We didn’t see the crash coming. Not because we were blind, but because we were looking for the wrong shape. A single, monolithic implosion—a 2000-style dot-com wipeout that vaporizes everything in one fell swoop. That’s the narrative the bears have been selling. But what if the AI bubble isn’t a balloon about to pop, but a series of smaller, shifting manias? What if it’s rolling?
Dhaval Joshi, chief strategist at BCA Research, dropped a perspective that sliced through the noise. He calls it a “rolling bubble.” The AI sector isn’t heading for a binary collapse, he argues, but rather a sequence of localized overvaluations that migrate across layers—infrastructure, models, tools, applications. Each layer inflates, then deflates, while the next one catches the wind. The capital misallocation is real, but it’s episodic, not terminal.
I’ve audited enough narratives to know that sentiment is a shifting tide, not a solid ground. Back in 2018, I poured 40 hours into reverse-engineering Raptor Protocol’s smart contracts, convinced I’d found the next DeFi unicorn. I published a 3,000-word bullish thesis just days before a $2 million exploit. The lesson? The market doesn’t reward the loudest preacher—it rewards the one who reads the room. Joshi’s rolling bubble thesis is that kind of room-reading: it acknowledges the hype without buying the doomsday script.
Context: The Layered Architecture of a Narrative
AI, like crypto, is a stack. You’ve got the bare metal—GPUs, data centers, cloud infrastructure. Then the model layer—OpenAI, Anthropic, Meta’s open-source Llama. Above that, the tooling and middleware—LangChain, vector databases, agent frameworks. And finally, the application layer—Palantir, Copilot, AI tutors, and a thousand SaaS wrappers. Each layer has its own lifecycle, its own capital demands, its own fanbase.
From 2023 to 2024, we saw capital cascade through this stack. First, NVIDIA’s market cap went ballistic, crossing $3 trillion. Then the model companies raised billions at valuations that made no sense by traditional metrics. Now, the narrative is shifting to applications: “AI adoption is real, look at the revenue.” But Joshi sees the pattern: the bubble doesn’t burst—it rolls. One layer gets hot, overshoots, then cools as money flows to the next layer. The problem? The misallocation in each layer leaves a trail of stranded assets—overpriced GPU rental contracts, model companies that can’t justify their burn rates, and apps that sign up users but fail to retain them.
Core: How the Rolling Mechanism Works
Let’s get into the mechanics. Imagine a series of interconnected balloons. You inflate the first one—infrastructure. It grows, stretches, but doesn’t pop because the capital inflow keeps it aloft. Then, before it can burst, a new narrative emerges: “The real value is in the models.” So capital rotates to the second balloon. The first one starts to deflate, but slowly, because the overall market is still euphoric. The second balloon inflates, then the third, and so on. The aggregate air (total capital) keeps increasing, but the pressure is distributed across different containers.
This is precisely what Joshi means by “rolling bubble.” The risk isn’t a sudden all-at-once crash, but a gradual accumulation of mispricing across the stack. At any given moment, one layer looks overvalued, another looks undervalued. Active traders can profit by riding the wave. But the passive investor holding the entire index is exposed to sequential corrections that never fully recover—like a series of small leaks that eventually sink the ship.
Data from the past 18 months supports this. NVIDIA’s revenue growth decelerated in Q2 2024, yet AI application stocks like Palantir soared. The capital didn’t leave AI—it just moved up the stack. Meanwhile, the total AI investment (CAPEX from Microsoft, Google, Amazon, Meta) exceeded $200 billion in 2024 alone, far outpacing the revenue generated by the model layer. That’s capital misallocation. But because the bubble is rolling, the market doesn’t feel the pain all at once.
In the ledger’s silence, the true story whispers. On-chain data for AI-related tokens (like Render, Fetch.ai, or Akash) shows a similar pattern: valuation spikes, then corrections, then a new narrative takes over. The correlation isn’t perfect, but the structural similarity is unmistakable. Capital flows where the attention flows, and attention is a finite resource.
Contrarian: The Blind Spot of the Single-Bubble Narrative
The mainstream take is that AI is a bubble and it’s going to pop—hard. The bears point to the 1990s internet bubble, the 2008 housing crash, and the 2021 crypto bull run. But those were single, monolithic bubbles (or at least perceived as such). The internet bubble did have a rolling component: first semiconductors, then portals, then e-commerce, then fiber optics. The final crash happened when the last layer (fiber) collapsed, but by then the earlier layers had already corrected. The damage was spread out.
Joshi’s insight is that the same pattern is playing out today, but with a twist: AI’s layers are more interdependent. The infrastructure layer (GPU hardware) is a sunk cost—you can’t repurpose a data center for something else. The model layer is a race to the bottom in terms of pricing. The application layer is still searching for product-market fit. Each layer’s failure is cushioned by the next layer’s hype, but the cumulative risk is growing.
Here’s the contrarian angle for crypto: if the AI bubble is rolling, it means capital will eventually rotate out of AI and into the next narrative. Crypto is a natural candidate—it’s the other high-risk, high-narrative asset class. But this rotation won’t be a flood, it will be a trickle. Each AI layer’s deflation will release a small wave of capital, some of which will find its way into Bitcoin, Ethereum, or DeFi. The key is to watch the timing: the infrastructure layer is already cooling, which means we might see the first wave of rotation in the next 6–12 months.
Every bull run is a myth waiting to be debunked. The AI bull run is no different. But debunking doesn’t mean a sudden end—it means a slow unraveling, a series of smaller disappointments that shake out the weak hands. For crypto investors, this is an opportunity to be the liquidity provider when the AI narrative loses steam.
Takeaway: What the Rolling Bubble Means for Your Portfolio
So, where does this leave us? The rolling bubble thesis implies that the market will not experience a single, clean crash. Instead, it will bleed through a series of mini-crashes, each one hitting a different subsector. For the long-term holder, this is a risk: the overall market may grind higher for years, but the rotational corrections will erode buy-and-hold returns. For the active trader, this is a gift: the ability to rotate between AI layers and even into crypto at the right moment.
My advice: stop looking for the big bang. Start mapping the narrative layers. Watch the data—GPU rental prices, model API revenue growth, application churn rates. When a layer peaks, don’t fight it. Let the capital roll. And when the last layer (applications) shows signs of exhaustion, that’s when you want to be positioned in crypto, waiting for the next tide.
Yield is the bait, liquidity is the trap. The rolling bubble is the trap. But if you understand the mechanism, you can step through the trapdoor and come out the other side.