Text by: AI
Foreign media reports indicate that Anthropic is expected to complete a new round of financing totaling approximately $300 billion as early as next week, with a valuation exceeding $9 trillion.
As recently as February, Anthropic's valuation stood at $3.8 trillion. In just three months, it has nearly tripled.
What's most interesting, however, is that Google has also participated in this funding round.
Google had previously committed in April to invest up to $40 billion in Anthropic. An initial $10 billion in cash was injected at a transaction valuation of about $3.5 trillion at the time, with up to $30 billion in subsequent investments tied to Anthropic's performance.
Google has Gemini, and just recently announced Gemini 3.5 at its I/O conference.
Wouldn't that $40 billion be better spent on R&D for their own products? Especially when it's being invested in a competitor.
Therefore, I believe Google may not necessarily want to buy Anthropic's models; what they might want to buy is a position.
It's not uncommon for competitors to invest in each other; Intel has invested in AMD, Microsoft in Apple, and Sony in Epic Games.
Large companies frequently invest in promising startups, reaping financial returns while staying in touch with new technologies.
But the scale and persistence of Google's investment in Anthropic seems a bit "excessive."
After all, it's $40 billion! Enough for Google to build another AI team from scratch.
So, what exactly has Anthropic done to captivate Google to such an extent? Besides Opus 4.7 and some minor product updates, the buzz around Anthropic centers solely on Mythos.
This is a model described by Anthropic as being so powerful that it absolutely must not be made available to ordinary consumers. Corresponding to Mythos is the Glasswing cybersecurity collaboration program.
It's akin to saying I've written a groundbreaking article, but I won't publish it because I fear everyone who reads it will become utterly obsessed and unable to extricate themselves.
The idea itself is absurd. Yet, Anthropic, through various overt and covert channels, has given tangible form to the power of Mythos in everyone's mind. Even if you haven't seen it, you can sketch its contours clearly through news and snippets of information.
That's why I think Anthropic's model products are not the core of this company. Its truly admirable aspect is its ability to tell a story.
It can transform something you can't see or touch into real gold and silver.
Latest Progress on Glasswing
We must first acknowledge a fact: the market accepts hype.
On May 23, Anthropic published an article titled "First Update on Project Glasswing," discussing the latest progress of Glasswing.
The article stated that Mythos Preview had scanned over 1,000 open-source projects, discovering 6,202 high or critical severity vulnerabilities. After evaluation by independent security research firms, 90.6% were confirmed as true positives, with 62.4% classified as high or critical severity.
It wasn't until I read that article twice that I realized it was essentially a "watered-down" piece.
The most subtle aspect of these numbers is the lack of a frame of reference.
Anthropic didn't specify the scale of those 1,000+ projects, nor did it disclose the total lines of code scanned, the time taken, or the improvement compared to traditional security tools.
Out of those 6,202 candidate vulnerabilities, only 1,752 were actually evaluated.
This fact itself is very un-Anthropic. Previous Anthropic articles were often reproducible. For instance, their use of Qianwen to simulate whether humans could control AI if it surpasses human intelligence.
Anthropic used Qianwen instead of their own Claude so that after reading, you could go back and replicate the experiment yourself.
But the Glasswing article didn't offer that. You can't reproduce it, nor can you verify it.
More crucially, Anthropic packaged the "decision not to publicly release" as an embodiment of responsibility. They fear misuse, hence the non-release. It's only released to specific users, and they use feedback from these users to indirectly validate the model.
Anthropic used such a marketing narrative to transform an "unreleased" product into proof of high enterprise technical prowess.
This granted it a special kind of immunity. It doesn't need to prove it's stronger than all competitors; it only needs to prove it's "so strong it can't be released casually."
Anthropic also mentioned in the update that industry practice is to publicly disclose vulnerabilities 90 days after discovery or 45 days after a patch is released. This means details of the vulnerabilities discovered by Mythos Preview cannot be fully disclosed yet, as it would expose end-users to risk.
This is a layer of protection.
It's similar to the Windows vulnerability mechanism. Research teams discovering Windows vulnerabilities send them to a specific email address. Only after Windows releases an update/patch to fix the vulnerability do these teams publicly disclose the details.
Anthropic's statement is logically impeccable. But it's not Microsoft, Claude isn't Windows; it's just a model. If I can't use Claude, I can use ChatGPT. If your computer can't use Windows, it's a brick (I'm exaggerating a bit; you could use Linux, etc., if you know how).
This is the most impressive aspect of Mythos. It's not just potentially powerful; it's designed within a product narrative where "the less public it is, the stronger it appears."
Anthropic seized on this characteristic. It doesn't need everyone to use Mythos; it only needs the market to believe Mythos represents a certain superior capability. And this capability, precisely because it cannot be widely verified, appears even more mysterious and powerful.
This is what Anthropic is doing.
It's using a very clever way to transform "unverifiable capability" into "imaginable value."
And this value ultimately manifests in valuation, fundraising, and the overall judgment of the market and society towards Anthropic.
The White House Re-enables Claude
If only Anthropic itself claimed Mythos was powerful, that would just be corporate marketing.
But if the US government, while viewing Anthropic as a supply chain risk, is also reportedly close to allowing intelligence agencies like the NSA to use its advanced models, the story is completely different.
In February, former President Trump indicated the US government would blacklist Anthropic, and the Pentagon labeled Anthropic a supply chain risk.
The most intriguing aspect at the time was that the White House wasn't saying Claude was unusable. On the contrary, the conflict stemmed precisely from "wanting to use it too much."
Foreign media reported that the Pentagon wanted to use Claude more freely in military and national security contexts, but Anthropic insisted on not relaxing restrictions on two types of uses: large-scale domestic surveillance within the US, and fully autonomous weapons.
The situation indeed seemed like the left brain attacking the right brain. The White House, on one hand, believed Anthropic's restrictions made it a supply chain risk, while on the other hand, acknowledged the technology in Anthropic's hands was sufficiently important.
But just after the publication of the first Glasswing update article, foreign media suddenly reported that the White House had reached a cooperation agreement with Anthropic, allowing specific agencies to use Claude, most likely Mythos.
If even the White House is willing to contradict its own ban and re-enable Claude, it shows this thing is genuinely powerful.
This contradiction itself endorses Anthropic.
While the news didn't state it directly, it conveyed a signal: Anthropic's technology has unique value.
Because after the White House banned Claude, OpenAI struck a deal, becoming the White House's AI supplier. Now, the White House's actions essentially tell you, "Anthropic is irreplaceable."
Anthropic doesn't need to prove to the public how powerful Mythos is; it only needs the market to know that Mythos is now a White House's Choice selection.
Procurement by government agencies, especially national security-related ones, signifies your technology has passed the highest standard of scrutiny.
Making it onto the procurement list means your company has met US government requirements in terms of supply chain security, technological reliability, and long-term support capabilities.
More importantly, government procurement often comes with long-term contracts and stable revenue.
This is crucial for investors, especially with Anthropic approaching an IPO.
Having such a major order demonstrates Anthropic doesn't just have technology; it has the ability to convert that technology into stable commercial revenue, and the most reliable customer.
What's even more interesting is that Anthropic didn't adopt a submissive, apologetic stance.
After being named by the Pentagon, it publicly emphasized it wasn't refusing to serve US defense, but couldn't abandon two red lines: large-scale domestic surveillance and fully autonomous weapons.
Amodei later stated in a company announcement that while the tone of internal statements could apologize, these principles would not change.
In other words, Anthropic packaged a procurement conflict first as "I have principles," and then, through subsequent news about the White House and NSA, turned it into "The US government still needs me in the end." This has more communicative value than simply securing the US government contract.
Yet, Anthropic hasn't mentioned a word about this cooperation.
If they loudly promoted "We're cooperating with the NSA," the market might suspect it's a marketing tactic. But this complete silence, on the contrary, makes the market even more convinced of Mythos's power, because it has compelled the US government to use it, even if it means backtracking.
Even if the final cooperation with the White House and NSA doesn't fully materialize, even if the specific content of the agreement differs from external perceptions, this narrative itself has already produced an effect.
It has already made the market believe Anthropic is a company capable of cooperating with national security agencies, a company whose technical capabilities meet White House standards.
Anthropic's greatest strength isn't convincing consumers; it's convincing the hardest-to-convince procurement party.
The decision-making process within the national security system is extremely complex, involving multiple stages like technical assessment, security review, policy considerations, and budget approval.
Being able to navigate these stages itself demonstrates a company's comprehensive strength.
All for Fundraising and Going Public
Technical narratives can be hyped, security narratives can be told, and US government cooperation may exist in various grey areas.
But fundraising is the most fundamental; capital ultimately converts stories into money.
Previously, when evaluating a company, investors looked at its revenue growth, profit margins, market share, and technical barriers.
But in the AI industry, these traditional metrics are often insufficient. Because even today, no one can clearly see an AI company's potential.
Foreign media reported that Anthropic told investors its annualized revenue would exceed $50 billion next month.
Anthropic's earliest commercial revenue growth wasn't primarily from regular Claude subscriptions, but mainly from enterprise APIs, cloud platform distribution, and code generation demand.
Foreign media reported in May 2025 that Anthropic's annualized revenue grew from nearly $1 billion in December 2024 to over $2 billion by the end of March 2025, reaching about $3 billion by the end of May, driven mainly by enterprise demand and code generation scenarios.
In the second half of 2025, Claude Code began to become an independent growth engine. Foreign media reported that Claude Code had an annualized revenue of about $400 million in July 2025, later approaching $1 billion.
Anthropic later confirmed in its official announcement of acquiring Bun that Claude Code reached a $1 billion run-rate revenue within six months of public release.
By February 2026, Anthropic officially disclosed the company's overall run-rate revenue reached $14 billion, with Claude Code alone exceeding $2.5 billion. In other words, Claude Code transformed from a developer's tool into a multi-billion-dollar product.
In April 2026, Anthropic, in its announcement of compute partnerships with Google and Broadcom, stated the company's overall run-rate revenue had exceeded $30 billion, continuing a sharp rise from about $9 billion at the end of 2025.
Simultaneously, the number of enterprise customers spending over $1 million annually increased from over 500 in February to over 1,000.
Investors are evidently willing to believe these numbers. Because various facts support Anthropic's story.
Google's continued investment further enriches Anthropic's narrative.
If an ordinary VC invested in Anthropic, the market might see it as normal venture capital behavior. But if an AI giant like Google invests in Anthropic, it indicates Anthropic truly has something exceptional.
Google isn't short on money, technology, or talent. Its investment in Anthropic isn't because Anthropic can fill its gaps, but because it judges Anthropic might become a key player in the AI field and doesn't want to miss the opportunity to get on board.
This judgment itself becomes an endorsement for Anthropic.
Investors often refer to other investors' choices when making decisions. This isn't called following the crowd; it's called "a rational way of gathering information."
Each investor has their own information channels and judgment capabilities. When multiple leading investors make the same choice, the probability of that choice being correct increases.
Google's investment is such a strong signal. It tells other investors Anthropic is worth investing in, and worth investing in heavily.
Thus, other investors begin to follow suit. Sovereign wealth funds, top VCs, large tech companies all express interest in Anthropic. This collective action further pushes Anthropic's valuation higher.
The rising valuation, in turn, reinforces Anthropic's narrative.
When a company's valuation reaches $9 trillion, the market naturally assumes this company must have exceptionally strong capabilities and vast prospects. Otherwise, how could it be worth so much?
Money plays the most critical role in the story.
You might not understand the technology, you might not understand the White House situation, but can you not understand dollars?
What Anthropic is securing now is precisely this kind of gold-and-silver endorsement.
This is the pricing function of capital. It ultimately converts all narratives, all imaginations, all expectations into a concrete number.
Of course, this consensus isn't necessarily correct. Figures including Sam Altman and Jensen Huang have acknowledged there are bubbles within AI.
A high valuation doesn't guarantee a company's success; massive fundraising doesn't guarantee technological leadership.
Historically, many highly-valued companies ended in failure.
But at this moment, Anthropic has indeed accomplished something remarkable. It packaged a capability unverifiable by ordinary users, a cooperation not officially announced by the government, and a revenue expectation that might be realized in the future, into a complete story. Then, using this story, it secured money from the market.
This is Anthropic's business model.
The brilliance of this mechanism lies in the fact that Anthropic doesn't need everyone to see Mythos firsthand. It only needs the wealthiest, most powerful, and most risk-savvy individuals to act as if they've already seen it.
And the actions of these individuals become the best proof for Mythos, more convincing than performance benchmarks or scores.
