Trust is a vulnerability we audit, not a virtue. The same applies to the AI market’s valuation narrative. Over the past seven days, while crypto markets churned in narrow ranges, a quieter storm brewed in the equity space: Dhaval Joshi, chief strategist at BCA Research, released a thesis that reframes the AI euphoria not as a single supernova ready to implode, but as a rolling bubble—a sequence of local overheatings that migrate across the AI stack, from infrastructure to model to application layer. This isn’t another doom prophecy; it’s a structural dissection of capital misallocation that every crypto investor should understand. Because what happens in AI doesn’t stay in AI—the same mechanisms of sequencer centralization, oracle manipulation, and deferred liquidity shocks are already baked into our own DeFi and Layer2 architectures.
Context: The Hype Cycle in Two Dimensions
Mainstream headlines scream “AI bubble will burst” every time Nvidia dips 5%. Joshi’s counterpoint is more nuanced: the bubble isn’t monolithic. It rolls through four layers of the AI stack—compute (GPU clusters, data centers), foundation models (LLMs like GPT-4), tooling (MLOps, inference engines), and applications (Copilot, Palantir). Each layer gets its own mania, its own capital inflow, and eventually its own correction, while the next layer heats up. This pattern mirrors the 1990s internet bubble: semiconductor → portal → e-commerce → optical fiber. The difference? AI’s underlying asset—compute—has residual utility, like a GPU that can still mine or render, whereas dark fiber rotted.
From my seat as a crypto security audit partner, this rolling pattern is uncomfortably familiar. I’ve seen it in DeFi: the 2020 yield farming boom (application layer) drove liquidity into protocols, then the hype shifted to L2 scaling solutions (infrastructure layer), then to liquid staking derivatives (derivative layer). Each wave left corpses of unbacked tokens and unaudited vaults. Logic dissolves when code meets human greed—and the same happens when capital meets a new narrative without a proven ROI.
Core: The Rolling Bubble as a Capital Misallocation Machine
Joshi’s central argument is that a rolling bubble introduces capital misallocation across the AI stack. Money flows into the hottest layer, inflating its valuation beyond any reasonable discounted cash flow, while the next layer starves—until the narrative pivots and the cycle repeats. This creates a systemic fragility: when the infrastructure layer overheats (e.g., Nvidia at $3T), the model layer might be underpriced, but the application layer is already overpriced on promises of adoption. The market becomes a series of delayed corrections, not a single cathartic crash.
Let me ground this with a personal audit. In 2020, during DeFi Summer, I spent 200 hours modeling Compound and Aave’s interest rate curves in Python. The risk parameters were mathematically sound—until they weren’t. The models assumed rational actors and liquid oracles. When the capital flow rotated from lending to yield farming to governance tokens, the rate curves became arbitrary. Complexity is just laziness wearing a mask. The same is happening in AI: the compute layer’s CAPEX exceeds $200B across hyperscalers, but the ROI on those GPUs is still unproven. If the next wave—applications—fails to materialize, the infrastructure bubble will deflate, and the capital misallocation will be exposed as a systemic bug.
Consider the signals: H100 GPU spot prices have dropped 30% from their peak, while cloud providers continue to build data centers. This is the classic “overbuilding during a narrative peak” pattern. In crypto, we saw it with the 2021 L1 chain boom—Solana, Avalanche, and Terra all raised billions for infrastructure that was never fully utilized. Silence in the blockchain is louder than the hack—the quietest time is when capital is flowing into a narrative that hasn’t been stress-tested.
Contrarian: What the Bulls Got Right
Yet the bulls aren’t entirely wrong. The rolling bubble framework acknowledges that AI’s underlying technology is genuinely transformative. The compute layer, unlike the 2000 fiber optic glut, has a floor value: GPUs can be repurposed for rendering, scientific computing, or even crypto mining. The foundation models have demonstrated real productivity gains in coding, drafting, and customer service. The application layer, while nascent, has shown early signs of sticky revenue (e.g., Microsoft’s Copilot attached to Office 365).
Interoperability is the illusion of safety—but in this case, the interoperability of capital across layers might actually soften the landing. If the infrastructure bubble deflates, the capital doesn’t disappear; it migrates to the next layer. This is exactly what happened in crypto during the 2022 bear market: the L1/L2 infrastructure bubble burst, but the capital that survived flowed into DeFi blue chips (Uniswap, Aave) and then into real-world asset tokenization. The rolling bubble creates a “soft crash” pattern, where the market corrects in slices rather than all at once.
However, the bulls ignore the latency of the correction. Just because the bubble rolls doesn’t mean it never pops. The 1990s internet bubble eventually popped, and the recovery took years. The current AI capital misallocation is building a balance sheet of deferred risk. When the macro environment tightens—rising rates, geopolitical shocks—the rolling waves may synchronize, causing a multi-layer collapse. Every summer has a winter of truth.
Takeaway: The Bridge Was Never Built, Only Imagined
Joshi’s analysis is a gift to the crypto analyst. It forces us to ask: where is the capital misallocated in our own stack? Layer2 sequencers are still centralized, and the “decentralized sequencing” narrative has been a PowerPoint slide for two years. DeFi lending protocols rely on arbitrary interest rate models that ignore real market supply. AI oracles are becoming the next attack vector—my 2025 deep dive into off-chain computation models revealed a centralization risk in the node selection algorithm that could enable coordinated manipulation.
The rolling bubble is not a prediction of doom; it’s a call to audit the assumptions. The bridge was never built, only imagined—and the same applies to the AI valuation narrative. Trust the code, not the capital flow. The market will eventually reconcile the difference between the two. When it does, the transparent ledger of crypto will be the first to expose the true cost of mispricing.
