We didn't see it coming. We never do.
Two weeks ago, I was sitting in a Zurich coworking space, debugging a cross-chain liquidity pool for a DeFi protocol that shall remain nameless. My phone buzzed with a Bloomberg alert: Dhaval Joshi, BCA Research's chief strategist, had just published a note warning that the AI bubble isn't a single balloon about to pop — it's a rolling series of mini-bubbles, each inflating and deflating across different layers of the tech stack. I stopped mid-debug. Because if Joshi is right, the implications for crypto are not just interesting — they're existential.
Here's the context. Joshi's argument is deceptively simple: AI valuations aren't monolithic. The market is rotating capital from infrastructure (Nvidia, data centers) to models (OpenAI, Anthropic) to applications (Palantir, etc.), and this rotation creates a "rolling bubble" — each layer gets its moment of euphoria, then cools, while the next layer heats up. The risk isn't a single crash, but a series of capital misallocations that compound over time. Sound familiar? It should. It's the exact same structure we lived through in crypto from 2017 to 2024.
I've seen this movie before. In 2017, I launched a white-label ICO called "ZurichChain" — a hybrid PoW/PoS consensus layer — and raised $4.2 million in 48 hours. The narrative was "decentralized sovereignty." The capital rotated from Bitcoin to ICOs to infrastructure tokens, then collapsed. In 2020, I audited AeroSwap's bonding curve and found a reentrancy vulnerability that would have drained $15 million. The rotation was from DeFi to yield farming to NFTs. In 2021, I ran a Zurich workshop on NFT provenance, connecting ERC-721 to digital identity. The rotation continued. In 2022, I joined LayerZero Labs and built cross-chain bridges in 72 hours. The rotation accelerated. In 2024, I helped a Swiss private bank design a decentralized custody solution for ETF-linked tokens. The rotation now includes institutional liquidity.
What I learned is this: rolling bubbles are the natural state of any narrative-driven, capital-intensive tech cycle. AI is just the latest iteration. And crypto is both a mirror and a potential beneficiary.
Let's get technical. The AI stack has four layers: infrastructure (chips, compute), models (foundation LLMs), tools (frameworks, middleware), and applications (vertical solutions). Joshi's "rolling bubble" implies that capital flows sequentially through these layers, driven by peak hype and diminishing returns. The key insight is that capital misallocation at one layer gets temporarily masked by the next layer's hype. For example, in 2023-2024, Nvidia's market cap exploded to over $3 trillion — that's infrastructure layer euphoria. But the actual ROI of those GPUs is still unproven at scale. Joshi warns that the moment the application layer fails to generate returns commensurate with the infrastructure spend, the bubble will roll back, causing a correction.
Now, overlay crypto's history. In 2017, the infrastructure layer (Ethereum, Bitcoin) was the first bubble. Then it rolled to ICOs (application layer for fundraising). Then to DeFi in 2020 (a new application layer). Then to NFTs in 2021 (digital identity). Then to cross-chain in 2022-2023 (interoperability infrastructure). Each time, the bubble didn't burst all at once — it rotated. The capital that rotated out of DeFi in 2021 didn't disappear; it went into NFTs. The capital that rotated out of NFTs in 2022 didn't vanish; it went into Layer 2s and intent-based protocols. The pattern is identical.
Here's the contrarian angle: Most crypto analysts are worried that AI will suck all the liquidity out of crypto. They see Nvidia's P/E ratio and panic. But Joshi's framework suggests the opposite. If AI is a rolling bubble, then capital will eventually roll out of the AI infrastructure layer (which is currently overheated) and into the next layer — which could very well be crypto-native AI infrastructure, or even decentralized compute. Because the next logical layer after AI models is decentralized inference and data sovereignty. And that's where crypto's edge lies.
I've been testing this hypothesis. In my recent work with institutional clients, they're asking two questions: (1) How do we hedge against AI concentration risk? (2) How do we participate in the next wave of AI without buying Nvidia at 50x earnings? The answer, in both cases, is decentralized compute networks. Projects like Akash, Render, or even emerging zk-proof-based AI verifiers are positioned to capture the capital that rolls out of centralized AI infrastructure. The capital misallocation Joshi warns about is exactly the opportunity for crypto to absorb overflow capital — but only if we build the bridges now.
But let's be honest about the risks. If the AI bubble rolls back before crypto's infrastructure is ready, we could see a double correction. Crypto is still a high-beta asset. If AI infrastructure corrects 30%, crypto could correct 50% or more. I saw this in 2022: when the entire tech stack corrected, crypto fell harder because it has less intrinsic yield. The rolling bubble model doesn't eliminate systemic risk; it just postpones and localizes it. The hidden danger is that the capital misallocation in AI could be so massive that when it finally reverses, it triggers a liquidity crisis that spills into every risk asset, including crypto. Joshi's note doesn't address this directly, but the math is straightforward: if $200 billion in AI infrastructure CAPEX is built on projected returns that don't materialize, the subsequent deleveraging will be brutal.
So what do we do? I'm not a trader; I'm a protocol PM. But I've learned to read the signals. The signal I'm watching now is the GPU rental spot price. H100s on the spot market have dropped 20% in the last quarter. That's a leading indicator that infrastructure demand is softening. If that trend continues, the AI bubble will roll from infrastructure to models — and the models layer (OpenAI, Anthropic) will face valuation pressure. That's when the capital starts looking for alternatives. That's our window.
My takeaway is this: The AI bubble isn't your enemy. It's a rolling wave. Crypto's job is to build the surfboard. We need decentralized compute, verifiable inference, and data sovereignty protocols ready to absorb the capital that will inevitably rotate out of centralized AI. The next 12 months are critical. If we can demonstrate that decentralized AI infrastructure delivers real utility — and I mean real, auditable, on-chain utility — then the capital will come. If not, we'll be left watching the bubble roll into the next hype cycle without us.
We didn't build for the last bull market. We build for the next rotation. And I'm betting that the next rotation points straight at crypto.