Three names quietly appeared on a shareholder list.

On May 29, a significant event occurred in the AI circle.

Anthropic, the developer of Claude, completed a Series H financing round of $65 billion. After the investment, its valuation soared to $965 billion, officially surpassing OpenAI and becoming the world's most valuable AI startup (Source: Cailian Press reported on May 29, "Anthropic Becomes the World's Most Valuable AI Startup! Official Announcement of Valuation Rising to $965 Billion").

The headlines of the media are all shouting about the figure of $965 billion.

But this article today is not about that figure.

What we're going to talk about today is that there are three unusual names quietly added to the shareholder list of Anthropic's latest financing.

Micron. Samsung. SK Hynix.

These three names hardly ever appeared together on the shareholder list of any AI company before. They are the "Big Three" in the global memory chip industry, and together they control more than 90% of the world's HBM (High-Bandwidth Memory) production capacity.

They all invested in an AI company today (Source: The Paper reported on May 29, "The Three Memory Giants Collectively Invest in Anthropic").

This is not a coincidence. It's a change in the industrial chain structure.

Why Did These Three Names Come Together?

To answer this question, we first need to know the usual state of these three companies.

Micron, Samsung, and SK Hynix are the world's three major oligarchs in the DRAM industry. In the past 30 years, they have been bitter rivals: competing in production capacity, technology, and price. The one with the lower price wins the sales. This is a typical strong-cyclical industry, and no one will give in to the other voluntarily.

But HBM is different.

HBM is the "special memory" in the AI era, specifically designed to support GPUs. A high-end AI chip (such as NVIDIA's H100 and B200) requires dozens of HBMs to operate around it, which is equivalent to the "peripheral organs" of AI computing power.

What's even more critical is that the production capacity of HBM is extremely scarce. The EUV lithography machines are in short supply, and the wafer consumption of one HBM is three times that of ordinary DRAM. As a result, Micron officially said that the HBM production capacity for 2026 "has been fully sold out," and the order fulfillment rate for some customers is only 50% to 70% (Source: The Paper reported on May 29).

Put simply: The supply of HBM is currently less than the demand. Deciding who to supply and who not to is a strategic issue.

So, the reason these three companies jointly invested in Anthropic this time is not about the financial return of Anthropic, but about something else.

They are tying themselves to the biggest "money-devouring beast" customer in the next five years in advance.

What Are They Betting On?

It sounds like "the shovel sellers are investing in the gold diggers" when memory giants invest in AI companies.

Is the logic reversed? No. This is actually the smartest way to bet.

First, Lock in Orders: Anthropic's ARR increased from $9 billion to $47 billion in five months.

Second, Smooth Out the Cycle: Replace the uncertainty of the strong cycle with the certainty of long-term contracts.

Third, Upgrade Status: Upgrade from selling standard parts to becoming a strategic partner of AI companies.

(Data source: Cailian Press and The Paper reported on May 29)

First, Lock in Orders. Anthropic's Annual Recurring Revenue (ARR) soared from $9 billion to $47 billion in five months (Source: Cailian Press reported on May 29). This growth rate devours an enormous amount of computing power. Investing in advance is equivalent to signing orders in advance.

Second, Smooth Out the Cycle. The biggest pain point in the memory industry is its strong cyclicality. Prices skyrocket during the boom and plummet during the recession. Tying up with a long-term money-devouring customer is equivalent to replacing the uncertainty of the "spot market" with the certainty of "long-term contracts."

Third, Upgrade Status. Upgrade from being a "standard parts seller" to a "strategic partner of AI companies." The latter has completely different say in the industrial chain.

Mizuho Securities provided a set of data: The global HBM market size will grow from $35.9 billion in 2025 to $246.1 billion in 2028, with a compound growth rate of about 90% (Source: The Paper reported on May 29).

What does this mean? It means that in the next three years, a significant portion of the money from the growth of global AI computing power will flow into the pockets of these three companies first.

What they are doing now is to lock in this pipeline in advance with capital.

The memory giants are not "shovel sellers," but "people who hide shovels in the inventory in advance."

Is the $965 Billion Valuation a Bubble?

Let's go back to the figure of $965 billion.

Anthropic currently has an annualized revenue of $47 billion. Dividing the $965 billion valuation by $47 billion, the price-to-sales ratio (PS) is about 20 times.

In the traditional software industry, a 20-fold PS is usually considered a bubble. But Anthropic is not a traditional software company.

Its gross margin of inference infrastructure has jumped from 38% a year ago to 70% now, and it is expected to achieve profitability for the first time in the second quarter of 2026 (Source: Cailian Press reported on May 29). 85% of its customers are enterprise customers, not the consumer market. The annual revenue of its single product, Claude Code, has reached $2.5 billion.

What's even more crucial is another thing.

A significant portion of Amazon's net profit in the first quarter of 2026 came from the floating profit of its equity investment in Anthropic (Source: 36Kr reported on May 12, "The Narrative of Technology Stocks Is Increasingly Dependent on Anthropic").

What does this mean? It means that Anthropic's valuation not only reflects its own profitability but also bears the "computing power premium" and "strategic position" of global technology giants in the AI era.

It's not just an AI company raising funds. It's several global giants jointly pricing the entire AI infrastructure era through it.

The following is an inference for readers' reference: The traditional price-to-sales ratio valuation model may have become ineffective in the case of the "AI operating system."

The Other Side of This Gamble

But every bet has a downside.

First, Will the HBM supply suddenly become excessive? Currently, the scarcity is due to the supply not keeping up with the demand. But after 2027, all three manufacturers are expanding production. It's not impossible for HBM to go from being a "bottleneck" to a "price war."

Second, Can Anthropic's explosive growth continue? A five-fold increase in ARR in five months is a historic speed, but all exponential curves will eventually return to the linear range. Once the growth rate slows down, the support for the valuation will be very fragile.

Third, Will the intensity of the AI competition between China and the United States change the situation? SK Hynix and Samsung are both in South Korea, and the key equipment and upstream materials of HBM are under the regulatory scope of multiple countries. This is an industrial issue, but the answer lies outside the industry.

There are no answers to these three questions. But they suggest one thing: $965 billion is not the end. It's the middle of a long-term gamble.

It's Both a Bet and a Treasure

Let's go back to that shareholder list.

The media tells today's story as an "AI company valuation miracle." But a more accurate version is: The power in the global AI industrial chain is being nailed down in advance by several companies with capital.

Whoever can lock in the HBM production capacity will have the power to "transfuse blood" to AI companies.

Whoever can get a share of Anthropic will have the "pricing power" in the AI era.

What these three memory giants did on May 29 was not a financial investment but an industrial positioning move.

The word "bet" here has two meanings: "betting" is like gambling, and "treasuring" means staking their own future five years as collateral.

The rest of the story depends on whether Anthropic's IPO can support this gambling game.

To the Friend Who Keeps Asking "Can I Still Get on the AI Bandwagon?"

The figure of $965 billion is actually not important. The three names on the list are.

The fact that three names that shouldn't have appeared together quietly showed up on a shareholder list means that the real positioning battle in the AI industry has shifted from the "model war" in the front stage to the "production capacity war" behind the scenes.

If you have friends who invest in technology stocks or people who always ask "Can I still invest in AI now," you can share this article with them.

It's not about telling them what to buy, but telling them one thing: The way to view the AI game may change from today on.

(Note: This article involves the trend observation of the evolution of the AI industrial chain and the memory industry from 2026 to 2028. Some inferences are marked as "the following is an inference." Please read independently based on your own judgment.)

This article is only for information sharing and industry analysis and does not constitute any investment advice, investment analysis opinion, or trading invitation. The market is risky, and investment should be cautious. Any investment decision made based on the content of this article, the risks and profits and losses shall be borne by the investor himself. The author and the publishing platform shall not bear any legal liability. The content marked as "inference" is the author's logical deduction based on public information and does not represent the official position.

This article is from the WeChat official account "BT Finance" (ID: btcjv1), written by Yining, and published by 36Kr with authorization.