“如果OpenAI崩盘，全球股市恐怕会被清算”：一篇长达15k字的“大空头”文章引爆AI泡沫之争

Source: Wall Street Vaunting

As OpenAI gets closer to its IPO, a long blog post of about 15,000 Chinese characters has once again pushed the controversy over an AI bubble to new heights.

Ed Zitron, a commentator who has long been bearish on AI and has a large readership in the tech industry, put forward what is arguably his most aggressive judgment yet in a blog post he published recently: the real AI bubble is essentially an “OpenAI bubble.” If OpenAI ultimately fails, it will become the “Lehman Brothers” of the AI era—piercing not only the entire logic behind AI investing, but also potentially triggering a large-scale repricing of data centers, AI infrastructure, and even global technology stocks.

These views quickly drew the attention of financial media. In the media’s view, Zitron’s most core point is not whether AI has value, but whether OpenAI has a business model strong enough to support an entire AI capital cycle. If the answer is no, then the financing, computing investment, and capital expenditure system built around OpenAI could face a chain reaction.

Of course, this is not a market consensus. Investors including Howard Marks, co-founder of OakTree Capital, have recently said that, compared with the earlier view that AI might just be a bubble, they are now more inclined to recognize AI’s long-term value as a General Purpose Technology. They believe the industry is still in the early stage of commercialization.

AI bubble, or OpenAI bubble?

Unlike most “AI bubble” arguments, Zitron makes a more disruptive judgment:

What is truly worth worrying about is not the entire AI industry, but a single company.

In his view, since ChatGPT burst onto the scene in late 2022, OpenAI has effectively become the “credit anchor” for the entire generative AI era.

Investors are willing to believe that: AI will change the world; mega-scale data centers are worth building; GPU demand will grow at a high speed for the long term; large-model companies will eventually turn profitable; and AI startups can create enough terminal demand.

And in Zitron’s view, all of this rests on the premise that OpenAI continues to grow rapidly. He argues that OpenAI has not only defined the current AI boom, but also shaped the valuation logic that capital markets apply to the entire AI industry chain. Therefore, once this core assumption is broken, the shock could be far greater than that caused by a single unicorn company itself collapsing.

In other words, OpenAI is no longer just a company—it is more like a “systemically important institution” for the entire AI investment cycle.

Why does he think OpenAI’s business model has a fundamental flaw?

Zitron’s skepticism focuses on three areas.

First, inference (Inference) costs remain too high.

As the number of ChatGPT users continues to grow, every user question means steadily increasing GPU, electricity, and server costs. If a large number of users remain long term on low-priced or even free plans, while enterprise revenue growth cannot keep pace with costs, then expanding scale may actually mean losses widening.

Second, capital expenditures are far outpacing improvements in cash flow.

At present, the biggest spending in the AI industry is no longer model training; it is inference compute, GPU procurement, and the building of global data centers.

OpenAI and its partners are pushing data center investments on the scale of hundreds of billions of dollars or even more, and these projects typically take years to recoup costs. If future AI demand grows less than expected, many pieces of infrastructure could see their utilization rates decline.

Third, ongoing reliance on external financing.

Zitron’s analysis says he believes that for many years to come, OpenAI will still need continuous funding to cover spending such as model R&D, compute procurement, and infrastructure construction. If risk appetite in the capital markets declines or financing conditions tighten, its business model will face even greater pressure.

These views are still Zitron’s personal judgment and have not been endorsed by OpenAI, but they do reflect recent market debates around AI return on investment (ROI).

Why are Oracle, CoreWeave, and data center operators becoming focal points?

Compared with OpenAI itself, Zitron is more concerned about the leverage effect across the industry chain.

Over the past two years, the U.S. tech industry has seen an unprecedented data center construction boom.

Super large-scale cloud providers (Hyperscalers) such as Microsoft, Google, Meta, and Amazon have all increased capital expenditures. At the same time, companies such as Oracle and CoreWeave are taking on an increasing share of AI compute infrastructure buildout tasks.

These projects rely heavily on: long-term leases, project financing, private credit, corporate bonds, and large-scale capital expenditures.

If, in the future, demand from core customers such as OpenAI falls below expectations, or if the capital markets re-evaluate AI returns, then utilization rates of data center assets, lease agreements, and even financing capacity could all be affected.

The media points out that Zitron believes that once OpenAI suffers a major setback, companies that depend on growth in AI infrastructure demand—such as Oracle and CoreWeave—could be hit first. This is because the high valuations the market previously assigned to these companies were, to a large extent, based on expectations that AI demand would continue to surge.

Of course, at present, major tech giants including Microsoft, Meta, and Alphabet are still expanding AI capital expenditures and generally emphasize that AI infrastructure investment aligns with long-term strategy. As a result, there are no signs in the market of a broad pullback in capital expenditures.

Why are Anthropic and SoftBank also getting pulled into the discussion?

In addition to OpenAI, Zitron also directs his attention at Anthropic.

His reason is that although the two companies take different paths of development, they share a common trait: both need to continuously invest huge amounts of money to build models, purchase compute, and rely on large technology companies to provide computing resources and financing support. If the pace of AI commercialization ends up being slower than expected, both companies could face profitability pressure.

Another repeatedly mentioned company is SoftBank.

In recent years, SoftBank has returned to the forefront of large AI investments and has actively participated in fundraising for AI infrastructure, chip, and model companies.

If the AI industry enters a period of valuation adjustments in the future, SoftBank’s large portfolio of AI assets will naturally become an object of market attention as well. However, for now, SoftBank remains firmly committed to betting on the long-term development of AI and regards it as an important direction for the next wave of technological revolution.

Has AI trading already become overheated?

In fact, debates on whether AI has entered a bubble stage have been ongoing on Wall Street for more than a year.

Those who support the “bubble” view argue that:

AI infrastructure investment growth is far faster than revenue growth; the profitability model for large models has not yet been fully validated; data center capital expenditures have set historical records; and market valuations increasingly depend on growth expectations for the next several years.

Optimists, meanwhile, argue that AI is a typical general-purpose technology revolution. Like the internet and electrification, early-stage investment often far exceeds short-term returns, but in the long run it can create new industries and business models.

Howard Marks recently said he has shifted from initially doubting that AI might be just a bubble to now being more convinced of its long-term value. He believes the reasoning, contextual understanding, and interaction capabilities demonstrated by modern AI have unprecedented characteristics, so it cannot be simply compared to historical speculative bubbles.

Some academic research has also reached more neutral conclusions: the current AI market has both genuine technological progress and localized valuation overheating, as well as issues of capital expenditures being ahead of schedule. Therefore, it is closer to “a technology revolution stacked with a localized bubble,” rather than pure speculative frenzy.

What truly deserves attention is not whether OpenAI will fall

Whether or not people agree with Zitron’s assessment, the problems he raised are becoming a focus for an increasing number of investors:

When exactly can AI investment be converted into stable cash flows?

Over the past year, capital markets have almost assumed that higher AI capital expenditures are always better.

But recently, whether it’s chip stocks, server makers, or cloud computing companies, investors have begun to pay more attention to another set of indicators: enterprise AI revenue growth; AI product paid conversion rates; the rate at which inference costs decline; data center utilization; and the AI investment payback cycle.

If these indicators continue to improve, then the current massive capital expenditures may ultimately prove to be a forward-looking investment, similar to early investments in the internet era. But if, over the long term, the pace of commercialization keeps lagging behind investment expansion, then the market’s valuation logic for AI trading may also face recalibration.

So, what Ed Zitron’s long-form piece truly sparks discussion over is not “whether OpenAI will definitely become the next Lehman Brothers,” but that it once again brings the most core question of the AI era to investors: after capital expenditures keep setting new records, can cash flow and profitability actually keep up? The answer to this question may be what ultimately determines the true direction of global AI trading in the coming years.