On April 17, 2026, the AI funding circle was once again buzzing with excitement.
A screenshot circulated wildly among investor circles: Anthropic's implied valuation had quietly crossed a line on secondary market and derivatives platforms like Caplight and Ventuals—$1 trillion.
Briefly, but truly, surpassing OpenAI.
There was no official announcement, no press release, no statement from CEO Dario Amodei. It was just the Pre-IPO market voting with its dollars.
Investors grew excited looking at the revenue curve. Anthropic's annualized revenue surged from $9 billion at the end of 2025 to $30 billion in just four months—a 233% increase—and began spreading the word: the AI leader has changed.
Let's be clear about one thing. Anthropic's latest official post-money valuation was $380 billion after the completion of its Series G in February 2026. Several venture capital firms have since offered $800 billion or higher, but Anthropic has not yet accepted.
That $1 trillion is the implied figure on secondary market platforms.
Almost on the same day, another piece of news emerged from Hangzhou.
DeepSeek is planning its first external funding round since its founding, targeting a valuation of over $10 billion and aiming to raise at least $300 million. The first time in three years.
One is being force-fed capital, chased all the way to the trillion-dollar threshold. The other has kept capital at bay for three years, then chose a moment it deemed right to crack the door open slightly.
Reading these two pieces of news together reveals the same thing: this spring, the two most important AI companies from two different countries have both reached the boundaries of their respective paths.
Anthropic's List of Allegiances
First, Anthropic.
On February 13, 2026, Anthropic completed its Series G funding round, raising $30 billion with a post-money valuation of $380 billion. The lead investors were Singapore's sovereign wealth fund GIC and hedge fund Coatue, with co-investors including Blackstone, Goldman Sachs, J.P. Morgan, Qatar Investment Authority, Temasek, NVIDIA committing up to $10 billion, and Microsoft up to $5 billion.
Read that list again: Singapore's sovereign fund, Qatar's sovereign fund, America's largest investment banks, NVIDIA, Microsoft.
This is a list of allegiances. Global capital is voting with real money: AI's discourse power should remain in the United States, in the hands of this company.
Two months later, the report card arrived.
According to data from corporate spending management platform Ramp, in March 2026, 73% of new funds from enterprises' first-time purchases of AI services flowed to Anthropic, while OpenAI's share dropped to 27%. Just 10 weeks prior, it was a 50:50 split.
The core weapon is Claude Code, with annualized revenue exceeding $2.5 billion, more than doubling since early 2026, and enterprise subscription users quadrupling.
This reversal can be understood this way. OpenAI is building a consumer-facing Disney, charging admission fees based on foot traffic. Anthropic is constructing a toll road leading to enterprises' core systems—tolls are much more expensive than tickets, and once a vehicle is on the road, it's unlikely to switch lanes.
Shortly after Anthropic announced its overtaking, an internal memo written by OpenAI's Chief Revenue Officer Denise Dresser was leaked, accusing Anthropic of inflating revenue by approximately $8 billion using the "gross method."
When customers purchase services through platforms like AWS or Google Cloud, Anthropic records the entire amount paid by the customer as revenue, including the portion that must be shared with the cloud service providers. Excluding this part, Anthropic's true revenue is about $22 billion, not exceeding OpenAI's $25 billion.
The wording of this memo reads more like two former colleagues exposing each other's shortcomings.
Understanding this memo requires context. Anthropic's private market valuation is about $600 billion, a significant premium over its last funding round, while OpenAI's secondary market valuation is around $765 billion, a discount of about 10% from its last round. The former employer is under pressure in the capital markets; releasing this document is both an opponent attack and a move to stabilize its own position.
Then there's that discordant number amid the celebration. Anthropic is not expected to be profitable until 2027. $30 billion in annualized revenue, $380 billion valuation—every funding round breaks records, but profitability is still a distant future. The higher the valuation, the greater the investors' expectations, the faster the burn rate, and the more urgent the next funding round. Anthropic cannot break this cycle actively; it can only maintain it by running fast enough. This is its invisible wall.
And DeepSeek—It Left the Entire Investment Circle Waiting for Three Years
Now, Liang Wenfeng.
After R1 went viral, the entire Chinese investment scene was thrown into chaos. Zhu Xiaohu, who had just said he "didn't看好创业公司做大模型" (didn't看好 startup companies doing large models), publicly stated that the price wasn't too important anymore; the key was to be involved. Tencent executives went, Alibaba executives went, VCs took turns visiting.
Rumors emerged: Alibaba investing $1 billion, a Series C raising $700 million—one after another, one after another denied.
Liang Wenfeng left the entire investment circle waiting outside the door for three years.
His reason was one sentence: "No short-term financing plans. The problem we face has never been money, but the embargo on high-end chips."
幻方量化 (Huanfang Quant) invested 3 billion RMB in the first phase of DeepSeek's R&D, entirely supported by the profits from quantitative私募 (private equity). He really doesn't lack money; he lacks chips, and financing can't solve the chip problem.
As for why he doesn't accept investment, he has another concern: external investors coming in might interfere with company decisions.
Liang Wenfeng, reading his background, gives a sense of consistency. Born in 1985 in Zhanjiang, Guangdong, graduated from Zhejiang University's School of Information and Electronic Engineering, didn't look for a job after graduation and directly went into quantitative investment, founded 幻方量化 in 2015, self-invested nearly 200 million RMB in 2019 to build the computing cluster "Yinghuo一号" (Firefly One) equipped with 1,100 GPUs.
As soon as the A100 was launched, he抢在众多企业前面 (raced ahead of many companies), becoming one of the first in the Asia-Pacific region to get the chips. In 2021, he invested another 1 billion RMB to build "Yinghuo二号" (Firefly Two), equipped with about 10,000 A100s. In 2023, he shifted computing power to large models and founded DeepSeek.
Everything he does has an engineer's forward-thinking: prepare the tools first, then get to work. Refusing financing is one of his tools.
But now, this tool is beginning to fail.
DeepSeek's absolute salary is not low, but it cannot match the equity incentives and valuation premiums of market giants like ByteDance, Alibaba, and Tencent. Liang Wenfeng has already begun推进公司估值工作 (pushing forward company valuation work), clarifying option pricing to give the team more certainty.
Without external financing, there is no market valuation, and thus no option value. For a top engineer, working at DeepSeek means you might be changing the world, but you can't produce an equity certificate that calculates wealth.
In January 2026,智谱 (Zhipu) rang the bell on the Hong Kong Stock Exchange, MiniMax followed closely with an IPO. Peers' options are cashing in, and the talent pressure on DeepSeek is becoming increasingly real.
Another issue is being forced out: DeepSeek and幻方's executives are still discussing whether the company should shift from "primarily focused on research" to "building a business that generates substantial revenue and eventually profits." This discussion itself is a crack in the door.
This first funding round targets a valuation of over $10 billion, while the company's valuation in 2025 was about $3.4 billion. If the funding is completed, the valuation will achieve a several-fold leap. $300 million against a $10 billion valuation means a dilution of less than 3%. This number is extremely restrained, like someone feeling the temperature on the doorknob before opening the door, confirming no danger, then gently pushing it open.
Liang Wenfeng used three years of independence to earn himself the strongest negotiating chips. He is opening the door at the moment he is most confident.
Two Civilizations at the AI Table
Putting these two stories together, a hidden line emerges.
Anthropic's Series G investors: Singapore's GIC, Qatar Investment Authority, Blackstone, Goldman Sachs, NVIDIA, Microsoft.
Behind this list lies a complete logic: AI's discourse power should remain in the United States; "safe, trustworthy" AI is the next infrastructure. Every dollar coming in is betting on this judgment.
DeepSeek's first funding round, potential investors include Alibaba, state-owned funds, and other leading domestic institutions. This is the first time Chinese capital is publicly pricing a top AI research institution. Betting on another logic: technological autonomy, open-source ecosystem, domestic computing power.
Two lists, placed on the same table, represent two civilizations各自押注 (each placing their bets).
Closed-source and open-source are also choices between two power structures in this博弈 (game).
Anthropic is全程闭源 (fully closed-source), relying on enterprise trust premium. Monthly active users can generate $211 in revenue. What it sells, besides model capability, is a sense of assurance endorsed by experts—you don't need to understand it, just trust it.
Liang Wenfeng says open-source is "more of a cultural act than a commercial one; contributing to open-source will earn us respect." The former concentrates the definition of "what is good AI" in the hands of a few; the latter hands it over to global developers for discussion.
These are two political propositions about the future of AI.
But both companies actually face the same question: when you grow large enough, how do you prove you're worth that price?
Anthropic's answer is revenue growth and enterprise clients, but profitability won't come until 2027, and the former employer is constantly nitpicking on the side. DeepSeek's answer is still forming.
Epilogue
This race has no referee yet.
Anthropic's valuation冲向万亿 (rushes towards a trillion), profitability probably waiting until 2027. How long are the world's most shrewd sovereign wealth funds and top investment banks willing to wait? AI's history is short enough that no one has seen how a company of this scale soft-lands, nor how it crashes hard. Everyone is groping in the dark,只是摸索的姿势各不相同 (just the postures of groping vary).
DeepSeek's problem is the cost of choice. After financing, external shareholders come in. How long can the independence Liang Wenfeng has一直守护 (always guarded) last? Once the door is opened, no founder in the world can fully control what comes in after it's opened.
Dario Amodei positions himself as "an explorer寻找第三条路 (finding a third path) between the narrow roads加速通往天堂和坠入地狱 (accelerating towards heaven and falling into hell)." People around Liang Wenfeng say AGI is his ultimate goal; money and commercialization are not high priorities.
Two people, each believing they are doing something more important than financing.
The capital market doesn't believe in faith; it only believes in profit statements.
Three years from now, or five years, when we come back to翻这张账单 (flip through this bill): Did the company whose valuation once rushed towards a trillion prove it was worth that price? Did the company that used three years of independence to earn respect, then decided to take the first step,守住它的初衷 (guard its original intention)?
Both paths are not yet finished.
