On May 21, 2024, Abby Joseph Cohen—the woman who called the 1998 bull run and spent decades at Goldman Sachs translating macro data into actionable strategy—issued a warning that cut through the market’s AI euphoria like a cold front. Her message was not a call to sell everything. It was a precise, clinical observation: the economy is uneven, and the current pace of AI investment is unsustainable.
For anyone who has spent years auditing on-chain flows and corporate balance sheets, her words land with the weight of a verified transaction hash. This is not about whether AI will transform industries—it will. The question is whether the capital being deployed today will generate returns that justify the prices being paid. My answer, based on the data I have seen across both traditional markets and crypto, is a cautious no.
Let me be clear about what I am not saying. I am not predicting an immediate crash. I am not dismissing the transformative potential of AI. What I am saying is that the current allocation of capital resembles the ICO mania of 2017 more than it resembles a rational build-out of infrastructure. And when that pattern emerges, the ledger does not lie—only the interpreters do.
Cohen’s warning centers on two distinct but interconnected phenomena. First, the economy is growing unevenly. Some sectors—semiconductors, data centers, AI-enabled services—are experiencing explosive demand. Others, including traditional manufacturing, consumer discretionary spending, and real estate, are stagnating or contracting. This is not a synchronized recovery. It is a bifurcated market where the top 10% of companies are capturing the vast majority of growth while the rest of the economy struggles to keep pace.
Second, the investment surge in AI is showing signs of overshoot. Capital is flowing into GPU clusters, model training runs, and AI startups at a rate that assumes near-perfect execution and unlimited future demand. But the revenue models for many of these ventures remain unproven. The gap between invested capital and realized revenue is widening, and that gap is the definition of an unsustainable trend.
From my perspective as someone who has traced the flow of capital through both traditional markets and blockchain networks, Cohen’s warning has a direct parallel in the crypto world. The current AI narrative is the crypto equivalent of the 2021 DeFi summer—except this time, the tokens are replaced by GPUs, and the yield farms are replaced by compute clusters.
I have spent the last 21 years observing markets, and I have developed a set of forensic habits that serve me well in times like this. When I review a protocol, I do not read the whitepaper first. I go to the code. I check the contract addresses. I verify the transaction history. I look for the mismatch between what the team claims and what the chain records. Ledgers do not lie—only the interpreters do.
Applying the same discipline to the AI investment boom, I see several red flags that deserve attention. First, the concentration of investment in a handful of companies and technologies creates systemic risk. If NVIDIA’s guidance disappoints, the entire AI trade unwinds. Second, the build-out of AI infrastructure is happening faster than the development of applications that can monetize that infrastructure. This is a classic capacity glut pattern, and it rarely ends well. Third, the regulatory environment is shifting. As MiCA and other frameworks take effect, the compliance burden on AI-driven financial services will increase, potentially compressing margins in a sector that is already trading at full valuation.
Let me break this down further, because the details matter.
The Concentration Problem
When I look at the current AI investment landscape, I see a portfolio that is dangerously undiversified. The top AI companies—NVIDIA, Microsoft, Google, Amazon—are capturing the vast majority of capital inflows. This is not a healthy market structure. It is a monoculture. And monocultures are vulnerable to a single pathogen.
The pathogen in this case could be a disappointing earnings report, a regulatory crackdown, or a geopolitical event that disrupts the chip supply chain. When that happens, the sell-off will not be gradual. It will be a cascade, because the entire market is positioned the same way.
I have seen this pattern before. In the crypto markets, we witnessed the same dynamic during the 2021 bull run, when every project was a "DeFi protocol" and every token was "the next Ethereum." When the music stopped, the projects with real usage survived, but the ones with only narratives collapsed. The same will happen in AI. The companies with actual revenue and clear paths to profitability will survive. The ones that are trading on narrative alone will be destroyed.
The problem is that we cannot easily distinguish between the two right now, because the market is pricing everything as if the narrative is true.
The Capacity Glut
The second red flag is the pace of infrastructure build-out. Companies are ordering GPUs months in advance, data centers are being constructed at record speed, and cloud providers are expanding capacity as if demand will grow exponentially forever. But demand is not infinite. It is constrained by the number of applications that can actually use this compute power effectively.
We are building a massive highway system, but we only have a few cars to drive on it. The cost of building that highway is being paid today, but the toll revenue is uncertain. This is not sustainable in the long term.
I have seen this pattern in the crypto world as well. In 2022, we saw a massive build-out of Layer 2 infrastructure—dozens of rollups, sidechains, and app-specific chains. But the user base did not materialize as quickly as the infrastructure. Many of those chains are now running at a fraction of their capacity, with their tokens trading at a fraction of their peak values.
The same dynamic is now playing out in AI. The infrastructure is being built, but the applications that will generate returns are still in their infancy. This does not mean the infrastructure is worthless—it means the timing of the returns is uncertain. And in a market that demands immediate gratification, uncertainty is punished.
The Regulatory Overhang
The third factor is regulation. In the European Union, MiCA is now fully in effect, and it imposes strict compliance requirements on digital assets. In the United States, the SEC is taking an increasingly aggressive stance on both crypto and AI-related financial products. The intersection of these two regulatory regimes is creating a compliance burden that is passed directly to honest users.
Most KYC procedures are theater. A few wallet holdings can bypass them. The costs of compliance are borne by the people who follow the rules, while the people who break them find loopholes. This is not a sustainable equilibrium.
In the AI space, the regulatory picture is even murkier. How do you regulate a model that generates code? How do you assign liability when an AI-driven trading algorithm makes a mistake? These questions are unresolved, and they create significant legal risk for companies operating at the intersection of AI and finance.
I have submitted formal complaints to the Polish Financial Supervision Authority about platforms that failed to implement real-time transaction monitoring. The response was slow, and the enforcement was even slower. This is the reality of regulation in a rapidly evolving technological landscape—it lags behind the innovation it is meant to govern.
The Counterintuitive Angle: What the Bulls Get Right
Before I continue with the bear case, I need to acknowledge the counterintuitive angle. The bulls are not wrong about the long-term potential of AI. They are wrong about the timing and the magnitude.
AI will transform industries. It will create new categories of products and services. It will generate enormous wealth for the companies that execute well. But the market is pricing in the best-case scenario for all of these outcomes, and the best case is rarely what happens.
In my experience, the best investments are made when there is a gap between perception and reality. When everyone believes something is true, it is usually already priced in. The opportunity lies in finding the areas where the market is wrong—where the risk is higher than the price suggests, or where the potential is higher than the market recognizes.
In the current AI trade, the risk is higher than the price suggests. The market is ignoring the concentration risk, the capacity glut, and the regulatory overhang. It is pricing in perfection. And perfection is a fragile thing.
There are, however, pockets of opportunity. The companies that provide the infrastructure for AI—the power companies, the cooling system manufacturers, the network providers—are less exposed to the bubble risk than the pure-play AI companies. They have real revenue, real customers, and real margins. They are not trading on narrative alone.
Similarly, in the crypto space, the projects that are building real infrastructure—the ones with actual users and actual transaction volume—are undervalued relative to the projects that are still in the whitepaper phase. The market is rewarding narrative over substance, and that creates opportunities for disciplined investors.
The Takeaway: Accountability and Adaptation
The bottom line is that Cohen’s warning is a signal, not a prediction. It is a reminder that markets are not rational—they are emotional. And when emotion drives prices, the eventual correction is often severe.
As investors, we have a choice. We can follow the crowd and hope the music continues. Or we can step back, examine the data, and make decisions based on fundamentals rather than sentiment.
I have seen too many investors lose everything because they chased a narrative without checking the code. I have seen too many projects fail because they raised money on hype instead of building real value. The ledger does not lie—only the interpreters do.
The AI investment boom is not a fraud. It is a reflection of genuine technological progress. But it is also a reflection of human nature—the tendency to extrapolate current trends into infinity, to believe that this time is different, to ignore the warning signs.
Cohen’s warning is a reminder that this time is not different. The cycle is the same. The actors are different, but the dynamics are identical. And the ones who survive are the ones who understand the cycle, who prepare for the downswing, and who position themselves to take advantage of the opportunities that the correction creates.
The current market is a bear market in disguise. The AI trade is masking the underlying weakness in the broader economy. When the mask comes off, the correction will be painful. But it will also be cleansing.
In the meantime, the smart money is positioning itself for the transition. It is diversifying into value stocks, defensive sectors, and real assets. It is reducing exposure to the AI trade and increasing exposure to the areas that will benefit from the rebalancing.
The question is not whether the correction will happen. It is whether you will be prepared when it does.
As for me, I will continue to do what I have always done: follow the data, verify the code, and trust the ledger. Because in the end, that is all we have.