Do You Think AI Is a Bubble?" — Yes | 42 Chapters

AI Bubbles and Structural Shifts — Mo Jielin’s Sixth Podcast Conversation

This is Mo Jielin’s sixth time joining our podcast. Every few months, we review the latest shifts in the AI market. This time: one of the hottest topics — the AI bubble. The bubble is just the emotional surface; what matters more are the structural changes beneath it.

> Original transcript: ~17,000 words

> Edited version: ~6,600 words

---

Is There an AI Bubble Right Now?

Qu Kai: Is there an AI bubble?

Mo Jielin: Yes.

Qu Kai: That direct? No dissection? (laughs)

Mo Jielin: Haha — if we define a bubble as expectations exceeding reality, then yes.

• A bubble isn’t always bad — it can push the industry forward.
• It doesn’t have to burst right away.

---

Separating Value from Price

Qu Kai: We should separate value from price:

• Value perspective: Among AI practitioners, few talk about bubbles. Many believe intelligence levels are already high enough; things are “going great.”
• Price perspective: Break into China vs. U.S., primary vs. secondary markets.

Primary market: Valuation

• In China: healthy overall; valuations up from last year but still 10x lower than equivalent U.S. projects.
• In the U.S.: More obvious signs of overvaluation (e.g., Cursor at nearly $10B while losing money).

Secondary market: Market capitalization

• U.S. valuations visibly high; more volatility in public market pricing.

---

Price vs. Expectations

Mo Jielin: Past bubble discussions often point to P/E ratios not being excessive because EPS rises.

Qu Kai: Right — which is why some argue there’s no bubble.

Mo Jielin: But revenue alone misleads. Two key factors:

• Expectation Variance — People have radically different estimates for:
• Speed of cost reduction
• ROI of data centers
• Position in the AGI journey
• Structural Development — AI split from the start: China vs. U.S., hardware vs. software, etc.

---

Shifts in Narrative: Pre-training to RL

Qu Kai: Last year’s narrative:

• Stage 1: Pre-training scaling law → NVIDIA’s rise.
• Stage 2: Post-training & RL dominance (o1 release → peak narrative → NVIDIA drop).
• Stage 3: DeepSeek hype fading; RL scaling unclear.

---

Mo Jielin: Current bubble arguments are fragmented, but summarize to one core issue:

> Model companies’ ROI is questionable — huge investment, inadequate output.

---

The ROI Question

At the start of the year, DeepSeek’s breakthrough was framed as “spend little, achieve big results”.

Now discussions are more varied: model capability, application scenarios, commercial models, costs, China’s open sourcing.

---

Qu Kai: This is not new — discussed last year too.

Bubble proponents often reuse old points. No clear leading “bubble flag waver.”

---

Investment vs. Returns

Mo Jielin: Recent chain of events:

• Meta talent poaching
• OpenAI & xAI investing heavily in data centers
• NVIDIA backing model companies

Common thread: Investment keeps growing ⇒ returns flattening ⇒ ROI anxiety.

Add macro volatility ⇒ bubble talk intensifies.

---

Market Sentiment and Structural Change

Qu Kai: U.S. stocks & NVIDIA down. Is it AI-driven or macro?

Mo Jielin: Not purely AI. Risk appetite lower, but impact on practitioners or Google buyers minimal.

Qu Kai: The market is “looking for a bubble.”

Mo Jielin: Probably just trimming over-optimism.

---

From AGI Dreams to ROI Calculations

Mo Jielin: Many see model capabilities as good enough.

The key question:

• Has large-scale pretraining ended?
• Are we moving to: apply existing models + focus on cost reduction?

Qu Kai: The industry shift toward applications and monetization is visible.

Platforms like AiToEarn官网 exemplify open ecosystems — creators leveraging AI for content generation, distribution, analytics, and model rankings.

---

Can We Judge Cycles?

Mo Jielin: Cycle judgment is inherently difficult:

• Few global decision-makers determine AI’s path.
• AI is capital-intensive; affected by macroeconomics.

Example: Apple avoids over-investment yet thrives.

Structural change: ROI mattered little before; now it’s central.

---

Qu Kai: Scaling Law justified large spend if it neared AGI — but RL shows no clear scaling curve.

Mo Jielin: Disagree — Scaling Law is intact but immeasurable now. Focus needed on cost, infrastructure, agents, and context.

---

Hardware vs. Software Outcomes

Mo Jielin: If training plateaus, some U.S. firms have bubbles. But NVIDIA’s stock held — market expects GPUs to remain essential.

Qu Kai: NVIDIA profited most; no standout software yet.

---

Mo Jielin: Bubble judgment is complex — price ≠ expectation. Emotion often drives “bubble” calls. Historically, most perceived bubbles prove false.

---

Without Breakthroughs, No “Belief Recharge”

Qu Kai: AI hype comes in pulses — recent lack of breakthroughs reduces excitement.

Mo Jielin: Future likely to diverge sharply. Winners will emerge across fields — critical to identify early.

---

Winners and Cycles

Mo Jielin: Beneath emotions lie two key questions:

• Cycle: New or ended?
• Winners: Who will dominate?

Recent underestimation examples: Chinese open-source models (2023), Cambricon (2024). Current skepticism: OpenAI; real profits: NVIDIA.

---

Channels Often Win Most

Qu Kai: Industries’ biggest earners often control the “channels” — e.g., ByteDance, Focus Media, malls.

Mo Jielin: Could be misjudging OpenAI — emotional conclusions often contradict long-term.

---

Secondary Market Retail Dominance

Mo Jielin: Retail investors dominate, shifting emphasis to sentiment/narrative over long-term trends.

Practitioners often provide denser, ground-level insights.

Key Q3 2025 observations:

• NVIDIA’s $100B OpenAI investment
• OpenAI’s AMD & Broadcom deals
• Google’s TPU investment
• ⇒ Signals an upstream compute power reshuffle.
• Ignored player in U.S. coverage: ByteDance — with strong talent reserves.

---

Deployment Cycle and Bubble Dynamics

Qu Kai: We’re between two cycles:

• Market moving from AGI “dream caps” → deployed-app P/E ratios.
• Bubbles inevitable in fast-growing fields; bursting nourishes survivors.
• Example: PayPal emerged from a bubble by maintaining strong cash flow.

---

Practical Monetization Tools

In this deployment phase, tools like AiToEarn help creators & teams:

• Generate AI content
• Publish cross-platform
• Analyze impact
• Track model rankings

---

Conclusion: Winners Will Emerge

Mo Jielin: Emotions matter short-term; cycles & structure matter long-term.

Every business cycle produces winners — focus on identifying and adapting to them.

---

42 Chapters

Thinking about the essence of things