On April 17, 2026, the AI ​​investment circle was once again excited.

A screenshot circulated among investors, and Anthropic's implied valuation quietly crossed a line on secondary markets and derivatives platforms like Caplight and Ventuals: $1 trillion.

Briefly, but truly, surpassing OpenAI.

With no official announcement, no press release, and no CEO Dario Amodei coming out to say anything, just the market voting in the Pre-IPO.

Investors looked at the revenue curve with excitement. Anthropic's annualized revenue jumped from $90 billion at the end of 2025 to $300 billion, a 233% increase in four months, and then they began to spread the word: the AI ​​leader has changed.

Let's make one thing clear. Anthropic's latest official post-money valuation was after the completion of the G round in February 2026, at $380 billion. Since then, multiple venture capitalists have offered $800 billion or higher, but Anthropic has not accepted.

That $1 trillion is the implied number on the secondary market platform.

Almost on the same day, another piece of news came from Hangzhou.

DeepSeek is planning its first external financing since its founding, targeting a valuation of over $100 billion with plans to raise at least $3 billion. The first time in three years.

One is being chased by capital, chased to the doorstep of a trillion. One has kept capital out for three years, then chose a time they felt right, and cracked the door open a bit.

Reading these two pieces of news together, what is read is the same thing: this spring, the two most important AI companies in two countries have reached the boundary of their own path.

Anthropic's List of Backers

First, let's talk about Anthropic.

On February 13, 2026, Anthropic completed its G round of financing, raising a total of $30 billion, with a post-money valuation of $380 billion. The lead investors were Singapore's government investment corporation GIC and hedge fund Coatue, with joint participants including Blackstone, Goldman Sachs, JPMorgan Chase, Qatar Investment Authority, Temasek, NVIDIA committing up to $10 billion, and Microsoft up to $5 billion.

Read through this list once: Singapore Sovereign Wealth Fund, Qatar Sovereign Wealth Fund, America's largest investment bank, Nvidia, Microsoft.

This is a list of alliances. Global capital is using real money to vote: the narrative of AI should stay in the United States, in the hands of this company.

Two months later, the report card arrived.

According to monitoring data from the corporate expense management platform Ramp, in March 2026, a whopping 73% of the new funds for companies' initial AI service purchases flowed to Anthropic, with OpenAI's share dropping to 27%. Just 10 weeks ago, the two were evenly split at 50:50.

The core weapon is Claude Code, with an annualized revenue of over $25 billion, more than doubling since the beginning of 2026, and enterprise subscription users quadrupling.

This reversal can be understood as follows. OpenAI is building a consumer-facing Disney, relying on foot traffic for ticket sales. Anthropic is constructing a toll road leading to the enterprise's core systems, with tolls much higher than tickets, and once the vehicle is on the road, it is unlikely to change lanes.

Shortly after Anthropic's announcement of its overtaking, an internal memo written by OpenAI's Chief Revenue Officer Denise Dresser was leaked, accusing Anthropic of inflating revenue by about $8 billion using the "gross settlement method."

When customers purchase services through platforms like AWS, Google Cloud, etc., Anthropic records the full amount paid by the customer as revenue, including the portion that needs to be shared with the cloud service provider. Excluding this portion, Anthropic's actual revenue is approximately $22 billion, which does not exceed OpenAI's $25 billion.

The wording of this document is more like two former colleagues airing each other's dirty laundry.

Understanding this memo requires some context. Anthropic's private market valuation is about $600 billion, a significant premium over the previous financing round, while OpenAI's secondary market valuation is approximately $765 billion, about 10% discount from the last round of financing. With its back against the wall in the capital markets, the release of this document is not only about attacking the competitor but also about shoring up its own position.

Then there is that discordant number in the celebration. Anthropic is expected to be profitable by 2027. With $30 billion in annualized revenue, a $380 billion valuation, setting new records with each financing round, but profitability is still in the distant future. The higher the valuation, the greater the expectations from investors, the faster the burn rate, and the more pressing the next round of financing. This cycle is something Anthropic cannot actively break; it can only survive by running fast enough. This is its invisible wall.

Meanwhile, DeepSeek Left the Entire Investment Community Hanging for Three Years

Let's talk about Liang Wenfeng.

After the R1 exploded in popularity, the entire Chinese investment community was in awe. Zhu Xiaohu, the person who had just said "I don't see much potential in startup companies pursuing a large-scale model," publicly stated that the price was no longer the main concern; the key was to be involved. Executives from Tencent went, executives from Alibaba went, and various VCs came knocking one after another.

Rumors of Alibaba's $1 billion investment came out, rumors of a $700 million Series C round came out, one after another, all of which were subsequently denied.

And Liang Wenfeng just left the entire investment community out in the cold, hanging for three years.

His reason was simple: "There is no financing plan in the short term. The issue we face has never been about money, but about high-end chips being restricted from import."

For the initial R&D investment in DeepSeek, Huansquare Quantitative contributed 3 billion RMB in self-funded capital, entirely supported by the profits from quantitative private placements. He truly doesn't lack funds; what he lacks is chips, and financing won't solve the chip problem.

As for why he didn't accept investments, he had another concern: external investors coming in might intervene in company decisions.

Upon reading about Liang Wenfeng, one might sense a consistent theme in his life. Born in 1985 in Zhanjiang, Guangdong, graduated from Zhejiang University's College of Information Science and Electronic Engineering, he went straight into quantitative investment after graduation. In 2015, he founded Huansquare Quantitative, and in 2019, invested nearly 200 million RMB to build the computing power cluster "Firefly No. 1," equipped with 1100 GPUs.

Upon the release of the A100, he was ahead of many companies, becoming one of the first in the Asia-Pacific region to obtain the chip. In 2021, he invested another 1 billion to establish "Firefly No. 2," equipped with around ten thousand A100 GPUs. By 2023, he shifted the computing power to focus on large-scale models and founded DeepSeek.

For everything he does, he has an engineer's preconceived notion: first, prepare the tools, then get to work. Rejecting financing is one of his tools.

But now, this tool is beginning to malfunction.

DeepSeek's compensation is undoubtedly high, but it cannot match the equity incentives and valuation premiums of market giants like ByteDance, Alibaba, and Tencent. Liang Wenfeng has begun working on the company's valuation, explicitly pricing options, to give the team more certainty.

Without external financing, there is no market valuation, and there is no option value. For a top engineer, working at DeepSeek means you may be changing the world, but you won't have a share certificate that can calculate wealth.

In January 2026, ZhiPu rang the bell at the Hong Kong Stock Exchange, followed closely by MiniMax's IPO. Meanwhile, competing options were being cashed out, and DeepSeek's talent pressure was becoming increasingly real.

Another issue was being forced out: DeepSeek and Illusion Square's executives were still discussing whether the company should shift from being "primarily research-focused" to "building a business that generates significant revenue and eventually turns a profit." This discussion itself was a glimpse of a doorway.

The target valuation for this initial funding round exceeded $10 billion, while the company's valuation in 2025 was around $3.4 billion. If the financing is completed, the valuation will increase several times over. $300 million is less than 3% dilution for a $10 billion valuation. This number is extremely conservative, like a person placing their hand on the door handle to feel the temperature before gently pushing it open to confirm there is no danger.

Over three years, Liang Wenfeng had earned himself the greatest bargaining chip. When he opened the door, it was at a time when he was most confident.

Two Civilizations of the AI Table

Putting these two stories together, a hidden thread emerges.

Anthropic's Series G investors include Singapore's GIC, the Qatar Investment Authority, Blackstone, Goldman Sachs, NVIDIA, and Microsoft.

Behind this list is a complete logic: the discourse on AI should remain in the United States. "Secure and trustworthy" AI is the next infrastructure, with every dollar coming in betting on this judgment.

DeepSeek's initial funding round, potential investors include Alibaba, state-owned funds, and other top domestic institutions in China. This is the first time Chinese capital has publicly priced a top AI research institution. The bet is on a different logic: technological independence, open-source ecosystem, and domestic computing power.

Placing these two lists on the same table represents two civilizations making their own bets.

Closed-source and open-source are also choices of two power structures in this game.

Anthropic remains closed-source throughout, relying on an enterprise trust premium. Each active user per month can generate $211 in revenue. Besides selling model capabilities, it offers a sense of reassurance endorsed by experts—you don't need to understand it, just trust it.

Liang Wenfeng said, "Open-source is more of a culture than a business act. Contributing to open source earns us respect." The former consolidates the definition of "good AI" in the hands of a few, while the latter hands it over to developers worldwide for discussion.

These are the two political propositions about the future of AI.

However, both companies are actually facing the same question: when you grow large enough, what do you use to prove your worth?

Anthropic's answer is revenue growth and enterprise customers, but profitability is expected by 2027, with the old-timers constantly nitpicking. DeepSeek's answer is still taking shape.

Epilogue

In this competition, there is still no referee.

Anthropic's valuation is heading towards a trillion, with profitability likely waiting until 2027. How long are the world's most astute sovereign wealth funds and top investment banks willing to wait? The history of AI is short enough that no one has seen how a company of this scale lands softly, nor how it crashes hard. Everyone is fumbling in the dark, each with a different posture of fumbling.

DeepSeek's dilemma is the cost of choice. After raising funds, external shareholders come in. How long can Liang Wenefeng protect the independence he has always guarded? Once the door is opened, there is no founder in the world who can completely control what will come in.

Dario Amodei positions himself as "an explorer seeking a third way between accelerating towards heaven and plunging into hell". Those around Liang Wenefeng say AGI is his ultimate goal, with neither money nor commercialization taking top priority.

Both individuals believe they are doing something more important than fundraising.

The capital markets do not believe in faith, only in the income statement.

Three years later, or five years later, when we revisit this bill: did the company whose valuation once soared to a trillion prove its worth? Did the company that, after three years of independence, earned respect and then decided to take the first step, hold onto its original intention?

Both paths have not yet been fully traveled.