The market is not staring at a single, monolithic AI bubble about to pop. It is watching a rotating sequence of smaller, sector-specific bubbles, each feeding off the previous one's debris. That is the core thesis of Dhaval Joshi, chief strategist at BCA Research, as reported by Crypto Briefing. And if you read carefully, the implications for both AI and crypto are far more nuanced than a simple 'buy the dip' or 'sell everything'.
I have spent the last six years dissecting smart contracts and zero-knowledge proofs. My INTP brain loves structural models. So when I see a framework like 'rolling bubbles,' I immediately map it to the four-layer stack of AI: infrastructure (chips, data centers), models (foundation LLMs), tooling (frameworks, middleware), and applications (vertical solutions). Joshi's argument is that capital flows are not evenly distributed across this stack. Instead, they swarm one layer, inflate valuations, then migrate to the next, leaving the previous layer to cool down.
This is not just a financial observation. It is a game-theoretic equilibrium. Capital seeks the highest marginal return on narrative, not on actual productivity. The market is a machine that optimizes for attention, not for truth. Math doesn't lie, but the distribution of capital does.
Context: The Single-Bubble Fallacy
From 2022 to 2024, the prevailing narrative was that AI was a classic bubble akin to the dot-com era—one giant balloon that would eventually deflate, taking down Nvidia, OpenAI, and every AI startup with it. But Joshi injects a contrarian twist: the bubble is not a single entity; it is a 'rolling bubble.' The same capital that pumped Nvidia in 2023 rotated into model providers in early 2024, and is now eyeing application-layer companies. The dot-com bust itself was not a single crash either—it was a sequence of sub-bubbles: semiconductors, portals, e-commerce, and fiber optics. The AI bubble is following a similar path, but with a crucial difference: the underlying assets (compute) have significant residual value.
Privacy is a protocol, not a policy. And in this context, the protocol of capital allocation is not transparent. We see only the surface-level price movements, not the underlying flows of misallocation.
Core: The Structural Anatomy of the Rolling Bubble
Let me break down the layers using the same rigorous lens I apply to smart contract audits.
Layer 1: Infrastructure (Chips and Data Centers)
This is the first bubble. Nvidia's market cap surged past $3 trillion, and cloud providers spent over $200 billion on capex in 2024 alone. The capital misallocation here is obvious: hyperscalers are building GPU clusters that, at current utilization rates, will take years to become profitable. But here's the twist. Unlike the fiber-optic cables of 2000 that were largely unused, compute has a durable use value. Even if the bubble deflates, those GPUs will still mine Bitcoin, run inference, or be rented out. The infrastructure bubble is a 'hard asset' bubble, not a 'paper asset' bubble. This makes it stickier.
Layer 2: Models (Foundation LLMs)
By mid-2024, capital rotated from infrastructure to model providers. OpenAI, Anthropic, and Mistral raised billions at valuations that implied future revenue growth of 10x or more. But the unit economics of selling API tokens are brutal. The cost of inference is dropping exponentially, while competition is commoditizing the output. The capital misallocation here is that the market is pricing models as if they have network effects, but they don't—they have data moats, which are rapidly eroding as synthetic data and open-source models catch up.
Layer 3: Tooling and Frameworks
This layer is the quietest bubble. Companies like LangChain, Weights & Biases, and Hugging Face have raised large rounds, but their revenue is often tied to the success of the layers above. Tokenomics of tooling is fragile: if the model layer crashes, tooling loses its raison d'être.
Layer 4: Applications
This is currently the most hyped layer. Palantir, Salesforce, and a host of AI-native startups are being valued on 'AI tailwinds' rather than actual retention metrics. The capital misallocation here is the most dangerous because it is the least visible. When a company claims 'AI-driven growth,' investors rarely verify that the customer churn rate is actually lower than the pre-AI era. I've seen this pattern before in DeFi protocols that claimed 'TVL growth' but had zero retention.
Contrarian: The Blind Spots and the Final Act
The common belief is that a rolling bubble is safer than a single bubble because it deflates gradually. I disagree. The rolling mechanism actually delays the inevitable correction, allowing misallocation to accumulate across multiple layers. When the music stops, the correction will be simultaneous across all layers, not sequential. This is the 'resonance risk.' If infrastructure, models, tooling, and applications all crash together, the contagion will be far worse than a single bubble pop.
Moreover, the rolling bubble assumes that each new layer can generate enough narrative to attract capital from the previous one. But what if the narrative supply runs out? What if the next layer (e.g., AI agents) fails to deliver the expected return? Then the bubble collapses back into the previous layer, creating a 'double bubble' that is even more fragile.
Another blind spot: the role of large tech companies. Microsoft, Google, and Meta can absorb the losses in one layer because they have diversified revenue. But the startups that ride the bubble in a single layer have no such cushion. They will be the victims of the rotation. The capital misallocation is not just about money; it is about talent. The best engineers will be sucked into the next hot layer, leaving the previous layer without the human capital to fix its fundamentals.
Takeaway: What This Means for Crypto
Crypto Briefing reported this thesis because the AI bubble directly affects crypto markets. When capital rotates out of AI infrastructure, some of it will flow into crypto as a 'risk-on' alternative. But the timeline is not linear. The rotation will happen in fits and starts, driven by macroeconomic triggers like interest rate changes or regulatory shocks. The key signal to watch is the real yield on 10-year U.S. Treasuries. If that rises above 2.5%, the rolling bubble will accelerate its rotation, and crypto will be a beneficiary—but only temporarily until the next liquidity crunch.
For crypto builders, the lesson is this: do not confuse AI hype with your own protocol's value. The DAOs that claim to be 'AI-powered' are just compliance shields. The real value lies in the infrastructure that is immune to the rotation—things like decentralized compute networks, zero-knowledge rollups, and on-chain data markets. Math doesn't, and neither does the market. Trust nothing. Verify everything. Again.
The rolling AI bubble is not a crash in slow motion. It is a structural reallocation of capital that will leave behind both winners and losers. The winners will be those who understand that each layer has a different time constant and a different risk profile. The losers will be those who chase the narrative without auditing the code.