3.7 trillion, the largest wave of AI IPOs in history is here, but the peak signal for the US stock market has also lit up
Recently, the global capital market has been waiting for three bells to ring.
SpaceX, Anthropic, and OpenAI are successively rushing towards IPOs.
If these three companies go public successfully, they will jointly create the largest round of financing wave in the history of technology:
The combined valuation of the three companies is close to $3.7 trillion, and the total planned fundraising scale is expected to be about $200 billion, which may rewrite the records of the US IPO market in recent years.
However, the more crazy it gets, the closer the risks become.
On June 5th, Bank of America Securities issued a warning: Approximately 70% of the bear - market signals have been triggered, and multiple valuation indicators of the S&P 500 are approaching historical highs.
In other words, the largest AI financing window is heading towards a market that may be close to its peak.
Is this IPO wave the golden moment for AI or the last carnival before the bubble bursts?
Behind the IPOs
In the past, when technology companies went public, the most concerned questions for investors were how fast the user growth was and how large the market space was.
But in the AI era, a new question has been put on the table: How much money do these companies actually need to spend to sustain their growth?
Currently, none of the three star companies rushing towards IPOs can answer when they will start making stable profits.
This is a structural dilemma in the business models of AI companies. Revenue growth does not necessarily bring about economies of scale.
Every time an AI model answers a question or executes a task, it consumes more computing power, and the inference cost also rises accordingly.
Judging from the publicly available data, OpenAI's financial pressure is the most obvious.
The company's annualized revenue has reached approximately $25 billion, but it is generally expected by the outside world that it will still incur a loss of about $14 billion in 2026. Multiple institutions predict that OpenAI will have a chance to achieve overall profitability at the earliest around 2029 - 2030.
In addition, OpenAI has committed to about $600 billion in future computing - power expenditures. This figure can only be self - consistent on the premise of continuous and high - speed revenue growth.
Anthropic's situation is relatively more optimistic.
According to CNBC, citing sources close to the company, its annualized revenue is about $44 billion, and it is expected that its operating profit will turn positive for the first time in the second quarter of 2026, reaching approximately $559 million.
However, operating profit does not equal net profit, let alone free cash flow.
For a model company that is still frantically purchasing computing power and building infrastructure, one or two quarters of profitability are not sufficient to prove that the business model has been successful.
SpaceX seems to be the closest to a mature enterprise.
In 2025, the company's total revenue reached $18.7 billion, but the net loss was still as high as $4.9 billion. The losses mainly come from two continuously money - burning areas: the Starship rocket project and xAI.
A Morningstar analyst publicly stated that there is an obvious over - valuation risk in SpaceX's current valuation, and xAI is regarded as an important uncertain factor that may erode the company's value.
Even in SpaceX's prospectus, it clearly warns that the company has a history of long - term losses, and it may not be able to achieve profitability in the future.
Although it is not uncommon for technology companies to go public without making profits, and the capital market is willing to pay a premium for future growth, investors never just look at the accounts of these three companies.
The GPUs purchased by OpenAI, Anthropic, and xAI come from NVIDIA; the computing services come from Microsoft Azure, Amazon AWS, and Google Cloud; and the data centers are connected to power companies, server manufacturers, and network operators.
Together, they form a huge AI infrastructure industrial chain, and this industrial chain is expanding at the same aggressive pace.
Therefore, the three AI companies' rush towards IPOs seems on the surface to be star enterprises entering the public market, but behind it is the entire AI industrial chain asking for money from the global capital market at the same time.
Accelerated Expansion of the AI Industrial Chain
But the question is, how much money is the entire AI industrial chain asking from the capital market?
The answer may be more exaggerated than most people think.
According to statistics from The Wall Street Journal, the cumulative bond - issuing scale of five technology giants, Google, Microsoft, Amazon, Meta, and Oracle, has reached $159 billion this year. The money mainly flows to data centers, chips, power, and networks.
What does $159 billion mean?
This has exceeded the annual fiscal revenue of many countries and is also close to the total annual capital expenditure of a group of super - large technology companies.
From an enterprise perspective, $159 billion is already the budget for a quasi - national - level infrastructure construction.
According to The Wall Street Journal, this debt - financing figure was $17 billion in 2024 and jumped to $108 billion in 2025.
It has nearly increased nine - fold from $17 billion to $159 billion in two years.
The risk of the AI financing wave lies not only in the huge absolute amount but also in its extremely fast acceleration.
This is different from the previous round of Internet expansion.
Internet companies' money - burning mainly occurs in traffic, subsidies, market, and human resources, and many costs can be reduced.
But AI infrastructure requires heavy - asset investment.
Once GPUs are purchased, data centers are built, power contracts are signed, and networks are laid, it is difficult to make a quick turn.
It is more like railway, telecommunications, and energy infrastructure, with huge upfront investment, a long return cycle, and whether it can make money in the end depends on whether there is enough computing - power demand later.
However, $159 billion is not the entire AI - related investment of these five companies this year.
Introl predicts that the combined capital expenditure of these five companies is expected to exceed $600 billion in 2026, a 36% increase from 2025, and about 75% of it will be directly used for AI infrastructure.
The predicted capital - expenditure - to - revenue ratio of these companies in 2026 ranges from 30.8% to 57%. Historically, only capital - intensive public utilities and telecommunications companies have had such a ratio.
For reference, AT&T's capital - expenditure - to - revenue ratio was 23.5% at the peak of the telecommunications bubble in 2000.
Currently, according to institutional predictions, Oracle's ratio is about 57%, Meta's is about 50%, Microsoft's is about 38%, Google's is about 44%, and Amazon's is about 30.8%.
These technology giants, once known for their light - asset operations, are completely transforming into infrastructure operators.
But there is even greater pressure ahead.
BCA Research estimates that the five ultra - large - scale technology companies plan to add about $2 trillion in AI - related assets to their balance sheets by 2030.
Calculated at an annual depreciation rate of about 20% for AI assets, the annual depreciation expense will reach about $400 billion at that time, exceeding the combined total profit of these companies in 2025.
That is to say, even if the AI demand grows as expected, the accounting profits of these companies will be significantly eroded by depreciation around 2028 - 2030.
If the demand fails to meet expectations, depreciation will turn from a paper loss into real financial pressure.
The Financing Window May Be Reaching Its Peak
The funds in the capital market are not infinite.
The timing chosen by the three AI companies to rush towards IPOs may not be a bull market without any shadows.
Just as the market is discussing that Anthropic, OpenAI, and SpaceX may jointly create the largest financing wave in the history of technology, Bank of America Securities is also reminding investors that the risk signals in the US stock market are rapidly piling up.
The Bank of America strategy team said that 70% of the bear - market signals have been triggered, and this proportion has approached the typical level near the market peak in history.
According to Bank of America's criteria, 17 out of 20 valuation indicators of the S&P 500 show over - valuation in a statistical sense; among them, 8 valuation levels are even higher than the peak during the 2000 technology bubble.
This means that today, when AI companies want to go public and raise funds, they are facing not a cheap market but a market that has fully priced in future expectations.
The structural problem is also obvious.
This round of the US stock market's rise is increasingly relying on a small number of AI and semiconductor companies.
The AI and chip sectors have contributed the majority of the S&P 500's gains this year, but sectors such as finance, healthcare, and consumption have begun to weaken.
That is to say, the index is still at a high level, but the forces supporting the index are becoming more and more concentrated.
Once the AI trading cools down, there are not enough sectors in the market to take over.
More importantly, AI infrastructure is in turn squeezing the cash flows of the technology giants themselves.
Reuters previously cited market estimates that in the next two years, AI capital expenditure may absorb about 94% of the operating cash flows of large technology companies.
Compared with the model in which cloud providers self - expanded through operating cash flows in the past few years, today's AI construction has increasingly relied on debt, equity financing, and the public market.
This is the most delicate part of the IPOs of these three AI companies.
They are not going public in an infinitely optimistic market but are seizing the window in a market with high valuations, concentrated gains, and rising capital costs and capital expenditures.
A large number of companies are waiting for this door to open. Model companies, cloud providers, chip companies, and data - center companies all need money.
The most dangerous moment for the capital window is often when everyone thinks they must catch this train.
Prosperity or Bubble?
Where will this AI financing wave ultimately lead?
In the 19th century, the construction of railways in the United States also relied on bond financing. Many railway companies later went bankrupt, and investors suffered heavy losses, but the railway network remained and ultimately reshaped the US economy.
If the demand for Agents truly explodes in the next two or three years, today's seemingly radical data - center construction may become the infrastructure for the next - generation economy.
Of course, it could also be a telecommunications - style bubble.
In the late 1990s, telecommunications companies also believed that Internet traffic would grow infinitely, so they frantically borrowed money to lay optical fibers.
As a result, the traffic growth failed to digest the production capacity in time. A large amount of optical fiber was left idle for a long time, companies went bankrupt, and shareholders were liquidated.
The infrastructure ultimately survived, but much of the capital that bet on it making quick money did not.
If the demand for Agents comes too slowly, today's AI data centers may become an optical - fiber - style bubble.
Currently, the AI industry is at a similar crossroads.
The question is, is the real demand sufficient to absorb these capital investments?
Bain & Company's estimate is that for the current batch of data centers to be financially viable, the global AI - related revenue must reach $2 trillion per year by 2030.
Some institutions estimate that the actual figure in 2025 is less than $20 billion.
A hundred - fold increase in six years is an implicit premise in the entire AI investment logic.
Meanwhile, McKinsey's survey data shows that 88% of enterprises are already using AI in at least one business module, but less than 40% of enterprises have scaled up AI deployment beyond the pilot stage.
Most enterprises' current use of AI is still limited to light - level scenarios such as auxiliary writing, summary generation, and code completion, and the computing - power consumption is far from sufficient to support the current scale of data - center construction.
The current reality is that the supply is being built, but the demand is still on the way.
The AI industry is telling two stories at the same time: one is the next industrial revolution, and the other is the next capital bubble.
No one knows which one will come first.
This article is from the WeChat official account “World Model Workshop”. Author: World Model Workshop. Republished by 36Kr with permission.