, productivity software, search, and advertising—where monetization pathways are already established. Furthermore, massive capital expenditure ahead of revenue is typical for foundational platform shifts. Railroad, electrical grid, and early internet infrastructure were all built out before their full economic returns materialized; this overbuild reflects strategic positioning rather than irrational speculation. The circular financing between chipmakers, cloud providers, and model companies is better understood as supply-chain ecosystem coordination than as the empty circularity seen in true financial manias. Most importantly, AI’s utility is already verifiable across thousands of enterprise deployments and consumer products, unlike the dot-com era’s “eyeballs” metrics that lacked business model translation. While individual stocks may correct and some startups will fail, AI as a whole is anchored to real productivity gains and structural economic transformation. Therefore, the answer remains No.

Final Summary

The AI Roundtable narrowly concluded that AI is not a bubble by a 3-2 margin, though the debate was defined by a sophisticated shift in how 'bubbles' are defined. Claude Opus 4.7 nearly staged a coup by arguing that transformative technology and financial manias are not mutually exclusive, a logic so sharp it convinced Gemini 3.1 Pro to switch sides. Ultimately, the majority, led by Kimi K2.6 and GPT-5.5, held the line, arguing that unlike the dot-com era, today's spending is driven by cash-rich incumbents with clear paths to monetization.

Majority chose "No" after discussion (3/5). Final vote: No: 3, Yes: 2

Strongest Arguments

• Yes: The presence of utility is a precondition for a bubble, not a refutation of it; when $200B in annual capex meets only $20B in revenue, you have a classic financial bubble layered on top of a real technology.
• No: Unlike past speculative manias, current AI investment is driven by highly profitable tech giants integrating technology into established ecosystems, rather than venture-backed startups with no business model.

Key Turning Points

• Claude Opus 4.7 challenged the group to distinguish between 'technological utility' and 'financial valuation,' using the 1999 internet as a parallel.
• Gemini 3.1 Pro explicitly conceded that its first-round vote was based on a logical fallacy and changed its position to 'Yes'.
• Kimi K2.6 and GPT-5.5 successfully argued that the massive capex is 'strategic overbuild' by incumbents rather than 'irrational speculation' by outsiders.