Recently, an unusual phenomenon has emerged in the US stock market.

The primary market continues to assign sky-high valuations to AI companies, but when it comes to going public, investors in the secondary market are starting to balk.

From mid-September to mid-October, in less than a month, at least four companies related to the AI industry chain or AI investment narratives announced postponements or withdrawals of their listings, covering smart hardware, AI computing power, data centers, and nuclear power.

On October 9, Firmus, an Australian AI data center company with a 7.2% stake held by NVIDIA, officially withdrew its IPO.

The company had secured support from investment institutions such as NVIDIA, Blackstone, and Coatue, with clients including OpenAI and Meta. Logically speaking, it had almost all the plus points—AI concept, star-studded shareholders, and major clients.

Yet, it still failed to complete its IPO.

Going back further, Oura, a smart ring manufacturer, also suddenly announced the postponement of its IPO on the day the issue price was to be finalized. The core reason was a failure to agree on valuation.

On one side, valuations of AI assets in the primary market continue to soar; on the other, the secondary market is starting to say 'no' to these high valuations. The pricing logic for this round of AI assets is undergoing subtle changes.

Today, Silicon-Based Gentleman will discuss this matter.

/ 01 / The Secondary Market Is Not Buying 'AI New Listings'

Among the AI companies hitting the brakes on IPOs, the most well-known is the smart ring manufacturer Oura.

On September 21, Oura officially kicked off its IPO roadshow, planning to raise up to $2.2 billion, corresponding to a fully diluted valuation of approximately $15.6 billion.

The company is not lacking in performance. Over the past 12 months, Oura achieved revenue of $1.21 billion, up 74% year-on-year, with expected revenue growth of 90% in fiscal 2026, and it has already turned a profit.

According to Reuters, Oura's IPO had already secured about four times oversubscription at that time. However, on September 29, the day originally planned to finalize the issue price, Oura suddenly announced the postponement of its IPO.

A similar situation occurred in the AI infrastructure sector.

On October 9, Firmus, an Australian AI data center company with a 7.2% stake held by NVIDIA, officially withdrew its IPO.

The company had secured support from investment institutions such as NVIDIA, Blackstone, and Coatue, with clients including OpenAI and Meta. Logically speaking, it had almost all the plus points—AI concept, star-studded shareholders, and major clients.

Previously, the company planned to raise about 7 billion Australian dollars, corresponding to a valuation as high as 44 billion Australian dollars. To attract investors, Firmus once considered lowering its issue price from 11 Australian dollars to 8.25 Australian dollars but still failed to secure sufficient support.

A few days later, on October 13, AirTrunk, a data center operator under Blackstone, also transmit [can be translated as 'reported'] the news of suspending its Singapore REIT listing.

Compared to Firmus, AirTrunk's business is much more mature. As early as 2024, Blackstone acquired the company at an enterprise value exceeding 24 billion Australian dollars, and its data centers already had over 800MW of customer-committed capacity.

Even so, AirTrunk's planned listing transaction, which originally aimed to raise about $1.5 billion, may still be postponed until 2027.

More noteworthy is that this change has already spread along the AI industry chain to the energy sector.

On September 17, US nuclear power company Holtec announced the postponement of its IPO and officially withdrew its listing application on September 25.

In its statement, Holtec directly mentioned that the nuclear energy sector had recently encountered an unfavorable market environment, with one significant reason being the uncertainty surrounding the prospects of AI data center construction.

From smart hardware to computing power infrastructure, then to data centers and nuclear power, the businesses of these companies vary greatly, yet they almost simultaneously encountered resistance to going public. This reflects a noteworthy change:

The secondary market is becoming more cautious about valuations of AI assets.

This scene also appeared during the dot-com bubble 26 years ago.

On March 10, 2000, the Nasdaq Composite Index hit an intraday high of about 5,132 points before beginning a sustained decline.

A month later, the US IPO market suddenly saw a large-scale wave of withdrawals.

On April 19, 2000, Wired published a report titled 'More Delays in IPO Land.' Of the 18 companies originally scheduled to go public that week, only one had completed its issuance as of the report.

Among them was AltaVista, one of the most well-known internet search engines at the time.

When AltaVista had most of its equity acquired by internet investment company CMGI in 1999, the transaction value was about $2.3 billion. In April 2000, AltaVista originally planned to raise $280 million through its IPO.

How popular was AltaVista at the time?

According to data from internet traffic monitoring agency Media Metrix, in December 2000, AltaVista still had about 19 million unique visitors, ranking 11th among the most visited websites in the United States.

Besides search, it also had a highly popular translation product, Babelfish, which processed over 1 million translation requests per day.

But despite being an internet star with a massive user base, AltaVista missed the last train for its IPO.

In April 2000, just before its official issuance, the Nasdaq began to plummet, and AltaVista suddenly announced the postponement of its listing. Subsequently, as the dot-com bubble continued to burst, this search giant ultimately failed to successfully enter the capital market.

And AltaVista was not an isolated case.

At the time, internet companies such as Coolsavings and Yupi Internet also postponed their IPOs, and even the most sought-after internet infrastructure enterprises began to be forced to lower their prices.

For example, 360networks, which built transnational fiber-optic networks, originally planned to raise $860 million but had to reduce its target fundraising scale to about $600 million before completing its issuance.

This change then quickly spread.

According to IPO market statistics at the time, in the first quarter of 2000, only 20 IPOs were withdrawn or postponed in the United States. By the second quarter, this number suddenly increased to 85. Companies lowering their issue price ranges also rose from 8 to 32.

It is worth mentioning that the IPO withdrawal wave in 2000 occurred after the Nasdaq peaked and began to plummet.

In this round, AI assets have obviously not reached that stage yet. However, at least for some AI assets, the high valuation expectations in the primary market have begun to diverge from the pricing of secondary market investors.

/ 02 / The Premium for Long-Term Growth Disappears

Behind this divergence are two main reasons.

First, the valuation gains of AI are being cashed in by the primary market ahead of time.

Over the past few years, a very clear trend in the global investment and financing market is that the steepest part of valuation increases for tech companies is increasingly happening before going public.

Two sets of data are worth noting.

In its US private market report released in July 2026, Cambridge Associates disclosed that in 2025, the return rate of the US venture capital index reached 21.1%, the best performance since 2021.

On the other hand, data from Renaissance Capital shows that in 2025, for US-listed tech companies, based on the issue price to the closing price on December 17, the average return was -7.3%.

Even when looking only at tech IPOs with a fundraising scale exceeding $100 million, by the end of the year, only about a quarter were still trading above their issue prices.

Of course, the statistical caliber [can be translated as 'criteria'] of these two sets of data are not entirely consistent, but the market contrast they reflect is still noteworthy.

In the past, IPOs were often an important starting point for value growth in tech companies. Nowadays, however, more and more star tech companies have completed most of their valuation expansion in the private market.

In the AI field, this trend is especially evident.

In September 2025, Firmus' private valuation was still only 1.85 billion Australian dollars. By August 2026, with the participation of institutions such as NVIDIA and Blackstone in financing, its valuation quickly surged to about 15 billion Australian dollars.

But just two months later, Firmus proposed a valuation of 44 billion Australian dollars in its IPO. In other words, in just one year, Firmus' valuation skyrocketed nearly 23 times.

Even after the company later considered lowering its issue price by 25%, the corresponding valuation would still reach 33 billion Australian dollars, more than double the previous round of private valuation.

Oura was no slouch either. At the end of 2024, Oura's private valuation was about $5.2 billion. By October 2025, this figure had risen to $11 billion. And when planning its IPO, the company hoped to further increase its valuation to $15.6 billion.

In less than two years, the valuation had roughly tripled.

This means that for these companies, multiple rounds of financing before going public have already completed a significant portion of their valuation expansion.

When the primary market cashes in on valuation gains years ahead of time, the potential return space left for secondary market investors is naturally squeezed.

Second, increased volatility in AI assets has led the secondary market to reassess the value of long-term growth.

In the second quarter of 2026, driven by the rebound in tech stocks and the AI investment boom, the Renaissance IPO ETF surged 42%. The strong market performance also led many companies preparing to go public to raise their issue price expectations.

However, entering the third quarter, as market volatility increased, investors began to demand that valuations truly match revenue structures, profitability, and capital investments.

Oura is a typical case. Based on the previously proposed issuance range, Oura's corresponding valuation was about $12.8 billion to $15.6 billion, equivalent to about 9 to 11 times its revenue over the past 12 months.

Such a valuation is not uncommon for a high-growth software company.

But Oura's actual revenue structure is still highly dependent on hardware sales.

According to its prospectus, in the nine months ending June 2026, Oura's subscription revenue was about $240 million, accounting for about 20% of total revenue, with the remaining 80% mainly coming from smart ring sales.

This means that although Oura has established a subscription business, it is still primarily a hardware sales company overall.

Hardware and software often correspond to two different valuation logics in the capital market.

Software subscriptions typically have strong revenue predictability and scale effects, while hardware requires ongoing costs for production, inventory management, and product iteration. This is the core reason for the valuation divergence in Oura.

Firmus faces a similar test.

Although Firmus proposed a construction plan for AI data centers with a capacity of up to 1GW, according to Reuters Breakingviews data, its actual operational capacity at the time was only 42MW, less than 5% of the target scale.

In other words, more than 95% of the planned capacity had not yet been put into operation.

This means that investors buying in at the proposed IPO valuation of 44 billion Australian dollars would largely be paying in advance for the company's construction and growth over the next few years.

Not only that, but to actually build the planned data centers, Firmus would need to continuously raise massive amounts of capital.

Morningstar strategist Lochlan Holloway believes that according to the company's projections, once all data centers are completed, Firmus' debt could reach $30 billion, about six times its projected operating income for 2028.

For a company with a relatively small actual operational scale, investors not only need to believe that AI computing power demand can continue to grow but also that the company can build data centers as planned, secure financing, and ultimately achieve expected profits.

Any issues in any of these links could affect the final investment return.

The valuation divergences in Oura and Firmus to some extent reflect that market concerns about risks are beginning to outweigh the fear of missing out on AI opportunities.

/ 03 / The Interest Rate Hike Cycle Also Adds Pressure

Besides market volatility, rising interest rates have further amplified this contradiction.

On September 16, the Federal Reserve announced a 25-basis-point interest rate hike, raising the target range for the federal funds rate to 3.75%—4.00%.

For AI companies reliant on long-term growth, rising interest rates mean two types of pressure.

On one hand, higher risk-free rates and market-required return rates reduce the value of future cash flows discounted to the present. The more a company relies on profits years down the line to support its valuation, the more pronounced the impact usually is.

On the other hand, for asset-heavy enterprises like AI data centers, rising interest rates also directly increase financing pressure.

Building data centers requires significant upfront capital investment, and companies often need to maintain expansion through debt financing. When borrowing costs rise, the expected investment returns originally anticipated may also be eroded by increasing interest expenses.

Interestingly, the bursting of the dot-com bubble 26 years ago also occurred during an interest rate hike cycle.

According to Federal Reserve historical data, from June 1999 to May 2000, the Federal Reserve raised interest rates six consecutive times, increasing the federal funds target rate from 4.75% to 6.5%, a cumulative increase of 175 basis points.

The Continuously tightening [can be translated as 'continuously tightening'] monetary environment raised capital costs and put pressure on internet companies that relied on long-term growth expectations to support their high valuations at the time.

Of course, today's AI industry is very different from the internet in 2000, and the degree of Federal Reserve tightening is far from reaching the levels of that year. A few companies' IPO setbacks are not enough to conclude that the entire AI bubble is bursting.

Of course, there are still significant differences between today's situation and that of 2000. A few companies' IPO setbacks are not enough to conclude that the entire AI bubble is bursting.

However, the consecutive cooling of AI asset IPOs also exposes an important change:

While AI's industrial growth may be far from over, the growth premium that capital markets are willing to pay in advance has already begun to shrink.

By Qi