In his latest memo, Howard Marks, a figure revered by Warren Buffett, examines the current market psychology and warns that we may be looking at an imminent bubble. According to Marks, the main fuel behind this euphoria is the fascination with novelty. Because these technologies have no precedents, emerging technologies do not offer a track record to counterbalance market expectations. This is the case with the startup Thinking Machines, which raised $2 billion in its seed round without even revealing its product. With rumors that its valuation already stands at $50 billion, the market confirms that pedigree today trumps any business plan.
“Tech giants like Nvidia invest billions in AI startups that, shortly thereafter, use that capital to buy computing power in the cloud or chips from those same investors”
Nevertheless, the most striking warning sign is the prevalence of “circular agreements”. Tech behemoths such as Nvidia invest billions in AI startups that, in turn, use that capital to buy cloud computing power or chips from those same investors. This creates the illusion of explosive revenue growth that looks solid on a balance sheet, but it lacks genuine external demand and a broad base.
On the left, a map of AI sector companies. On the right, a comparison between Nvidia’s value and a number of European firms. Photo: Bloomberg and multiples.vc
Meanwhile, as markets pour hundreds of billions into data centers and GPUs (Graphics Processor Units), Ilya Sutskever, cofounder of OpenAI, argues that brute-force scaling is hitting its limits and will stop delivering significant advances. Instead, he predicts progress will depend on fundamental research, placing human-level intelligence at a distance of five to twenty years.
Additionally, the industry is increasingly reliant on aggressive leverage to finance assets with uncertain lifespans. There are firms issuing thirty-year bonds to build data centers full of chips that could become obsolete in as little as three years. When capital starts chasing bad ideas because the good ones have run out, we approach what economists call a Minsky moment: the final exhalation of euphoria.
The Law of Uncertainty
However, spotting a bubble is not the same as predicting its bursting. Despite these clear warning cues, we cannot know for sure if the music will stop, or when it will happen.
This uncertainty was illustrated when Michael Burry, the investor who correctly predicted the 2008 housing collapse, shut down his hedge fund. For nearly two years, Burry warned about the AI sector, taking substantial short positions against semiconductor and software firms during the boom, without success. As he warned long ago: “The market can stay irrational longer than you can stay solvent.”
Turning Point or Reversion?
If history teaches us that gravity always prevails in the end, what kind of collapse should we expect? Marks draws a fundamental distinction between two types of bubbles, and his diagnosis offers a glimmer of hope.
“The AI boom resembles past turning-point bubbles: the 1860s ‘railroad fever’ or the 2000 dot-com bubble”
We are unlikely to face a mean-reversion bubble, such as the 2008 subprime mortgage crisis. That one was a bubble fed by toxic debt and financial engineering, which, when it burst, left behind wreckage and recession. By contrast, the AI surge resembles those turning-point bubbles of the past: the 1860s railroad fever or the dot-com bubble of 2000.
In those cases, investors lost more than their shirts; yet the physical infrastructure (railways and fiber-optic cables) remained to propel the next century’s growth. Think of Netscape: it was the face of the Internet in the mid-1990s and ended up crushed by competition. Its fate reminds us that picking winners in the heat of a market is a gamble, not a science. The key question, therefore, is not only how to survive the burst, but how to manage the powerful infrastructure that will remain as a legacy.
What to Do After the Bubble Bursts?
Once the dust settles, we will face the hard task of integrating this technology into our economy. This requires a concrete strategy, for both investors and policymakers.
“Regarding policy and society, Daron Acemoğlu and Simon Johnson argue that we must reject the idea that new technologies automatically help workers”
For investors, Marks advises avoiding extremes. The all-in approach risks ruin in the event of a collapse, while an entirely outsized stance risks missing the technological shift of our generation. Investors should recognize they operate in a high-uncertainty zone. This means abandoning the lottery-ticket mindset that justifies any price for a potential winner and instead seeking firms with tangible cash flows capable of surviving a Minsky moment.
Regarding policy and society, Nobel laureates Daron Acemoğlu and Simon Johnson argue that we must reject the myth of the “productivity wagon”: the idea that new technologies automatically aid workers. To ensure that the post-bubble AI era benefits the many, four specific reforms are needed:
- Invest in labor, not just capital. Our current tax system often subsidizes the purchase of equipment (automation) while taxing human labor. Governments should create incentives for firms to use AI to augment workers’ abilities rather than replace them.
- Antitrust laws. Data is the fuel for AI. If, after the bubble, a handful of monopolies control the world’s data, they will control the economy. Regulators should enforce data interoperability and prevent information hoarding to ensure a competitive environment.
- Prepare for replacement. MIT’s new Iceberg Index warns that 11.7% of the US labor market is already economically replaceable by current AI. We need to bolster “compensatory powers,” revitalizing unions and civil-society groups to give workers a voice in decision-making.
- Treat AI infrastructure as a public service. Just as electricity and telecommunications became regulated essentials, access to computing and cloud must be affordable and universal. If the AI bubble bursts, governments should repurpose surviving infrastructure so its benefits do not vanish with failed companies.
The bubble may burst tomorrow, or it may not. But the choices we make about the infrastructure it leaves behind will shape the economic reality of the next decade.
