Fork in the road ahead. Steve Eisman, the 'Big Short' legend who made a fortune betting against subprime mortgages, just dropped a bomb on the AI narrative—and nobody in crypto is listening. But they should be. The same revenue concentration that doomed Terra-Luna is now infecting the AI tech stack. And I've seen this pattern before: a circular dependency between hype, capital expenditure, and a handful of key players that looks bulletproof until it's not.
Liquidity evaporation detected. Let me rewind. Eisman recently told a public audience that the AI boom's growth story is dangerously tied to OpenAI and Anthropic, two companies whose revenue streams are the primary pillars of the 'AI revenue' narrative for big tech. He warned that 'cheaper alternatives' could erode their market share, making large tech companies' growth 'unstable.' This is not a casual remark. It's a structural indictment of the entire AI value chain. And as someone who spent years on the front lines of crypto's boom-bust cycles—from the 2017 Ethereum Classic hard fork sprint to the 2022 Terra-Luna crash—I can tell you that the parallels are screaming evidence of a systemic risk that the market is blissfully ignoring.
Context: Who Is Eisman, and Why Should Crypto Care?
Steve Eisman is not a crypto analyst. He's a value investor who made his name by identifying hidden concentration risks in opaque markets. In 2008, he saw that mortgage-backed securities were built on a fragile web of subprime loans, and he bet against the entire system. Today, he's looking at the AI industry and seeing the same architecture: a small number of dominant players (OpenAI, Anthropic) whose revenue is the linchpin for the AI growth thesis of Microsoft, Google, and Amazon. The fact that he's speaking out now—after a two-year rally in AI-related stocks—is a signal that should flash red for anyone holding AI-themed crypto tokens, GPU mining operations, or even just a bullish position on the tech sector.
But why should crypto specifically care? Because the crypto market has a history of amplifying narrative-driven booms and then crashing when the underlying revenue assumptions fail. We saw it with DeFi in 2020—liquidity mining APY was a subsidy, not real user demand. We saw it with NFTs in 2021—metadata storage failures exposed ownership risks. And we saw it with Terra-Luna in 2022—a circular dependency between a stablecoin and its native token that worked until it didn't. Eisman's warning is about a similar circular dependency: AI revenue feeds cloud provider growth, which justifies huge capital expenditure, which in turn props up semiconductor and infrastructure stocks. Break one link—the revenue from OpenAI and Anthropic—and the whole chain collapses.
Core: The Technical Anatomy of the Revenue Concentration Risk
Let me break down the mechanics. Eisman's core argument is that OpenAI and Anthropic are the two primary 'revenue pillars' of the AI narrative. But what does that mean in numbers? Based on my own experience parsing SEC filings and on-chain data, I can tell you that the 'AI revenue' reported by Microsoft, Google, and Amazon is a composite of multiple streams: cloud services (Azure, GCP, AWS) that include AI model inference, direct API sales from their investments (OpenAI for Microsoft, Anthropic for Google/Amazon), and internal AI usage. The problem is that the growth of these streams is heavily concentrated in the API sales of the two leading model companies.
Metadata mismatch found. The market narratives treat AI as a monolithic growth story, but the underlying data tells a different story. OpenAI's estimated annualized revenue for 2024 is around $50-80 billion—impressive, but still a fraction of the capital expenditure poured into AI infrastructure. Cloud providers are spending an estimated $200 billion annually on AI-related capex, and a significant portion of that is justified by the expectation that OpenAI and Anthropic will continue to generate high-margin API revenue. But here's the catch: the pricing of large language model APIs has dropped by over 90% since early 2023. Cheaper alternatives—from open-source models like Llama, Qwen, and DeepSeek to smaller, specialized models—are closing the performance gap while costing a fraction of the price. Eisman's 'cheaper alternatives' are not a hypothetical threat; they are a real, accelerating trend.
Pattern emerging from chaos. I recall a similar dynamic in the 2020 Uniswap V2 debate. At the time, the market believed that AMMs were bulletproof liquidity aggregators. I published a thread deconstructing the constant product formula, arguing that it created hidden impermanent loss traps for retail users. The same dismissive response happened then: 'it's just a technical detail, the growth story is intact.' But the hidden risk was real—and it reshaped the entire DeFi landscape when the market corrected. Today, the hidden risk in AI is the revenue concentration. If cheaper alternatives capture even 10% of the API market share from OpenAI and Anthropic, the revenue growth for cloud providers could slow dramatically. That would trigger a repricing of the entire AI capex cycle, hitting everything from NVIDIA's GPU orders to AI-focused crypto tokens like Render or Akash.
But let's go deeper. The capital expenditure irreversibility is the key. Once cloud providers build data centers and order GPUs, those costs are sunk. They cannot easily scale back. If AI revenue growth disappoints, companies will still have to depreciate those assets, crushing margins. This is exactly the 'liquidity evaporation' I saw in the 2022 Terra-Luna crash: the UST-LUNA circular dependency created an illusion of stability, but once the revenue from new deposits dried up, the entire system collapsed. The AI revenue cycle has a similar circularity: Microsoft's Azure AI growth is partly driven by OpenAI's own consumption of compute, and OpenAI's revenue comes from API and subscription users. If those users migrate to cheaper alternatives, the entire chain shrinks. The market is pricing in a smooth growth curve, but the second derivative is about to turn negative.
Contrarian: The Unreported Angle—Google Is the Silent Winner
Here's the contrarian take that most analysts are missing. Eisman's warning implicitly highlights a winner: Google. While the narrative focuses on OpenAI and Anthropic as the duopoly, Google has a unique position. It owns its own TPU hardware, it has the Gemini model family, it has a massive distribution channel (Google Cloud, Android, Search), and it has a strong open-source ecosystem (via TensorFlow and JAX). If the 'cheaper alternatives' trend gains steam, Google is best positioned to offer competitive, low-cost models without sacrificing margins. In fact, Google's AI revenue is less dependent on a single model company than Microsoft's (which relies on OpenAI) or Amazon's (which relies on Anthropic). The duopoly's pricing power erosion could be a net positive for Google, allowing it to capture market share in the 'efficiency race' while the others struggle to maintain premium pricing.
Furthermore, the market is ignoring the application layer's rising bargaining power. Enterprise customers are moving from 'single model vendor lock-in' to 'multi-model routing and hybrid deployment.' This is analogous to the DeFi composability trend: users want to pick the best tool for each task, not bundle everything from one provider. That structural shift directly undermines the pricing power of model layer companies. The 'cheaper alternatives' are not just a threat to OpenAI and Anthropic; they are the vehicle for the application layer to capture more value. This is the same pattern I identified in the 2021 Bored Ape Yacht Club metadata investigation: the centralized IPFS gateways were a single point of failure that the market ignored until 0.5% of the images were corrupted. Today, the single point of failure is the revenue concentration in two model companies. The market will wake up when the first major cloud provider misses AI revenue guidance.
Takeaway: What to Watch Next
So where does this leave us? The next watchpoint is the earnings reports from Microsoft, Google, and Amazon for the next two quarters. Specifically, look for any deceleration in the 'AI services' revenue line, and any commentary about increased competition from cheaper models. If the cloud providers start to hint at a 'normalization' of AI growth, the market will reprice the entire capex cycle. That will hit not just tech stocks, but also crypto tokens that are tied to GPU compute (like Render, Akash, or even tokens related to AI infrastructure). The bull market euphoria is masking a structural risk—and Eisman is the canary in the coal mine.
My advice: look at the 'cheaper alternatives' as a catalyst, not a threat. For the crypto world, this could be the moment when decentralized AI compute networks (like Akash) gain traction as enterprises seek cheaper, more flexible infrastructure. The 'efficiency race' will favor protocols that can offer low-cost, verifiable computation over centralized cloud providers. That's the fork in the road ahead. The question is whether the market will see it before the liquidity evaporates.
