I watched the silence break the noise of 2021. That year, every conversation in crypto was a crescendo—NFTs, DeFi, scaling. The silence came when the music stopped. Last week, I saw that same silence creep into an AI funding round. A founder in Bangalore, who raised $50M for an AI agent platform in 2024, told me his investors are now asking for proof of ROI, not just TAM. The pitch deck has shifted from 'revolution' to 'rotation.' That shift is the core of Dhaval Joshi's thesis: the AI bubble isn't a single balloon waiting to burst—it's a series of smaller, rolling bubbles, each inflating and deflating across different layers of the tech stack.
Joshi, a chief strategist at BCA Research, has been making waves with a counter-narrative that sits between the permabulls and the doomsayers. He argues that the market is not facing a monolithic AI crash, but a 'rolling bubble' where capital misallocation rotates from infrastructure to models to applications. My own experience tracking sentiment during the 2024 ETF era—where I analyzed 200 key Twitter accounts and saw language shift from 'store of value' to 'institutional yield play'—taught me that narratives are the real drivers of market cycles. The AI narrative is now undergoing a similar transition.
Context: The Four Layers of the AI Stack To understand the rolling bubble, you have to map the AI stack. There are four layers: infrastructure (chips, data centers, energy), foundation models (LLMs like GPT-4, Claude), tooling (frameworks, middleware, verification), and applications (enterprise solutions, AI agents, creative tools). Each layer has its own valuation cycle. In 2023, the bubble was in infrastructure—NVIDIA’s market cap hitting $3 trillion, hyperscalers spending billions on GPUs. In early 2024, it rotated to models, with OpenAI’s valuation soaring to $300 billion. Now, we are seeing the bubble shift to applications, where companies like Palantir and ServiceNow are riding the wave. But the catch, as Joshi warns, is capital misallocation: money is flowing into layers based on narrative heat, not on actual demand or return on invested capital.
Core: The Mechanism of the Rolling Bubble Based on my research into social listening data across AI and crypto markets, I can see the pattern clearly. The narrative shifted from 'we need more compute' to 'the model is the product' to 'the app is the killer use case.' Each shift creates a local boom. The infrastructure layer boomed when GPU scarcity was the story. Then models boomed when GPT-4 launched. Now, applications are booming because everyone wants to show real-world traction. But the underlying data tells a different story. According to a recent analysis of 100 AI startups, only 15% have achieved positive unit economics. The rest are burning cash on customer acquisition, hoping the next narrative rotation will save them.
I spent three weeks in a cabin after the LUNA collapse, analyzing the psychological breakdown of that community. The same fragility is present in AI today. The rolling bubble structure masks the cracks—each layer’s correction is temporary because capital rotates to the next big story. But the underlying risk is that the rotation itself creates a systemic dependency. If the next narrative fails to materialize—say, if AI agents don't deliver on their promise—all layers could correct simultaneously. This is the 'silent crash' scenario Joshi alludes to. The ETF didn't eliminate volatility; it just delayed it.
Contrarian: The Hidden Resilience of Rolling Bubbles Here is the counterintuitive angle: a rolling bubble might actually be the healthiest way for a transformative technology to mature. Unlike the 2000s dot-com bubble, where capital was poured into fiber optics that sat unused, AI infrastructure—GPUs, data centers—has long-term utility. Even if the application layer crashes, the compute capacity remains. I spoke with a data center operator in Chennai who confirmed that his facility’s utilization rate is still below 70%, but he expects it to climb as AI inference demand grows. The rolling bubble allows for incremental capital deployment, preventing the all-at-once waste of the 1990s.
But the blind spot is timing. Joshi’s thesis assumes that the rotation can continue indefinitely. History doesn't guarantee that. The narrative shifted from 'scaling laws will continue forever' to 'we need more data' to 'we need synthetic data.' Each shift is a smaller narrative spark. The real risk is that the market runs out of new stories to tell. When that happens, the rolling bubble becomes a single, fragile sphere. The silence I heard in Bangalore is the sound of that narrative fuel running low.
Takeaway: What Comes Next? The next narrative will likely be about 'AI verifiability' or 'decentralized inference'—the intersection of crypto and AI. I see early signals in the rise of zk-proofs for model integrity and MPC for AI identity. The regulatory endpoint is clear: every AI output will need an on-chain attestation of origin. The rolling bubble will rotate toward that story. But the question remains: can the market sustain this rotation long enough for the fundamentals to catch up? Or will the silence eventually break the noise?
I don't have the answer. But I know that the silence in Bangalore was louder than any pitch deck. And that silence is the signal to watch.