AI's Rolling Bubble: A Structural Truth or a Delayed Crash?

The data doesn't lie. AI capital expenditure in 2024 hit $200 billion across the Big Four hyperscalers. Yet the narrative remains singular: a monolithic bubble waiting to burst. Dhaval Joshi of BCA Research offers a different framework. He calls it a rolling bubble. Not one explosion, but a series of localized overvaluations that migrate across the AI stack. This is not a simple warning. It is a structural diagnosis that demands a technical audit. I have spent the last decade tracing the gap between code and capital. From auditing smart contracts in 2017 to modeling AI-agent tokenomics in 2026, I have learned one thing: volume lies. Liquidity speaks. Joshi's thesis aligns with what I observed in the 2020 DeFi summer—yield farming APYs were a narrative, not a revenue model. The same pattern now grips AI. The difference is the scale of capital misallocation. Let me lay out the architecture. The AI stack has four layers: infrastructure (GPUs, data centers), models (foundation APIs), tools (development frameworks, middleware), and applications (industry solutions). Joshi argues that the bubble does not inflate uniformly. It rotates. In 2023, infrastructure was the epicenter. Nvidia's market cap crossed $3 trillion. In 2024, the focus shifted to model providers—OpenAI, Anthropic, and their equivalents. The next rotation, if the pattern holds, will target applications. This is not a random wave. It is a capital flow mechanism driven by narrative cycles. From my experience managing a $2 million DeFi portfolio, I know that stability is a narrative itself. During the bZx hack, my rigid exit rules saved 95% of capital. The same principle applies here. The risk is not that the AI bubble bursts today. It is that capital chases the hottest layer, leaving the previous one underfunded. This creates a structural vulnerability: when the next rotation fails to sustain momentum, the entire stack may collapse simultaneously. Joshi's framework implies that the crash is postponed, not cancelled. Code is law, until it isn't. The infrastructure layer, for instance, has a unique buffer. GPU clusters and data centers have long-term utility. Even if the 2024 capex overbuilds, the assets will not rot like fiber-optic cables in 2000. They can be repurposed for future workloads. This is what I call the 'economic viability buffer.' But it has limits. Energy constraints, cooling costs, and the physical limits of Moore's law put a ceiling on the buffer. The market often ignores these constraints. I saw the same in 2017 when my ICO audit flagged integer overflow vulnerabilities. The committee rejected my report. They prioritized hype over code. The same mistake is being made today. Now, the contrarian angle. What if the rolling bubble is not a series of local shocks but a disguised supercycle? Consider the 1990s internet bubble. It rolled through semiconductors, portals, e-commerce, and optical networking. Each rollover was a correction, but the aggregate was a systemic crash. However, the internet infrastructure built during that bubble enabled the next decade of growth. The same could happen with AI. The capital misallocation today may be wasteful in the short term, but it seeds the compute capacity for the next paradigm. The real question is whether the timeline of revenue generation aligns with the timeline of capital patience. Based on my 2026 audit of Render Network, I found a critical flaw in its tokenomics: it did not account for agent transaction fees. The narrative of 'AI agents on blockchain' was hot, but the economic model was fragile. When the market corrected, the narrative collapsed. The same will happen to any AI layer that relies on token emissions without real user retention. The signal to watch is not market cap. It is user retention rates and revenue per compute unit. I learned this during the NFT ice age of 2022. I accumulated Axie Infinity when floor prices were low because user retention metrics were stable. The same principle applies to AI stocks. Joshi's framework is valuable, but it has a blind spot. It assumes that the rotation is orderly. In reality, capital flows are driven by FOMO and regulatory shifts. The 2024 Bitcoin ETF approval was a regulatory clarity trigger that reshaped capital allocation. A similar regulatory event—like a SEC ruling on AI model liability—could force a simultaneous devaluation across all layers. The rolling bubble would then become a cascading waterfall. This is why I include a 'Regulatory Risk Assessment' in every report. The legal landscape is the ultimate narrative driver. For crypto investors, this rolling bubble presents both opportunity and trap. The AI-crypto narrative is already leaking. AI agents executing blockchain transactions is a hot topic. But the economic viability of these tokens is questionable. My analysis of the AI-agent crypto integration framework in 2026 showed that most projects lack sustainable token models. They are riding the AI narrative wave, not building real value. The smart capital will differentiate between protocol-generated revenue and token emission incentives. The rest will be left holding the bag. Takeaway: The next narrative shift is not from AI to crypto. It is from hype to revenue. Joshi's rolling bubble is a call to audit the fundamentals. I will be watching the price of H100 spot rentals as a real-time indicator. If that drops while AI stocks rise, the bubble is rolling away from infrastructure. That is the signal to rotate. Data doesn't lie. The narrative does.