The market is holding its breath for an AI collapse. But the real story is subtler – and more dangerous for crypto.
Dhaval Joshi, chief strategist at BCA Research, just dropped a warning that cuts through the hype: AI isn’t a single bubble about to pop. It’s a rolling bubble – a sequence of mini-booms that shift across the tech stack, from chips to models to apps. Each wave inflates, then deflates locally, while the next sector catches fire.
Why should crypto care? Because capital doesn’t stay in silos. When AI’s infrastructure layer (GPUs, data centers) starts to cool, that liquidity doesn’t evaporate – it rotates. And crypto has been the beneficiary of every major tech rotation since 2020. If you think the AI bubble bursting is bad for crypto, you’re reading the map wrong.
Context: Why “Rolling” Matters More Than “Popping”
For the past 18 months, the narrative has been binary: AI is either a world-changing revolution or a tulip mania. Joshi’s framework breaks that binary. He argues that the dot-com era wasn’t a single implosion either – it was a cascade of sector rotations: semiconductors in 1998, portals in 1999, e-commerce in 2000. Each mini-bubble inflated, then rolled into the next.
Today, AI’s stacked layers mirror that pattern. Infrastructure (Nvidia, cloud CAPEX) peaked in 2023. Model layers (OpenAI, Anthropic) sucked in capital through 2024. Now the spotlight is shifting to applications and agents. The risk isn’t a sudden crash – it’s that capital misallocation builds up in each layer, and when the next layer fails to deliver, the whole stack wobbles.
From my 7x24 surveillance seat, I’ve seen this before. In DeFi Summer 2020, liquidity rotated from DEXs to lending protocols to yield aggregators – each wave inflated, then dumped. The same mechanical pattern is playing out in AI, but at 10x the scale.
Core: The Technical Anatomy of Capital Rotation
Let’s get granular. The rolling bubble has four stages, each with a distinct crypto parallel:
- Infrastructure (2023-2024): Nvidia’s valuation tripled on AI GPU demand. Code is law, but vigilance is the price of entry. The parallel? Ethereum’s L1 gas fees spiked during NFT mania – infrastructure bottlenecks create temporary value, but the real money is in the applications built on top.
- Models (2024-2025): OpenAI’s $150B valuation vs. $3.7B revenue. That’s a 40x multiple – higher than any crypto L1 token. Models are the new “base layer” – but unlike Ethereum, they have no token to capture value. When the model bubble deflates, the capital will flow to the next layer that offers something more tangible.
- Tools & Agents (2025-?): This is the stage we’re entering. AI agents, autonomous workflows, verifiable compute. In crypto terms, this is the “L2 scaling” moment – zk-rollups, AI copilots, and decentralized inference networks. Modularity isn’t the freedom to scale – it’s the freedom to fragment. Every new AI agent platform is a new chain, with its own token, its own liquidity. The same fragmentation we saw in the L2 wars is now happening in AI.
- Applications (2026+): The final stage. Real revenue, real users. If stage 3 fails to deliver, the entire rolling bubble could stall. But if it succeeds, the capital that rotated out of infrastructure will pour into apps – and crypto’s app layer (DeFi, gaming, payments) is the most natural landing pad.
Contrarian: The Blind Spot Everyone Misses
Here’s the counter-intuitive take: The rolling bubble is actually good for crypto – in the short term. Every time an AI sector overheats and wobbles, the capital that leaves looks for a new narrative. Crypto is the default alternative for risk-seeking, tech-native capital. We saw it in 2022 when the macro selloff hit both AI and crypto, but then crypto recovered faster because its narrative (decentralization) was distinct from AI’s (centralized compute).
But there’s a catch. The same capital rotation that pumps crypto can also poison it. If AI’s rolling bubble finally collapses into a synchronized crash (because all layers are overvalued simultaneously), the contagion will hit crypto hard. The correlation between BTC and tech stocks has been rising – 0.6 in 2024 vs. 0.3 in 2022. A multi-sector AI rout would drag crypto down.
Takeaway: What to Watch Next
I’m not betting on a crash. I’m betting on rotation. The next 12 months will see capital shift from AI model tokens (if they exist) to AI infrastructure tokens (like Render, Akash, or new L1s focused on verifiable compute). Then to applications. The key signal: watch the GPU rental spot price. If it drops below $1.50/hour for H100s, the infrastructure layer is cooling – and crypto will be the next stop.
When the AI bubble rolls, will crypto catch the capital – or the contagion?
