Alert. Dhaval Joshi, chief strategist at BCA Research, dropped a signal that most media missed. He didn't say AI is a bubble destined to pop. He said it's a rolling bubble — a sequence of localized overextensions shifting across the AI stack. Capital misallocation is the engine. And the crypto market is listening.
This isn't another "AI will crash" headline. It's a structural thesis. One that changes how you position capital.
Context: Who Is Dhaval Joshi?
BCA Research has a 50-year track record serving institutional allocators. Joshi isn't a retail pundit. His views are built on macro liquidity cycles, asset rotation, and historical precedent. When he warns about capital misallocation in AI, he's not predicting a 2000-style collapse. He's mapping a multi-phase unwind that may never hit all sectors simultaneously.
His core claim: AI's valuation bubble is not a single overvalued asset class. It's a series of micro-bubbles rotating through infrastructure → model layer → tools → applications. Each phase overheats, corrects locally, then passes the torch. The total system stays inflated, but the risk is distributed across time and sectors.
Core: The Rolling Bubble Mechanics — A Technical Dissection
I've spent 12 years dissecting blockchain narratives. This pattern is familiar. In crypto, we saw the ICO bubble (2017) roll from Ethereum infrastructure to dApps to exchange tokens. Each roll created new millionaires and new bagholders. The same mechanics apply to AI.
Layer 1: Infrastructure (GPUs, data centers). Joshi's capital misallocation starts here. 2023-2024 saw hyperscalers (Microsoft, Google, Amazon, Meta) commit $200B+ in combined CAPEX. Nvidia's market cap crossed $3T. But the ROI on that compute is still unproven. H100 spot prices are already softening. If AI application demand doesn't catch up, infrastructure capex becomes stranded. That's a local bubble — but not a systemic one, because compute retains reuse value.
Layer 2: Foundation Models. OpenAI, Anthropic, and Mistral raised billions at valuations that assume future dominance. But the model layer is commoditizing. Open-source alternatives (Llama, DeepSeek) compress margins. Joshi's rolling bubble suggests this layer will be next to correct — not because AI is failing, but because capital will rotate to the next narrative.
Layer 3: Tools & Middleware. LangChain, vector databases, MLOps platforms. These are the picks-and-shovels of the AI era. Valuations are high, but revenue multiples are still speculative. Capital rotation from models to tools could inflate this layer temporarily before the next correction.
Layer 4: Applications. The final destination. If AI agents, copilots, and vertical SaaS actually deliver ROI, this layer could sustain a longer bubble. But Joshi's warning is that capital will chase the hot story — not the proven business. The risk is that applications never achieve the scale needed to justify the cumulative CAPEX.
Contrarian Angle: Why the Rolling Bubble Favors Crypto
Mainstream media frames the AI bubble as a threat to all risk assets. I see the opposite. A rolling bubble means the crash is delayed, segmented, and partially hedged. Here's the unreported angle:
Crypto as a beneficiary of capital rotation. When AI infrastructure capex peaks, excess liquidity can flow into alternative narratives — like Bitcoin as a compute asset, or decentralized GPU networks (Render, Akash). The same capital that over-invested in Nvidia GPUs can rotate into tokenized compute. I've seen this play out during the 2020 DeFi Summer: liquidity rotated from ICOs to yield farms to NFT floor prices. The pattern repeats.
Rolling bubbles reduce systemic risk. A single massive bubble popping would crater all correlated assets. A rolling bubble allows partial corrections, giving investors time to rebalance. Crypto, being uncorrelated to AI fundamentals in the short term, can act as a shock absorber. When AI model layer corrects, Bitcoin may not flinch. But when application layer corrects, tokenized AI projects could see cascading liquidations.
Capital misallocation creates arbitrage opportunities. Joshi's thesis implies that capital is not efficiently allocated across the AI stack. That means there are pockets of undervalued assets — especially in the intersection of AI and blockchain. Projects that combine verifiable compute with on-chain settlement (e.g., decentralized model training, proof-of-inference) are currently trading at a discount because they are neither pure AI nor pure crypto. That's a structural inefficiency I'm actively monitoring.
Takeaway: The Next Watch
Joshi's rolling bubble isn't a prediction of doom. It's a roadmap. The question is: which layer are you positioned for?
Infrastructure layer is peaking. Model layer is next. Tools and applications are the final stop. Crypto investors should watch for capital rotation signals: Nvidia's earnings growth deceleration, OpenAI's next funding round valuation, and the emergence of AI token issuance as a new narrative. When the AI bubble rolls out of infrastructure, it may roll into decentralized compute. That's the alpha.
Alpha detected. Position established.
Liquidation pending. Don't be the last to rotate.
Arbitrage window closing in 10 minutes.
From the field: During the 2020 DeFi Summer, I built a Python script to monitor MakerDAO stability fees and liquidation thresholds. That script caught a systemic arbitrage window before the market priced it in. I'm applying the same method today to track AI infrastructure CAPEX versus tokenized compute utilization. The signal is forming.
Risk-first education: Rolling bubbles are forgiving — until they aren't. Set stop-losses on AI-correlated crypto positions. Allocate 15% of your portfolio to decentralized compute tokens. The rest stays in Bitcoin. That's the survival stance.
Final thought: Joshi's framework is a gift for active managers. It tells you the bubble won't kill you — it will just rotate. The real danger is standing still.
