Lately, whenever I open my phone, the group chat is mostly discussing these topics:

NVIDIA has hit a new high, the US stock market has also reached a historical high; the memory sector has surged, with Micron more than quadrupling this year and Intel achieving its strongest weekly gain since 2008, even the storage sector in the A-share market is taking off.

At the same time, group members are discussing "what is the next target to get in on?" and "Is this a replay of the dot-com bubble at its peak?"

It sounds contradictory, but it's actually the same sentiment—fear of missing out, yet fear of a crash.

However, in reality, the "bubble" we are currently discussing is likely not the true bubble of this AI wave. Or more accurately, the most dangerous part of this AI bubble is not where you can see when you open your trading account.

A few days ago, OpenAI was revealed to have arranged a stock sale for employees in October last year. 75 people cashed out at the maximum limit of $30 million, while over 500 other employees took home an average of about $6 million each. Originally, the company intended to raise $6 billion, but due to too many external investors, this was temporarily increased to $10.3 billion. This round of funding valued OpenAI at $500 billion, more than three times its valuation six months ago.

This event took place in October last year, but most people only found out about it in May this year. Without the report by The Wall Street Journal, many people might still be unaware. And during these 7-plus months, OpenAI's valuation has surged from $500 billion to $852 billion, a further 70% increase.

While the memory sector is booming and NVIDIA is hitting new highs, all of this is true, but they are not the most dangerous part of this AI bubble. The real bubble is increasingly occurring where you cannot see or buy into.

This time, it's not that ordinary people didn't see the bubble. It's that when they did see the bubble, the most critical trades had already concluded.

Once the Valuation Surge is Over, You Might Not Even See It

Yesterday, OpenAI issued a statement on its website, stating that OpenAI's equity cannot be traded privately, and any unauthorized transfer or pledge is invalid. The announcement specifically banned several products: selling equity to investors through a shell company, converting equity into crypto tokens and selling them on the blockchain, as well as using "forward contracts" to promise buyers a share of the profits once OpenAI goes public.

If we compare this to the dot-com bubble of 2000, the biggest difference is that when the bubble burst, Google, Amazon, Yahoo, and various .com companies were already listed, allowing retail investors to directly purchase shares of these companies with P/E ratios of 100x, 200x. The bubble formed in the public market and also collapsed in the public market.

OpenAI is now valued at $852 billion, up from $157 billion a year and a half ago. Anthropic is valued at nearly $900 billion, up from $615 billion a year ago, more than a tenfold increase. xAI, founded just 3 years ago, is now valued at $2.5 trillion, while Databricks saw its valuation surge from $620 billion to $1.34 trillion in a year. However, these skyrocketing numbers, faster than a rocket's ascent, have not emerged from the public markets.

This AI bubble cycle is happening in a space where the public cannot participate.

When anxiety can't find an entry point, it seeks alternatives. Recently, there were numerous media reports of Anthropic surpassing a $12 trillion valuation and overtaking OpenAI. This number came from a blockchain-based decentralized pre-IPO platform that packaged Anthropic's equity into tradable synthetic assets (the type of trading explicitly prohibited by OpenAI). However, the platform's actual trading volume is less than $1.4 million in 24 hours, with only over 300 participants.

Users are not buying actual Anthropic common stock but a kind of "anxiety exposure." The $12 trillion valuation is not Anthropic's true value; it's more like an outbreak of AI anxiety at a liquidity breaking point. Silicon Valley big shots understand this anxiety too well; they even hope for more significant outbreaks of anxiety, enabling them to sell more anxiety products.

Last month, Silicon Valley's most renowned investor, Naval, launched a "retail fund" called USVC, aiming to allow ordinary people to invest in AI companies. The fund's portfolio includes shares of the hottest AI companies like OpenAI, Anthropic, and xAI, allowing non-accredited investors to buy in starting at $500.

However, this is a closed-end registered fund; its shares are not traded on exchanges, with a quarterly repurchase limit of 5%, and the board can decide not to repurchase. Upon careful review of the prospectus, you will find that it encourages investors to "consider the shares as illiquid assets," prompting many on social media to criticize it as a "dumping fund."

The surge in the memory module sector also follows a similar logic. With leading companies like Mag 7, especially NVIDIA, being too expensive and unavailable, investors are looking elsewhere along the AI supply chain: chips, memory, power, and even helium, copper, and silver.

What you see and discuss as a bubble in the public markets is actually liquidity anxiety spilling over from the private market.

Cashing Out Like Breathing, Exiting Before Even Waiting for an IPO

In the old Silicon Valley, hard tech companies required employees to wait 7 to 10 years to cash out, either by enduring until an IPO or waiting for acquisition by an industry giant. In the post-internet era, this cycle has been compressed to around 5 years, with options unlocking, secondary market transfers, the IPO lockup period, and wealth distribution now having multiple nodes, but IPOs still being the biggest one.

As we entered the AI era, cashing out has been completely front-loaded to the pre-IPO stage.

This time, OpenAI set the stock sale threshold for employees at only two years. ChatGPT was released in November 2022, and the incoming employees began unlocking their stock sale eligibility in the second half of 2024. This group of people coincided perfectly with the $6.6 billion cash-out in October last year.

It's not just within OpenAI. The founders and core teams of AI companies are all using a new way to exit early, without needing to be acquired or IPOed.

In 2024, Google acquired Character.AI, which, in the old Silicon Valley, would not have been a true acquisition. Google didn't buy the entire company; instead, they spent $3 billion to acquire the rights to use Character.AI's technology, with $2.5 billion allocated to distribute to Character.AI's existing shareholders and the remaining $500 million as a technology licensing fee.

In simple terms, it's technology licensing plus team migration. The company itself is still there, but the most valuable people and the most critical technology have already exited in a single private transaction. The two co-founders of Character.AI, holding over 30% of the company, could receive nearly $1 billion just from this deal.

Similarly, Microsoft acquired Inflection AI for $650 million, brought the technology over through a licensing agreement, and directly hired the founders and core team. Amazon also used this method to acquire Adept AI.

By early 2025, the Federal Trade Commission (FTC) had initiated an investigation into these types of transactions, focusing on whether large companies were using this structure to circumvent merger reviews. However, the "acquisitions" mentioned earlier all took place in 2024, without regulatory scrutiny, and without the need to list names in a prospectus.

Looking at it from a primary market perspective, today's AI doesn't even need to be compared to the internet bubble of the past because the hype has long exceeded it by several orders of magnitude.

For any AI startup, funding rounds start at several billion dollars. Most importantly, the team and founders don't need to wait for an IPO to exit. The money in the private market is already substantial, and more of this money is finding its way into the pockets of employees and founders in increasingly covert ways.

Prior to OpenAI's employee stock sale in October last year, they had conducted two similar internal transactions. Large unicorns like Anthropic and Databricks have also done the same. AI companies no longer need to wait for an IPO; they have a "liquidity window" every so often.

Founders also have their own channels. Silicon Valley is now seeing a trend in "founder-led secondary" transactions, where entrepreneurs sell a portion of their equity without leaving the company. This way, they can benefit from the continued increase in company valuation and receive cash upfront.

Alternatively, they can opt for equity-backed loans. There is a company called Pluto that specializes in this, helping AI founders and early investors use their private stock as collateral to receive cash, with a loan-to-value ratio of 20% to 35%. This way, they can get cash without selling stock.

Early-stage investors do not have to wait for the company to IPO to cash out for their limited partners (LPs). They can establish a new fund using the original VC to sell the star assets from the old fund to the new fund. The old LPs can choose to cash out and exit, or they can continue to hold with the new fund. This method is called a "GP-led continuation fund," and the size of such transactions in the first half of 2025 was close to $50 billion, doubling from 2024.

Another indirect exit route is through starting new ventures. At least 7 unicorn companies have been founded by people who left OpenAI, including Anthropic, Thinking Machines Lab, and SSI. With the original team leaving, regrouping, and raising funds again, each departure triggers a new round of wealth distribution.

Each of the exit methods mentioned above does not require regulatory scrutiny or the need to disclose valuations in an IPO prospectus. AI has emerged as the biggest beneficiary because a large number of high-quality AI assets are currently unable to IPO.

AI Infrastructure, More Like a Real Estate Bubble

Many people compare today's AI to the Internet in 2000, but that's a flawed analogy. The current AI bubble is actually more akin to the real estate bubble in 2008.

During the 2008 subprime mortgage crisis, the houses were real, the rent was real, but the house prices, mortgages, ratings, securitization—everything was based on an excessively optimistic outlook. When Lehman collapsed, the bonds packaged with home loans became worthless.

Now, a similar financialization is happening in AI data centers, GPUs, and computing power contracts, but on a larger scale.

AI training and inference require data centers, which in turn need land, power, water, cooling, networking, and long-term customers. Therefore, data centers are no longer just the back-end machine rooms of tech companies but assets contended for by real estate funds, private credit, and insurance funds.

Last year, Meta announced a partnership with Blue Owl to develop the Hyperion data center in Louisiana, with a total development cost of $27 billion, which is nearly enough to build 30 Shanghai Center towers. Blue Owl's managed funds hold 80% of the project, with a significant portion of the funding raised through private debt issuance. Meta holds 20%, contributing the land and ongoing construction, and then signs a 4-year operational lease with the joint venture, along with a 16-year residual value guarantee. If the lease is not renewed at the end of the term, Meta would incur a loss based on the data center's value at that time.

Meta did not simply say, "I'm going to spend $27 billion to build a data center"; instead, it turned the data center into a joint venture, transformed capital expenditures into leases, residual value into guarantees, and then sold a portion of the project's debt to private bond investors. This logic is strikingly similar to the packaging of mortgages into financial derivatives in 2008.

CoreWeave is another example. In 2023, it completed a $23 billion debt financing using Nvidia chips as collateral. In 2024, it signed another $75 billion debt financing round led by Blackstone. In 2026, it completed an $85 billion GPU-backed financing, obtaining an investment-grade A3 rating from Moody's, making it the first-ever investment-grade rated GPU-backed financing.

And it's not just CoreWeave. This year, Lambda completed a $10 billion senior secured credit; Crusoe secured a $750 million credit from Brookfield, in addition to $11.6 billion for building OpenAI's Stargate computing power plant; and Broadcom reportedly is in talks with Apollo and Blackstone for a $35 billion AI chip financing.

Each of these transactions is transforming AI computing assets into financeable, securitizable credit products.

Regulators have already given a name to this phenomenon. In its 2026 report, the Bank for International Settlements referred to this structure as "shadow borrowing." Tech giants hold data center assets through joint ventures and SPVs, taking on debt through long-term leases and guarantees, but these debts are not recorded on the company's balance sheet. They borrow money to buy GPUs to build data centers while waiting for GPU depreciation. Moreover, the borrowed money has a long term, but GPUs depreciate quickly.

The bubble risk on this path actually didn't need to wait for the AI wave to validate it. The recent private equity fund explosion was like a preview.

In 2020, the private equity fund Vista Equity bought a SaaS company called Pluralsight for $3.5 billion. The debt holders who lent to it were all top players in private credit, including Blue Owl, Ares, Goldman Sachs, and BlackRock. By 2024, Pluralsight couldn't hold on anymore, so Vista had to "hand over" the entire company to the debt holders, resulting in a $4 billion loss for itself and co-investors.

The reason for not being able to hold on was not "how much money the company is making now," but "how stable the company's future subscription renewal revenue will be." When AI changed the renewal logic of the software market, all "seemingly stable cash flows" needed to be reinterpreted. The moat of SaaS private credit suddenly turned from water to sand.

Blue Owl, one of the top players in private credit, lent to Pluralsight. Earlier this year, its OCIC's private credit fund was redeemed by retail investors by 40% due to AI impacting SaaS. However, Blue Owl still lent to AI data centers. Apart from Meta's data center mentioned earlier, it is also a major blood supplier behind OpenAI's Stargate computing power project.

The most dangerous aspect of private credit is its opacity, which leads to widespread valuation distortions. The underlying assets of the fund are impossible for external investors to verify.

In August last year, HPS, the private credit division of BlackRock, was cheated out of over $400 million by an Indian-origin telecom entrepreneur using forged invoices. HPS lent to several telecom companies owned by this entrepreneur, with the collateral being these companies' accounts receivable. It wasn't until an HPS employee noticed an issue with the customer email addresses that the entire collateral was found to be nonexistent.

Even top players of BlackRock's scale like BlackRock can't clearly see if the money they lent out has real collateral. How much can the investors who buy its fund shares know?

All of this AI data center financing, GPU collateralized loans, and new SPV structures are built on one assumption: that the underlying assets are valuable.

But how fast do GPUs depreciate? Will data center customer contracts be renewed? Will AI inference demand materialize to support this computing power? These questions, even rating agencies assessing the assets and banks underwriting the funds, can only provide judgments "based on existing information." What ordinary investors see is just a prospectus, a rating report, and a name.

The Real Bubble Doesn't Necessarily Quote You First

Going back to the initial question, "What is the next target that one can get on board?"

Currently, what most people can get on board with is actually the shadow cast by the overflow of core assets. In the 2000 dot-com bubble, the peak was in the public market, and the crash was also in the public market. You could see it, feel it, and read about it in the news that day.

This time, the most bubbly and most dangerous part is happening where you can't see. By the time you see these, the most critical trades have already concluded.  

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia