Post

Author: @jonah_b, Researcher at Blockchain Capital Translated by: Jiahua, ChainCatcher It's hard not to notice the striking similarities between the current AI frenzy and previous rounds of crypto market mania. The crypto market is well-suited to observe how people behave in the face of a major technological wave: there are good aspects, bad aspects, and the worst aspects. This is because the crypto market cycles move quickly, early projects can gain liquidity through tokens that other markets rarely offer, and market behavior is by default public since blockchain data is inherently open. We have learned a lot from this market, and these lessons also apply to AI. If you are investing in AI, this article is for you. Lottery-Style Betting Huge successes trigger follow-the-leader behavior and create FOMO. Once an asset class produces a game-changing giant winner, investors rush in trying to replicate that success. But second-wave projects rarely reach the heights of the first wave; many end up as castles in the air, and a lot of money is lost. Bitcoin became a trillion-dollar asset. Then Ethereum reached hundreds of billions, and Solana became a multi-billion-dollar asset, proving this market can produce more than one giant winner. Thus, a wave of investment in new public blockchains followed. Venture capital firms treated these projects like lottery tickets: hitting one could earn multiples of the entire fund size. Today, many emerging AI labs are also valued on lottery-style expectations. OpenAI and Anthropic are both approaching trillion-dollar valuations. The formula behind this is simple: first, an extremely hyped market, then a proven successful follower. So every new entrant is seen as the next lottery ticket to wealth. Back then, many L1 projects got multi-billion-dollar valuations almost solely based on a whitepaper and a founding team. Their stories were compelling: What if the global economy runs on our chain? Similarly, emerging AI labs have raised billions based on a research philosophy and founding teams poached from OpenAI, Anthropic, or Google DeepMind. What if they really build a "machine god"? However, often these investments are just bets that a higher-priced next buyer will appear. Many early investors don’t necessarily evaluate the company’s current fundamentals or future prospects. They know that the addition of a star talent or a deal with a hyperscale cloud provider can push valuations higher. Plus, with increasing liquidity in secondary markets, they assume the next buyer will always show up. Market Chaos As BCAP GP @CremeDeLaCrypto said on the Bankless podcast, when large amounts of capital flow into a market, there will always be "a crowd of speculators and scammers chasing quick money wherever the hype is." For over a decade, crypto investors have witnessed this: wave after wave of projects claiming "tokens are the product," pushed to market by suspicious market-making tactics, investor-unfriendly high FDV/low circulating supply structures, unfavorable SAFT agreements, and endless other market tricks. Now, the AI industry is showing a similar scene. For example, those three-layer SPV structures with outrageously high fees... But there is one key difference: in crypto, prices are public and tokens are openly traded. AI company prices and valuations form in opaque, illiquid secondary markets. Even ignoring speculation, AI investors can see from crypto that changes in market structure determine who ultimately captures profits. For example, what AI investors are heavily betting on now may eventually become a commoditized product. Crypto Market Case: Block Space Becomes a Commodity Block space was once scarce and expensive, attracting capital trying to expand supply. But the industry went a bit too far: more and more L1 chains launched, and Ethereum added L2s. Eventually, block space went from scarce to abundant, even oversupplied. This is good for technology development but not necessarily for investors. Many alternative L1s still have very limited revenue. Early markets generally bet on the "fat protocol" theory, but the "fat application" theory ultimately prevailed. As block space became cheaper, users paid less for the underlying infrastructure and more for applications on top. Application-layer projects like Tether, Hyperliquid, Aave, and Polymarket ended up capturing the bulk of revenue. AI Case: Models May Become Commoditized AI may be replaying the early crypto market experience. Chinese AI labs are increasingly releasing stronger model weights publicly. I have explained the incentives behind this. If model weights become standardized commodities and model prices continue to fall, value will shift upstream and downstream in the tech stack. Applications will be the main beneficiaries: if the marginal cost of using a model approaches the marginal cost of running it, applications no longer have to pay for high model-layer profits. OpenAI and Anthropic are less vulnerable to this change because they already have user bases, enterprise clients, and developer distribution channels. But they are exceptions. In this sense, they are more like Hyperliquid in AI rather than L1s that only provide underlying infrastructure. In other words, OpenAI and Anthropic have achieved vertical integration. Other AI labs that cannot directly reach end users may face tougher challenges. Another result of declining model profit margins is that the lowest layers of the tech stack—energy and hardware—may gain higher margins. This trend may be more pronounced especially when compute supply is physically constrained. In short, hardware and application layers may take more profits while the model layer gets squeezed. Yet a lot of investment capital flows into the model layer. None of This Is New From a bigger cycle perspective, this is not surprising. Carlota Perez in "Technological Revolutions and Financial Capital" argues that such technological revolutions go through stages: infrastructure installation, frenzy, crash, and deployment. Financial capital overinvests in infrastructure during the frenzy, but these overbuilt, now cheap infrastructures support the next generation of applications. This helps explain why the crypto market saw block space overbuild. AI model expansion may be the next case. Investing in emerging AI labs assumes they can generate huge returns on R&D investment. The history of crypto markets and alternative L1s should at least make us question this assumption. Of course, if AGI arrives, none of the above judgments may hold. Because we have no idea what the economy will look like after AGI comes.