Written by: @jonah_b, Researcher at Blockchain Capital
Compiled by: Jia Huan, ChainCatcher
It's hard not to notice the striking parallels between the current AI boom and previous crypto market manias.
The crypto market is an ideal lens for observing how people respond to a major technological wave: it reveals the good, the bad, and the downright ugly. This is because crypto cycles unfold rapidly; early‑stage projects can leverage tokens to access liquidity that other markets struggle to provide, while market behavior is, by default, transparent, given that blockchain data is inherently open.
We've learned a great deal from this market, and those lessons apply equally to AI. If you're investing in AI, this article is for you.
Lottery-style betting
Massive success breeds copycats and fuels FOMO. Once an asset class produces a game-changing, runaway winner, investors rush in, seeking to replicate its trajectory. Yet second‑wave projects rarely match the heights of their predecessors; many end up as castles in the air, leaving vast sums of capital stranded.
Bitcoin has become a trillion-dollar asset. Subsequently, Ethereum has become a hundred-billion-dollar asset, and Solana has also become a tens-of-billion-dollar asset, proving that this market can produce more than just one giant winner.
As a result, a wave of investment in new blockchain projects swept through the industry. Venture capital firms treat these ventures like lottery tickets—betting that even a single successful bet could yield returns several times the size of the entire fund.
Today, the valuations of many emerging AI labs are also based on lottery‑style expectations.
Both OpenAI and Anthropic are on track to reach trillion-dollar valuations. The underlying formula is straightforward: start with a highly coveted market, then bring in a latecomer that has already proven its success. As a result, every new entrant is seen as the next golden ticket to riches.
Back then, many L1 projects secured valuations in the billions of dollars with little more than a white paper and a founding team. Their narratives were equally compelling:
What if the global economy were all running on our blockchain?
Similarly, emerging AI labs have raised billions of dollars, backed by a clear research vision and founding teams poached from OpenAI, Anthropic, or Google DeepMind.
What if they really could create a "God of Machines"?
However, all too often, this investment hinges solely on the expectation that a higher‑priced buyer will eventually step in.
Many early-stage investors aren't necessarily evaluating a company's current fundamentals or its future prospects. They know that the addition of a star‑level talent, or a partnership with a hyperscale cloud provider, could be enough to prompt investors to push the company's valuation up yet again. Coupled with today's increasingly robust secondary‑market liquidity, they assume that another buyer will always come along.
Market chaos
As BCAP GP @CremeDeLaCrypto noted on the Bankless show, whenever large amounts of capital flow into a market, there's always a swarm of speculators and scammers eager to chase quick profits, rushing wherever the hype is hottest.
For more than a decade, crypto investors have been witnessing a recurring pattern: wave after wave of projects, touting "tokens as products," are launched into the market—fueled by dubious market‑making practices, highly unfavorable FDV‑to‑circulating‑supply ratios, investor‑unfriendly SAFT agreements, and an endless array of other market‑manipulative tactics.
Today, the AI industry is witnessing a similar scenario. For example, there are structures composed of three tiers of SPVs that charge exorbitant fees…
However, there is a key difference between the two: in the crypto market, prices are publicly disclosed, and tokens can be traded openly. By contrast, the pricing and valuations of AI companies are determined in opaque, illiquid secondary markets.
Even without considering speculative behavior, AI investors can observe in the crypto market that shifts in market structure ultimately determine who captures the profits.
For example, what AI investors are currently betting heavily on may ultimately turn into a standardized commodity.
A Case from the Crypto Market: Blockspace Has Become a Commodity
Blockspace was once scarce and expensive, prompting capital to flow in in an attempt to expand supply.
But the industry later went a bit too far: more and more L1 chains were launched, and Ethereum also added L2s. In the end, blockspace went from scarce to abundant, even oversupplied.
This is good for technological advancement, but not necessarily for investors. Today, many alternative L1s still generate very limited revenue.
Early in the market, there was widespread betting on the "fat protocol" thesis, but in the end, the "fat application" thesis prevailed.
As blockspace becomes cheaper, users spend less on the underlying infrastructure and more on upper-layer applications. Application‑layer projects such as Tether, Hyperliquid, Aave, and Polymarket, in turn, capture the lion's share of revenue.
AI Case: Models May Be Commercialized
AI may be repeating the scenario that the crypto market experienced in its early days.
China's AI labs are increasingly making public the weights of ever-more-powerful models. I have previously explained the incentives driving this practice.
If model weights become standardized commodities and model prices continue to decline, value will shift upstream and downstream within the technology stack.
Applications will be the primary beneficiaries: as the marginal cost of using a model approaches the marginal cost of running it, applications will no longer have to bear the high profits generated at the model layer.
OpenAI and Anthropic are unlikely to be directly impacted by this shift, as they already have established user bases, enterprise customer relationships, and developer‑focused distribution channels. However, they are the exception.
In this sense, they resemble Hyperliquid in the AI space more than L1s that merely provide underlying infrastructure.
In other words, OpenAI and Anthropic have already achieved vertical integration. Other AI labs that cannot directly reach end users may find themselves in an even more challenging position.
Another consequence of declining model profit margins is that the lowest layers of the technology stack—energy and hardware—are poised to enjoy higher margins. This trend may become even more pronounced when computing power supply is constrained by physical limitations.
In short, the hardware and application layers may capture a larger share of profits, while the model layer faces margin compression. Yet substantial investment capital is flowing precisely into the model layer.
None of this is new.
From a broader perspective, this is actually not surprising.
In "Technological Revolutions and Financial Capital," Carlota Pérez argues that such technological revolutions proceed through several phases: the infrastructure‑building phase, the euphoria phase, the crash phase, and the deployment phase.
Financial capital tends to overinvest in infrastructure during boom periods, but this overbuilt infrastructure—now priced low—will in turn underpin the development of next-generation applications.
This helps explain why the crypto market has seen excessive overbuilding of blockspace. The expansion of models in the AI sector could become the next such case.
Investing in emerging AI labs presupposes that they can deliver substantial returns on R&D spending. The history of the crypto market and alternative L1s should at least prompt us to question this assumption.
Of course, if AGI were to emerge, none of the above assumptions might hold. After all, we have no way of knowing what the economy will look like once AGI arrives.