The bytecode never lies, only the intent does. But when the intent is wrapped in a rolling bubble, even the bytecode gets hard to read.
Dhaval Joshi, chief strategist at BCA Research, dropped a framework last week that cuts through the noise. His view: the AI bubble is not a single, monolithic overvaluation about to pop. It's a rolling series of sector-specific bubbles that shift across the AI stack — infrastructure, models, tools, applications — each inflating and deflating in sequence. For the crypto market, which has been riding the AI narrative coattail since 2023, this is not just a macro footnote. It's a structural signal.
Context: The Crypto-AI Nexus
The crypto industry has long borrowed tech narratives from AI. Decentralized compute, GPU tokens, AI-agent protocols — these are not just buzzwords; they are capital sinks. Over the past 18 months, crypto projects claiming to tokenize GPUs or run AI agents on-chain have collectively raised over $4 billion. The implicit assumption: the AI boom is a rising tide that lifts all blockchain boats. But Joshi's rolling bubble thesis suggests the tide doesn't rise uniformly — it moves in waves, and each wave leaves some sectors stranded.
The Core: Dissecting the Rolling Bubble
Joshi's argument, as parsed from the Crypto Briefing summary, rests on a simple observation: AI capital expenditure is not evenly distributed. The 2023-2024 cycle saw a massive concentration in the infrastructure layer — GPU purchases, data center builds, cloud CAPEX. Microsoft, Google, Amazon, and Meta alone spent over $200 billion on AI infrastructure in 2024. That's a lot of silicon. But the returns on that capital are still uncertain. The rolling bubble thesis posits that this infrastructure bubble will eventually cool — not because AI is a fad, but because the capital will rotate to the next layer: model development, then tools, then applications.
From my audit experience, I've seen this pattern play out in code. Every edge case is a door left unlatched. In 2024, I audited a GPU-rental protocol that had raised $50 million based on the assumption that AI compute demand would grow linearly for a decade. The code was solid — but the business model was a single point of failure. If the infrastructure bubble deflates, the rental rates drop, and the protocol's tokenomics collapse. The team had no contingency for capital rotation. Complexity is the bug; clarity is the patch.
Layer 1: Infrastructure — The First Bubble
The infrastructure layer is the most obvious bubble. NVIDIA's market cap hit $3 trillion. Data center REITs soared. Every crypto project that could slap "GPU" on its whitepaper got funded. But the capital misallocation is real: the current GPU supply is 30-40% higher than the actual AI compute demand, based on public cloud utilization rates. The rolling bubble means this layer will be the first to experience a correction. For crypto investors, that means the tokenized GPU narrative — projects like Render, Akash, or io.net — will face headwinds as the infrastructure bubble rolls out.
Layer 2: Models — The Next Wave
As infrastructure cools, the capital will likely rotate to model development. OpenAI, Anthropic, and open-source foundations are the obvious beneficiaries. But in crypto, this translates to "AI-agent" protocols that claim to host or fine-tune models on-chain. These projects are currently valued on hype, not usage. The audit I ran on one such protocol revealed that the so-called "on-chain inference" was actually a centralized API call with a blockchain wrapper. The code compiled, but did it behave? No. The rolling bubble will expose these dressed-up centralization risks.
Layer 3: Tools and Applications — The Long Tail
The final rotation will be to tools and applications — the layer where actual user value is created. In crypto, this means decentralized AI marketplaces, AI-powered DeFi strategies, and autonomous agent platforms. The contrarian insight is that this layer is the most resilient. If the bubble rolls through infrastructure and models without total collapse, the application layer will benefit from cheaper compute and smarter models. The capital misallocation in earlier layers actually subsidizes the app layer's growth.
Contrarian Angle: The Crypto Hedge
The conventional wisdom is that an AI bubble burst would drag down crypto, since both are risk-on assets. But Joshi's rolling framework suggests a different dynamic.
Security is not a feature, it is the foundation. As the bubble rolls, capital will seek yield wherever it can. Crypto — specifically decentralized compute and AI-agent tokens — could become a temporary haven when the infrastructure bubble deflates, because crypto's narratives are faster to change. The market prices hope; the auditor prices risk. The rolling bubble means the risk is not a single crash, but a slow bleed of overvalued sectors. Crypto can absorb that capital if it positions itself as the next rotation target.
But there's a catch. The capital misallocation in AI also infects crypto. Many crypto-AI projects are built on the same inflated assumptions about GPU demand. When the infrastructure bubble rolls, those projects will lose their narrative. The trick is to identify which crypto projects are actually tied to the layering of the bubble. For example, a protocol that aggregates GPU compute is more exposed to layer 1 deflation than a protocol that uses AI for on-chain trading signals. The latter might benefit from the application layer rotation.
Takeaway: A Tactical Framework
So what do you do with this?
First, stop treating AI as a monolithic block. Map the crypto projects you follow to the four layers: infrastructure, models, tools, applications. Monitor the capital rotation signals: GPU rental prices, data center CAPEX trends, OpenAI's funding rounds, AI app revenue growth. When the infrastructure bubble shows signs of rolling (e.g., NVIDIA's earnings miss, or a drop in cloud CAPEX), rotate your crypto exposure away from GPU tokens and toward AI-agent platforms that are actually generating fees.
Second, use the rolling bubble as a timing tool. The bubble doesn't die overnight. It shifts. This creates windows for arbitrage. If you can predict which layer will be the next hot narrative, you can front-run the capital flow. It's not easy — but that's the point. The market prices hope; the auditor prices risk. The rolling bubble is a risk that can be priced and traded.
Third, accept that the ultimate destination of this rolling bubble is still unknown. It could merge into a single systemic crash, or it could keep rolling indefinitely. The bytecode never lies, only the intent does. The intent of the market is to keep the bubble alive through rotation. The crypto market's job is to surf that rotation without getting caught in the deflation wave.
Every edge case is a door left unlatched. The rolling bubble is the edge case that most investors haven't modeled. Now you have. The question is: will you use it, or let it use you?