We didn’t learn from the 2022 Terra-Luna collapse. We didn’t learn from the ICO boom’s centralization of token distribution. Now, we are watching the same pattern repeat in the AI industry—except this time, the assets are not on-chain, but the governance failure is identical. Steve Eisman, the investor who bet against subprime mortgages and became the protagonist of The Big Short, has publicly warned that the AI boom rests on a fragile concentration of revenue. His target: OpenAI and Anthropic, the two closed-source model providers that have become the de facto pillars of the “AI revolution” narrative. Eisman argues that “cheaper alternatives” will erode their pricing power, destabilizing the entire growth story of Big Tech.
Governance isn’t just about smart contracts. It’s about understanding where power accumulates.
In the crypto world, we audit intent, not just syntax. We recognize that true decentralization requires distribution of economic power, not just code execution. The AI industry, driven by massive capital expenditure and narrative euphoria, has built its cathedral on a single point of failure: the revenue streams of two private companies. As a DAO Governance Architect who has spent years designing systems to resist centralization, I see the parallels with terrifying clarity. This is not a financial analysis—it is a governance diagnosis.
Let me state my bias upfront: I have been a vocal critic of the “AI hype cycle” since 2023, but not because I doubt the technology. I doubt the structural integrity of the economic model that has been erected around it. My background in auditing Ethereum smart contracts taught me to look for where risk is concentrated, not where returns are promised. When I designed the quadratic voting mechanism for Aave’s V2 governance, I learned that power silos are the root of systemic fragility. The AI industry has built exactly such a silo.
Context: The Two Pillars and the Narrative Machine
Steve Eisman’s warning, as reported by Crypto Briefing and other outlets, centers on the idea that the AI growth narrative of major tech companies—Microsoft, Google, Amazon—is overly dependent on the revenue performance of OpenAI and Anthropic. These two companies are the face of frontier AI models. OpenAI’s GPT-4o and Anthropic’s Claude 3.5 are the gold standards by which all other models are measured. Their APIs are the primary revenue generators for the AI cloud services of their respective backers.
But here is the structural flaw: the industry’s revenue concentration is not a secret. It is a feature of the current business model, not a bug. The large tech companies have invested billions into AI infrastructure—data centers, GPUs, networking—based on the assumption that demand for premium AI models will grow exponentially. The revenue of OpenAI and Anthropic has become the key performance indicator for this entire capital expenditure cycle. If that revenue falters, the entire house of cards trembles.
Eisman’s specific concern is that “cheaper alternatives” will eat into the market share of these dominant players. This is not a fringe hypothesis. Since early 2023, the price of API calls for large language models has dropped by over 90% per million tokens. Open-source models like Llama 3, Qwen 2.5, DeepSeek-V3, and Mistral have achieved performance levels that rival—and in some benchmarks surpass—closed-source models, at a fraction of the cost. The “open AI” ecosystem is not just a hobbyist playground; it is a competitive threat that is structurally undermining the pricing power of the duopoly.
Every line of code writes a history of power. The open-source AI models are writing a history of distributed accessibility.
Core Analysis: The Revenue Concentration Cascade
I will now dissect the mechanism that Eisman describes, using the forensic skepticism that I honed during my years of auditing smart contracts. The core argument is not that AI is a bubble—it is that the revenue cascade is fragile. Let me break it down step by step.
Step 1: The Duopoly’s Pricing Power
OpenAI and Anthropic currently command premium pricing because they offer the best performance. But “best” is a moving target. The performance gap between the frontier models and the open-source alternatives is narrowing rapidly. In the past 12 months, we have seen open-source models achieve scores on MMLU, HumanEval, and other benchmarks that are within 5-10% of the top closed models. For many use cases—customer support, content generation, code assistance—the gap is negligible. The premium is becoming harder to justify.
Based on my experience auditing DeFi protocols, I know that once a market reaches a certain maturity, the “best” product often becomes a commodity. The same dynamic is playing out here. The question is not if the duopoly will lose pricing power, but when and how quickly.
Step 2: The Cloud Revenue Correlation
Microsoft’s Azure AI growth is heavily tied to OpenAI’s consumption of its compute resources. Amazon’s AWS AI services rely on Anthropic’s models. Google’s cloud AI revenue is tied to its own Gemini model, but also to the broader ecosystem. If OpenAI and Anthropic face revenue pressure, they will reduce their compute consumption, which directly impacts the cloud providers’ AI revenue. This is not a hypothetical—it is a contractual dependency.
During my time as a governance architect for early DeFi protocols, I observed that protocols that relied on a single liquidity provider or a single market maker were inherently unstable. The same principle applies here. The cloud providers have become the “liquidity providers” for the AI model market, and their returns are dependent on a small number of “traders” (model companies).
Step 3: The Capital Expenditure Irreversibility
This is the most critical point that Eisman’s warning implies but does not explicitly state. The capital expenditure on AI infrastructure—especially data centers and high-end GPUs—is largely irreversible. Once a billion-dollar data center is built, it cannot be easily repurposed. The GPU orders are non-cancellable. If the revenue growth of the model layer slows down, the cloud providers cannot simply halt their spending. They are committed.
In the crypto world, we have a term for this: “locked value.” When a protocol locks up billions of dollars in liquidity, it creates a vulnerability. If the yield drops, the liquidity cannot flee instantly—it is trapped. The same is true for AI infrastructure. The capital is locked, and the returns are dependent on a narrative that is increasingly fragile.
Truth emerges from transparency, not from silence. The VCs and tech executives are silent about the capital expenditure irreversibility because it is their greatest vulnerability.
Step 4: The Earnings Multiplier Effect
Eisman’s warning is not about absolute revenue decline—it is about the rate of growth slowing down. The market has priced in a certain trajectory of exponential growth. If that trajectory shifts to linear or even parabolic deceleration, the valuation multiples will compress. The entire AI ecosystem—from Nvidia to the smallest AI startup—will be repriced.
Let me draw on my experience from the 2022 bear market. When the Terra-Luna collapse happened, the entire crypto market was repriced not because of a fundamental change in technology, but because the narrative of “risk-free yield” was shattered. The same principle applies here. The narrative of “AI will change everything, and the profits will be infinite” is the risk-free yield of the 2024-2025 stock market. If that narrative is undermined, the repricing will be brutal.
The Hidden Variables: Google and the Application Layer
There are two factors that Eisman’s public commentary may not fully address, but which are crucial to the analysis.
First, Google is the wildcard. Google owns its own TPU chips, its own Gemini model, its own massive distribution channels (search, Android, YouTube), and its own open-source ecosystem (TensorFlow, Gemma models). If the duopoly’s pricing power erodes, Google is uniquely positioned to benefit because it can offer a vertically integrated, cost-effective alternative. The “cheaper alternatives” that Eisman fears may actually be Googles orchestrated push into the open-source space.
Second, the application layer is gaining bargaining power. Enterprise customers are increasingly adopting multi-model routing strategies—they use different models for different tasks, switching based on cost and performance. This reduces lock-in and increases price sensitivity. The days of “we only use GPT-4” are ending. This structural shift in the buyer side is a direct threat to the duopoly’s pricing power.
Contrarian Angle: The Blind Spots of the Narrative
Now, let me challenge the prevailing narrative. Most analysts and investors are still bullish on AI because they believe the technology is transformative. They are right about the technology, but wrong about the business model. The contrarian view is not that AI is overhyped—it is that the current revenue concentration is a bug, not a feature. The market is pricing in a linear extrapolation of the status quo, but the status quo is changing.
Here is the blind spot: everyone assumes that OpenAI and Anthropic will remain the leaders because they have the best talent and the most data. But the open-source community is not a single entity—it is a globally distributed innovation network. The rate of progress in open-source AI is accelerating, not slowing. The “cheap alternatives” are not just small players; they are backed by major tech companies like Meta (Llama), Alibaba (Qwen), and even the Chinese government (DeepSeek). The geopolitical dimension adds another layer of complexity.
Another blind spot is the assumption that the large cloud providers will continue to subsidize the AI infrastructure indefinitely. But their shareholders are not infinitely patient. If the AI revenue growth disappoints, the pressure to cut capital expenditure will be immense. This is a governance failure: the decisions to invest in AI infrastructure were made by a small group of executives without a proper long-term risk assessment. Sound familiar? It is the same governance failure that led to the DeFi hacks and the Terra collapse.
We didn’t audit the intent of the AI infrastructure investment. We assumed that the executives knew what they were doing. They didn’t.
Takeaway: The Convergence of Lessons
What does this mean for the crypto and blockchain space? It means that the lessons we learned about centralization, governance, and risk concentration are not confined to our industry. They are universal. The AI industry is now facing the same structural challenges that DeFi faced in 2020-2022: a concentration of economic power, irreversible capital commitments, and a narrative that is disconnected from reality.
I am not saying that AI will crash and burn. I am saying that the current business model of AI is fragile. The solution is not to abandon AI, but to decentralize its economic structure. This is where blockchain-based governance mechanisms can play a role: decentralized AI marketplaces, staking models for compute resources, and transparent on-chain auditing of AI revenue flows.
As an early advocate for the convergence of AI and crypto, I have been working on a framework called “Verifiable AI” that ensures autonomous agents provide cryptographic proof of their actions. This is not just a technical solution—it is a governance solution. It ensures that the power of AI is distributed, not concentrated.
Truth emerges from transparency, not from silence. The silence of the AI industry about its revenue concentration is its greatest weakness.
Let me leave you with a question: If you were designing a governance system for the AI economy, would you allow a single point of failure like OpenAI or Anthropic to control the entire revenue stream? Or would you build in redundancy, checks, and balances? The answer is obvious. The challenge is execution.
