AI Trends

Anthropic Completes Approximately $81.3 Billion IPO, Seven Co-Founders’ Fortunes Surge

36Kr · AIUpdated Nov 20 08:31

AI summary

Anthropic debuted on Nasdaq under the ticker “ANT” on November 19, U.S. Eastern Time, selling approximately 980 million shares and raising about $81.3 billion. Its seven co-founders sold a combined 23.6 million shares, cashing out roughly $2 billion, while each retained shares worth approximately $19 billion to $20 billion at the offering price. Nvidia, Google, Microsoft, AMD, Broadcom, Micron, and TSMC additionally subscribed for a combined $26 billion worth of shares at the offering price.

Why read this

SelectedAI score 86

The article connects the scale of Anthropic’s public offering with changes in founder and employee shareholdings and revenue, outlining its capital expansion and commercialization path.

Topics

Tags

Body · AI translation

On November 19, U.S. Eastern Time, Anthropic rang the Nasdaq bell, completing this epic IPO under the ticker symbol “ANT.”

According to foreign media reports, the company sold approximately 980 million shares in the offering, raising about $81.3 billion. This surpassed the roughly $75 billion record set by SpaceX earlier this year, making it the world’s largest IPO.

Nvidia, Google, Microsoft, AMD, Broadcom, Micron, and TSMC separately subscribed for a combined $26 billion worth of shares at the offering price.

Notably, the company’s seven founders sold a combined 23.6 million shares in the offering, cashing out approximately $2 billion. This also means that the seven co-founders have officially made the leap from researchers to billionaires.

And that represents only a small fraction of their holdings. At the offering price, each still owns approximately 230 million to 240 million shares, worth around $19 billion to $20 billion.

Moreover, according to estimates by Sacra, the vested equity held by approximately 2,800 current and former employees may be worth more than $1 million each before taxes. Private equity trading platform Hill estimates that about 598 non-founder employees will each hold shares with a book value exceeding $100 million.

It was the first Christmas holiday of the pandemic. The streets were deserted, and most offices stood empty. On that day, Dario Amodei left OpenAI.

He was OpenAI’s vice president of research and a key figure behind the creation of GPT-2 and GPT-3. Just a few months earlier, GPT-3 had sent shock waves through the entire technology industry. By any conventional measure, he was at the pinnacle of his career.

Yet he chose to turn away at the summit. To understand that decision, one must begin with his father.

Dario was born in San Francisco in 1983. His father, originally from Italy, was a leather craftsman, while his mother worked in project management at a library. As a teenager, Dario was obsessed with mathematics and physics. He first attended Caltech before transferring to Stanford to study physics. Judging by the path he was on at the time, he should have become a theoretical physicist.

Around that time, however, his father developed a rare disease and died in 2006. What made it even harder for him to come to terms with was that, within just a few years, advances in medicine transformed the disease from a terminal illness into one that was treatable in most cases.

What if science had moved just a little faster?

That question stayed with him for the rest of his life. He shifted from physics to biology, earning a PhD in biophysics at Princeton, where he studied neural circuits, and later completing a postdoctoral fellowship at Stanford Medical School. But he gradually realized that biology was too complex—so complex that the human brain might simply not be up to the task. Perhaps a more intelligent tool was needed.

In 2014, he joined Baidu’s research institute in Silicon Valley. While working on speech recognition, he observed a phenomenon that would change his destiny: the larger the model, the more data it had, and the greater the computing power behind it, the more reliably its performance improved, almost like a law of physics. Many people regarded this as an engineering rule of thumb, but he saw in it a path toward powerful intelligence.

It was also then that he began to feel afraid. If this path was real, this power would arrive faster than anyone expected. It could cure diseases like the one that had killed his father, but it could also spiral out of control. At Google Brain, he and his colleagues wrote an influential early paper in the field of AI safety.

Then, in 2016, he joined the recently founded OpenAI.

Five years later, he personally proved that “bigger is stronger” was correct. It was also at this point that he and OpenAI’s leadership reached an impasse over the question of “how AI should be built.”

Six others walked out with him.

The first was his younger sister, Daniela. She had studied English literature, worked in politics, handled risk management at Stripe, and overseen safety and policy at OpenAI. Her brother could see the future; she knew how to organize a group of people and lead them there.

There was also Tom Brown, the lead author of the GPT-3 paper; Jared Kaplan, the theoretical physics professor who expressed the “scaling laws” as formulas; his collaborator Sam McCandlish; Jack Clark, who left journalism at Bloomberg to work on AI policy; and Chris Olah, who never finished college but pioneered research into neural network “interpretability.” Olah was only in his twenties at the time, making him the youngest of the seven.

In early 2021, the seven registered a company in San Francisco and named it Anthropic, meaning “relating to humans.” They incorporated it as a public benefit corporation, meaning it was not legally required to exist solely to make money for shareholders. The seven divided the equity equally, and none was the “big boss.”

Many people said, seven co-founders? They are bound to split up. Five years later, eight of OpenAI’s 11 co-founders have left, while not one of Anthropic’s seven has departed.

The most poignant moment in this story came in the summer of 2022.

At the time, an early, conversational version of Claude was already sitting on their servers. Had they released it, it would have reached the world before ChatGPT. Dario held it back. He worried that once it was released, it would ignite an arms race that no one could stop.

A few months later, ChatGPT burst onto the scene, surpassing 100 million users in two months as acclaim and capital poured toward his former employer. The arms race came anyway; he simply was not the one holding the starting pistol.

That year, Anthropic was a classic example of a company that was “one step behind.” No one imagined that, four years later, it would be standing at the center of Nasdaq.

The Fate of Four Checks

The first check came from a man who had been worried for a decade.

In the spring of 2021, Anthropic was still a company with no product, no revenue, just seven people and an idea. The person willing to pay for it was Estonian entrepreneur Jaan Tallinn.

Tallinn, a co-founder of Skype, had long since become financially independent. But over the previous decade, what concerned him most was not making money, but a question few people took seriously at the time: What if humanity created AI it could not control? In these seven people, he saw a group who were both capable of building the most powerful AI and, like him, afraid of it.

He led a $124 million investment. Joining him in the bet were Facebook co-founder Dustin Moskovitz and former Google CEO Eric Schmidt. What they bought at the time was less a company than a hypothesis.

Five years later, the hypothesis was worth roughly 300 times as much on paper. Tallinn later invested more in the Series B. Moskovitz, meanwhile, transferred his stake to a nonprofit entity, meaning the money would eventually go to charity.

The second check came from the Bahamas.

In 2022, the crypto world’s most dazzling young star, Sam Bankman-Fried, led Anthropic’s Series B. His company FTX invested about $500 million in total for a stake of roughly 8%. ChatGPT had not yet been released, and few people understood why he was putting so much money into a quiet research company.

It was the best investment he ever made. But he did not get to see it bear fruit.

That November, FTX collapsed within a week, and Bankman-Fried later went to prison. The Anthropic stake became a bankruptcy asset and was auctioned off for about $1.3 billion, which was returned to creditors. Today, it might be worth tens of billions of dollars. It was perhaps the costliest case of selling too soon in the history of capital markets.

The third check came from a giant said to have fallen behind.

After ChatGPT became a sensation, Silicon Valley’s deck was reshuffled overnight. Microsoft aligned itself with OpenAI, and the other tech giants began to worry.

Amazon was the most anxious. Despite operating the world’s largest cloud service, the company was widely seen as having fallen behind in the AI wave. In September 2023, it put up an initial $1.25 billion, later increasing its investment in several tranches to about $8 billion. Google also entered that year, receiving in return Anthropic’s commitment to use its in-house chips.

At the time, many people thought the two giants were spending money on a consolation prize.

By this spring, Amazon’s $8 billion investment had grown to a paper value of about $74 billion. In the first quarter of this year alone, it recognized $16.8 billion in pretax gains from the investment. At the IPO valuation, its cumulative investment could become worth more than $200 billion, while Google’s could be worth more than $100 billion. The consolation prize had become the jackpot.

The fourth check was written by all of Wall Street.

Starting in 2025, the story suddenly accelerated. That spring, Anthropic was valued at $61.5 billion; by autumn, that had become $183 billion; this February, $380 billion; and three months later, Sequoia, Altimeter and other institutions lined up with $65 billion, pushing its valuation to $965 billion.

That day, Anthropic surpassed OpenAI for the first time to become the world’s most valuable AI startup. The company that had been “one step behind” four years earlier had left its former parent in the dust.

The most intriguing scene in today’s IPO was Nvidia and Google each putting up about $10 billion, with AMD, Broadcom, Micron, Microsoft and TSMC also joining in, subscribing for a combined $26 billion at the offering price. They bought nonvoting shares. Chip sellers, memory manufacturers, foundries and cloud providers—the entire computing-power supply chain—lined up behind Anthropic.

They do not want a say in the company. They simply want to be tied to it.

Why Everyone Loves Claude

If you ask Silicon Valley programmers about their first impression of Claude, the answer is often not that it is “powerful,” but that it is “reliable.”

From its first day, Anthropic devoted enormous effort to areas others considered “unprofitable”: studying what exactly was going on inside a model’s mind, writing a public “constitution” to tell it what it should and should not do, and repeatedly testing whether it would act on its own. The persistence once dismissed as “academic” later became the reason banks, law firms and hospitals chose it.

It also made a promise rare in Silicon Valley: Claude would carry no advertising. As rivals began monetizing through ads and traffic, Anthropic positioned Claude as a colleague that works for you, not an app competing for your attention.

But what truly transformed Anthropic from a company with “a good reputation” into a highly profitable one was a tool developed internally.

The protagonist of this story is engineer Boris Cherny. He built a small command-line tool that allowed Claude to read code on a computer, modify files and run tests directly. It had no polished interface, just a black window. Yet one colleague after another began using it—and then could not put it down.

The tool was later given a name: Claude Code.

It officially went on sale in May 2025, and for the first time programmers felt as though they had gained a colleague who could work independently alongside them. Six months later, its annualized revenue had surpassed $1 billion, making it one of the fastest-growing products in enterprise software history. Within Anthropic, it already writes most of the code.

At the beginning of this year, Anthropic taught the same skills to people who do not write code. Its new product, Cowork, handles files, research and reports in the same way Claude Code handles codebases. Lawyers, fund managers, consultants and doctors began adopting it in droves.

Anthropic’s annualized revenue surged from about $9 billion at the end of 2025 to approximately $81 billion by the end of October this year, increasing roughly ninefold in ten months.

In all of 2024, Anthropic earned less than $400 million. A year later, that figure had risen to $4.6 billion. In the first half of this year alone, it brought in more than $16 billion in six months—over three times its revenue for all of last year. In October alone, its annualized revenue increased by about $10 billion, more than double its entire revenue for last year.

Even more striking was the change in its customer base. A year ago, fewer than 1,000 companies were paying it more than $100,000 annually; a year later, that number had grown to more than 11,000. Many companies initially let only a few engineers try Claude, but a year later, their spending on it had increased ninefold on average. They were no longer merely experimenting with something new; they were handing over more of their work.

Of course, this path is also extremely costly. In the first half of this year, Anthropic spent nearly $9 billion on computing power alone, while its long-term computing contracts have reached hundreds of billions of dollars. Its books also show a staggering $80.7 billion loss, but the overwhelming majority of that was an accounting loss generated by financing instruments such as convertible bonds after the company’s valuation soared. Its actual operating loss was only around $700 million and was narrowing rapidly.

To sustain this ever-accelerating growth curve, it needs more electricity, more chips and more money. That is why it is making its Nasdaq debut tonight.

Seven People’s Fortunes Soar

If someone had told Chris Olah at the beginning of this year that, eleven months later, he would be among the richest people in the world, he might first have investigated whether the statement was a model hallucination.

Olah is 34 this year and the only one of the seven founders born in the 1990s. He has no university degree and has spent more than a decade doing one thing: taking neural networks apart to see what they are actually thinking. Most of his co-founders are also still quite young: Daniela and Tom Brown are both 39, Jack Clark is 38, Jared Kaplan is 42, and Dario himself is 43.

Link their wealth trajectories together, and they resemble a film racing forward at ever-increasing speed.

In January this year, Forbes estimated that each was worth a little over $3 billion. At the time, the seven had just made a decision together: eventually, each would donate 80% of their wealth. Dario wrote in an article that AI could bring an unprecedented concentration of wealth and that action had to begin before the trend became uncontrollable.

On May 28, after the Series H financing was completed, their fortunes more than doubled in an instant to about $16.6 billion each. The Bloomberg Billionaires Index noted that this was the first time in its history that a single company had produced seven billionaires at once.

Tonight, that figure has reached another new level.

This wealth has some distinctly “Anthropic” details.

The first is equality. The seven founders’ shares were divided almost like seven identical bamboo steamer baskets. Although Dario is CEO and Daniela is president and chair of the board, neither received more than the others. Their salaries are not excessive either: Dario’s base salary last year was $650,000.

The second is restraint. In this IPO, the shares sold by all seven founders combined amounted to less than 2% of their holdings. The company specifically noted that the proceeds would not go to the company.

Viewed in the context of the wider industry, their wealth appears even more dramatic. OpenAI CEO Sam Altman does not directly own shares in his company, while co-founder Ilya Sutskever testified that his stake was worth about $7 billion. Each of the seven people who once left their side is now wealthier than either of them.

The company’s employees have risen along with them.

Private-market research firm Sacra estimates that, based on the $1.35 trillion valuation implied by the midpoint of the offering price range—about RMB 9.02 trillion at current exchange rates—the vested equity held by each of Anthropic’s roughly 2,800 current and former employees could be worth more than $1 million before taxes, or about RMB 6.681 million at current exchange rates.

Separately, Hill, a San Francisco platform for trading shares in private companies, estimates that after Anthropic goes public, 598 non-founder employees will each hold shares worth at least $100 million.

This is not entirely good news for Dario. According to reports, he once worried that enormous financial incentives would lead people to join for the money rather than the mission. It is a peculiar concern: one of the company’s greatest risks is that it has become too successful.

Tonight, the price of “ANT’s” first trade flashes onto the screen. “ANT”—a tiny ticker for a very large company.

This article is from the WeChat public account Rongzhong Finance (ID: thecapital), written by Abu and edited by Wuren. Published by 36Kr with authorization.

Source

36Kr · AI