Trust nothing. Verify everything.
Let’s start with a hard fact. Over the past 12 months, the four largest cloud providers—Microsoft, Google, Amazon, Meta—have committed over $200 billion in combined capital expenditures, with the bulk directed toward AI infrastructure. Yet AI-related revenue growth, while impressive, still lags these outlays by a significant margin. The data does not lie: a structural gap exists between capital deployed and value realized.
This is not, however, the classic apocalyptic bubble narrative that dominates headlines. According to Dhaval Joshi, Chief Strategist at BCA Research, the AI market is not a single supernova waiting to explode. It is a “rolling bubble”—a sequence of localized overvaluations that migrate across the technology stack, from silicon to applications. I have spent the last 14 years dissecting cryptographic protocols and market structures, and this framework resonates with what I observed in the 2022 Terra-Luna collapse: a single point of failure masked by a cascade of smaller, compounding misallocations. The ledger does not forgive, and neither does capital.
Context: The Rolling Bubble Thesis
Joshi’s argument, as reported by Crypto Briefing, challenges both the “AI will rise forever” and the “AI is about to crash and burn” camps. The key insight is that AI’s valuation bubble is not monolithic. It moves through four distinct layers of the technology stack: infrastructure (GPUs, data centers), models (foundation LLMs), tooling (frameworks, middleware), and applications (end-user solutions). Each layer attracts a wave of speculative capital, peaks, then deflates as the narrative shifts to the next layer. This prevents a single, synchronized collapse—but it also creates a persistent state of capital misallocation.
From my own work stress-testing Polygon zkEVM’s proof generation, I know that latency in one layer can cascade into inefficiencies in another. The same principle applies here: when capital overflows into infrastructure, it inflates GPU prices and data center valuations, but the models and applications that consume that compute have not yet proven their ROI. The bubble merely rolls forward, postponing the reckoning.
Core: A Technical Dissection of the Capital Flow
Let’s get granular. The AI stack mirrors the layered architecture of blockchain networks. Infrastructure is the base layer—equivalent to Ethereum’s consensus layer. Models are the execution layer—like the EVM. Tooling is the middleware—think of oracles or indexers. Applications are the dApps. In crypto, we saw a similar rolling bubble from 2020 to 2022: DeFi → NFTs → Gaming → Metaverse. Each layer experienced a spike in total value locked (TVL) followed by a brutal correction. The same pattern is unfolding in AI.
Using my analytical framework, I have mapped the current capital flow. In 2023, infrastructure was the hot spot: NVIDIA’s market cap surged past $3 trillion, and data center REITs soared. By mid-2024, the narrative began rotating to foundation models—OpenAI’s $86 billion valuation, Anthropic’s $18 billion round. Now, in early 2025, we see the first signs of application-layer fever: Palantir, C3.ai, and a host of AI-powered SaaS startups commanding multiples that defy traditional revenue metrics. The capital is rolling, but the underlying revenue generation remains weak.
Based on my audit experience, I have identified three critical failure points in this roll:
- Infrastructure overhang: The $200 billion in CAPEX assumes a compound annual growth rate (CAGR) of 40%+ in AI compute demand. But if applications fail to achieve mass adoption, GPU utilization rates will drop, and spot prices for H100 instances will plummet. I have seen this in crypto when Layer 2 sequencers spent millions on hardware before TVL materialized. Complexity is the enemy of security, and overbuilt infrastructure is a liability.
- Model layer commoditization: As open-source models (Llama, Mistral) narrow the gap with proprietary ones, the moat for foundation model companies shrinks. Capital that poured into valuation rounds will have nowhere to go but down. This is the equivalent of a DeFi protocol with no sustainable yield—the liquidity evaporates when the next hot thing appears.
- Application layer dependency: Most AI applications today are thin wrappers over APIs. They lack sticky network effects. If the underlying model providers raise prices or change terms, the entire application layer crumbles. I have seen this on-chain when a single oracle failure cascades across multiple protocols.
Contrarian: The Blind Spot – Crypto as the Next Roll
Here is where the conventional analysis misses the mark. Most commentators assume that a rolling AI bubble will remain contained within the AI sector. But capital is not siloed. The same macro forces—low interest rates, search for yield, fear of missing out—drive both AI and crypto markets. When the AI bubble rolls out of one layer, the capital does not disappear; it rotates to another high-risk, high-narrative sector. Historically, that has been crypto.
Consider the data: In 2021, when AI hype was nascent, crypto TVL peaked at $180 billion. In 2023-2024, as AI infrastructure soared, crypto remained subdued. Now, with AI application-layer valuations looking stretched, the next rotation could bring capital into decentralized infrastructure projects—DePIN, GPU marketplaces, or even AI-related blockchains like Bittensor. The narrative would be: “AI needs decentralized compute to avoid censorship.” That is a story that can attract billions.
But here is the risk: the rolling bubble postpones the inevitable aggregate correction. The market is not safer; it is simply more complex. As I wrote in my 2024 report on the Terra collapse, the absence of a single crash does not mean the system is solvent. It means the failure is distributed across time and layers. When the macro environment shifts—say, a rate hike or a geopolitical shock—all layers may deflate simultaneously. The rolling bubble becomes a rolling avalanche.
Takeaway: Vulnerability Forecast
Where does this leave us? The next 12 months will be critical. I will be watching the GPU rental spot price as a leading indicator of infrastructure demand. If it drops below the cost of compute, the infrastructure layer’s bubble has burst. I will also monitor foundation model API pricing—if OpenAI cuts prices aggressively, it signals a commoditization race that will compress margins across the stack.
For crypto investors, the question is not whether AI is in a bubble, but when the capital flowing through AI layers will spill into decentralized alternatives. The ledger does not forgive — but it also does not discriminate. Every layer of misallocation will eventually be reconciled. The only question is: will you be holding the bag when the music stops?
Trust nothing. Verify everything. Complexity is the enemy of security.