Steve Eisman, the investor who famously shorted the 2008 housing market—immortalized in The Big Short—recently dropped a bomb on the AI narrative. His warning? The AI boom rests on a fragile duopoly of OpenAI and Anthropic, and cheaper alternatives are already eroding their pricing power. For the crypto markets, where AI tokens have been the darlings of 2024 and 2025, this isn't just a stock market signal. It's a narrative earthquake.
Let me state this clearly: Structure beats speculation every time. And right now, crypto's AI narrative is built on speculation, not structure.
The Hook: Eisman's Contrarian Bet
Eisman's critique is not about AI's technical potential—it's about the revenue concentration at the top of the value chain. He argues that the AI-driven growth of big tech (Microsoft, Amazon, Google) is disproportionately tied to the income of just two companies: OpenAI and Anthropic. If cheaper alternatives—open-source models, distilled variants, or emerging competitors like DeepSeek—capture market share, the entire AI revenue cycle could contract. This is a classic value investor's structural concern: pricing power is the key, and it's under threat.
For crypto, the implication is immediate. The AI narrative has fueled a wave of token launches—decentralized compute networks (Akash, Render), AI agent platforms (Fetch.ai, Bittensor), and verifiable inference protocols. These projects assume a world where AI demand grows exponentially, requiring infinite compute. But if Eisman is right, that demand growth may slow, and the supply of compute—both centralized and decentralized—could face a glut.
Context: The Crypto AI Hype Cycle
2017 called. It wants its lessons back. Back then, I spent months dissecting ICO whitepapers, analyzing over 500 projects. I found that 85% of them had no viable roadmap—just a pitch deck and a promise. The same pattern is emerging in crypto AI today. Projects raise millions on the promise of "decentralized AI" without a clear demand side. The assumption is that AI will eat the world, and crypto will be the infrastructure layer. But infrastructure without demand is a ghost town.
Since 2023, the crypto AI sector has absorbed over $5 billion in venture funding, according to industry estimates. Tokens like TAO (Bittensor) have rallied 10x, and RNDR (Render) has seen similar surges. The narrative is intoxicating: AI agents need to rent compute, and decentralized networks offer cheaper, censorship-resistant alternatives. But the underlying assumption is that the AI industry's demand for compute will continue to grow at a breakneck pace, justifying the build-out of these networks.
Eisman's warning directly challenges that assumption. If the AI boom is driven by a duopoly whose pricing power is eroding, the entire demand curve could shift. Cheaper AI models mean less revenue for cloud providers, which means less capital expenditure on GPUs and data centers—and potentially less demand for decentralized compute.
Core: The Anatomy of the Revenue Chain
Let's break down the revenue chain that Eisman is questioning. The current AI economic model looks like this:
- OpenAI and Anthropic sell API access and subscriptions to enterprise and consumer clients. OpenAI's 2024 revenue is estimated at $50-80 billion annually (industry estimates), with a significant portion coming from Microsoft's Azure consumption.
- Cloud providers (Microsoft, Amazon, Google) invest hundreds of billions in AI infrastructure, partly to host these models. Microsoft's AI-related capital expenditure alone exceeded $50 billion in 2024.
- Hardware suppliers (NVIDIA, AMD) sell GPUs to these cloud providers, driving the entire semiconductor cycle.
If the duopoly's revenue slows—due to cheaper alternatives that offer 90%+ price reductions per token—the entire chain shifts. Cloud providers may cut capex, NVIDIA's orders may be delayed, and the narrative of "AI will need infinite compute" collapses.
For crypto, the decentralized compute narrative is an extension of this chain. Projects like Akash and Render assume that the demand for AI inference will spill over to decentralized networks because they offer lower costs. But if the cost of centralized inference drops to near-zero (thanks to open-source models and hyperscaler competition), the value proposition of decentralized compute becomes marginal.
Data point: As of early 2025, the price of inference via GPT-4o has dropped by over 90% from its peak in 2023. Open-source models like Llama 3.1 and DeepSeek-V3 now match or exceed GPT-4 on many benchmarks, at a fraction of the cost. The "cheaper alternative" that Eisman mentions is already here. It's not a future threat—it's a present reality.
First-person experience: In my 2020 DeFi analysis, I saw a similar pattern. The "yield farming" narrative collapsed when liquidity providers realized that the underlying protocols had no sustainable revenue. Today, I'm analyzing crypto AI projects with the same lens. I ask: Who is paying for this compute? Is the demand real, or is it subsidized by token emissions? The answer, in most cases, is the latter.
Contrarian Angle: Why Eisman May Be Wrong (and What It Means for Crypto)
Eisman's warning is sharp, but it's not the whole story. The contrarian view is that the commoditization of AI models—the rise of cheap alternatives—could actually increase total demand for compute. This is Jevons paradox: as the cost of a resource falls, usage rises. If AI inference becomes nearly free, it could be embedded into every application, from search to logistics to gaming. The demand for compute could explode, not contract.
In this scenario, decentralized compute networks could become the backbone of the next wave of AI applications. Low-cost inference means more experimentation, more agents, more autonomous systems—all requiring compute. Cryptocurrencies like Bittensor, which incentivize the creation of open-source AI models, could benefit from a world where model commoditization is the norm.
But here's the catch: The crypto AI market is currently pricing in a scarcity premium. Tokens assume that decentralized compute is a scarce resource that will become more valuable as AI demand grows. In reality, compute is becoming less scarce. The barrier to entry for running AI models is dropping, thanks to cheaper hardware (Apple's M-series, AMD's MI300) and more efficient software (vLLM, TensorRT). Decentralized networks may be competing with a market that has infinite supply.
Blind spot: The market is ignoring the "decentralized sequencing" trap I've seen in Layer2. Just as Layer2 sequencers are centralized nodes dressed up as decentralized, many crypto AI projects are centralized compute providers with a token wrapper. Akash, for example, relies on a small number of providers. Render's network is dominated by a few large GPU farms. The narrative of "decentralized" is often a marketing gimmick.
Takeaway: The Next Narrative Shift
Eisman's warning is a canary in the coal mine for crypto AI. The narrative that has driven token prices since 2023 is built on the assumption of infinite demand growth. But the data shows that AI pricing power is eroding, and the revenue chain is fragile. Crypto projects that are merely infrastructure-for-hire will face a reckoning as centralized alternatives become cheaper and more efficient.
The next narrative shift will likely be from "AI infrastructure" to "AI verifiability"—the need to prove that a model was executed correctly, or that data was generated by a specific agent. This is where blockchain's unique value proposition (trust, transparency) intersects with AI's core problem (black-box execution). Projects like Ritual, io.net, and Gensyn are already positioning themselves here. But they need to prove real demand, not just narrative.
Structure beats speculation every time. I've seen this cycle before—in 2017 with ICOs, in 2020 with DeFi, and in 2021 with NFTs. The narrative always overshoots reality. The best investors wait for the narrative to collapse, then buy the survivors. Eisman is doing the same with AI stocks. For crypto, the lesson is simple: Don't bet on the narrative. Bet on the data.
2017 called. It wants its lessons back. The AI crypto narrative is due for a structural recalibration. The question is whether you're ready to read the writing on the wall.