AI Trends

The History of Anthropic’s Split from OpenAI Behind Its Trillion-Dollar IPO

TMTPost · AGIUpdated Nov 20 13:53

AI summary

Anthropic debuted on Nasdaq at an offering price of $83, closing with a market capitalization of approximately $1.27 trillion and raising about $81.3 billion through the IPO. The article revisits the conflicts between Dario Amodei and others and OpenAI over AI safety strategy, organizational management, and personal trust, while tracing how Anthropic built its commercial ecosystem around enterprise customers, Claude, and Claude Code.

Why read this

SelectedAI score 89

The article examines the conflicts over strategic direction, trust, and power in the split between Anthropic and OpenAI, providing context for understanding the subsequent competitive landscape between the two companies.

Topics

Tags

Body · AI translation

By Cui Peng (CP)

On the morning of November 19 local time, Anthropic made a record-breaking debut on Nasdaq.

Anthropic’s shares were priced at $83 and closed at $97.17, giving the company a closing market capitalization of approximately $1.27 trillion. The stock held by each of the company’s seven co-founders will be worth more than $20 billion, making them billionaires in every sense of the word.

The IPO raised approximately $81.3 billion, surpassing SpaceX’s previous IPO fundraising record of $75 billion and becoming the largest public listing in global history.

It took Anthropic just over five years to go from its founding in 2021 to crossing the threshold of a trillion-dollar IPO.

Many members of its founding team were early core figures at OpenAI. Founder Dario Amodei previously served as OpenAI’s vice president of research.

He, Altman, and others were brought together by similar ideals and jointly laid the foundation for OpenAI’s success, only to part ways later amid a crisis of trust.

The rift between the two CEOs persists to this day. In internal communications, Dario compared the legal battle between Altman and Musk to “Hitler fighting Stalin” and sharply criticized OpenAI for being dishonest.

At the New Delhi AI summit this February, as those present joined hands and raised them overhead, Dario and Altman merely touched elbows. The awkward scene attracted considerable media attention.

Wall Street, however, does not care. If anything, it is grateful for that rupture years ago, because without it, there would have been no trillion-dollar IPO.

At the time, Dario was conducting AI research at Google and had already become fairly influential. His sister, Daniela Amodei, worked at payments company Stripe. Also sharing their home was her fiancé, Holden Karnofsky, who later joined OpenAI’s board.

OpenAI co-founder Greg Brockman was a former colleague of Daniela’s. He was then recruiting talent from far and wide in hopes of assembling a research team capable of advancing AI.

Brockman once tried unsuccessfully to recruit Dario. That did not stop the group from frequently exchanging ideas in their shared house on Delano Avenue in San Francisco and imagining the future of AI.

Dario and his roommate Holden Karnofsky saw things differently. They believed sensitive matters concerning AI capabilities should first be reported to government agencies, rather than rushed into the public eye.

At the time, however, the differences between the two sides were confined to verbal discussions. It was only years later that Brockman came to view them as reflecting the philosophical divide between the two companies.

In 2016, after having dinner with Altman at an Indian restaurant, Dario decided to join OpenAI.

After joining the company, Dario often stayed up until midnight with Brockman, training AI to play video games. Later, at Valve’s 2018 TI (The International), OpenAI Five—a team of five bots—even played an exhibition match against a professional DOTA 2 team.

Before long, Dario became a key leader of the research team and helped drive the development of the GPT series of models.

At the time, a small group within OpenAI, including Dario, held two firm beliefs:

First, putting more computing power into models would continue to improve their performance, with virtually no end in sight.

This conviction later also became one of Anthropic’s most important technical principles.

Daniela later joined OpenAI’s core management team as vice president of safety and policy. Together with Sam McCandlish, Jared Kaplan, and others, they gradually formed an important research force within OpenAI.

These people did not merely discuss safety issues in meetings. When GPT-2 was released in 2019, OpenAI did not immediately make the full model public. Instead, it first released a smaller version and temporarily withheld the larger model.

At the time, the company was concerned that a more powerful model might be used to produce fake news, send emails while impersonating others, or generate harassing content at scale.

During its first few years, OpenAI was still a nonprofit research organization whose vision was to ensure that the development of AI would benefit all humanity.

Dario helped write most of the OpenAI Charter. The document stressed that while developing powerful AI, the company also had to ensure that it could benefit humanity safely.

At that stage, scaling and safety research were not at odds. Dario believed AI would become increasingly powerful, and he also hoped OpenAI would proceed cautiously enough along that path.

But as OpenAI later shifted into a commercial company, disagreements within the team over research direction, organizational management, and development philosophy deepened, gradually becoming irreconcilable.

“There’s no point arguing anymore”

In fact, the fractures within OpenAI did not appear overnight.

In 2017, an early OpenAI project ran into difficulties. Musk, the company’s principal financial backer, demanded that management assess employees’ contributions, after which the company carried out a round of layoffs.

According to an investigation by The Wall Street Journal, Dario was deeply dissatisfied and believed the approach was excessively harsh.

That same year, OpenAI also discussed a controversial idea internally. Company co-founder Brockman believed AGI could be sold to the permanent members of the United Nations Security Council. But Dario believed that selling such powerful technology to rival countries could pose serious risks.

After Musk left OpenAI in 2018, Altman assumed a leadership role as a co-founder. Dario spoke with him, hoping that fellow co-founder Brockman and chief scientist Ilya Sutskever would no longer dominate the company’s management.

Altman agreed, but Dario believed that Altman had also made promises to other executives—and that those promises contradicted one another and could not possibly all be fulfilled.

At the time, Brockman wanted to become more involved in research on the GPT series of models, but both Dario and his sister Daniela opposed the idea. The dispute involved not only disagreements over research direction, but also management authority, team ownership, and individual contributions.

Dario believed that he and his team had not received due recognition for their contributions to the GPT series. He was once upset that Brockman had spoken publicly about the OpenAI Charter because, in his view, he had contributed far more to the document than Brockman had.

In 2019, Dario asked to be promoted to vice president of research. Altman agreed and promised the board that Dario would report directly to him and receive a level of public-relations exposure comparable to Brockman and Sutskever.

But this brief period of calm did not last long.

In early 2020, Altman called Dario and Daniela into a conference room and questioned whether the siblings had been encouraging employees to submit negative assessments of him to the board. They denied the accusation and immediately demanded that Altman identify his source.

When the executive in question was called into the conference room, the person said they knew nothing about it. Altman then immediately denied ever accusing the siblings, prompting Dario and his sister to begin shouting at him.

The meeting ultimately descended into a fierce argument.

This account from The Wall Street Journal reveals a fact little known to outsiders: the split between OpenAI and Anthropic was not merely an academic dispute over approaches to AI safety, but was also entangled with long-simmering personal tensions and organizational power struggles.

At the end of 2020, Dario decided to leave OpenAI. After learning the news, Altman personally went to Dario’s home to persuade him to stay with the company.

But Dario set conditions: he had to report directly to the board, and he could no longer work with Brockman. Given that Brockman was an OpenAI co-founder, those conditions were clearly difficult to meet.

In a later interview with Bloomberg, Dario attributed his departure primarily to a lack of trust in Altman. He said that when two people have different visions and do not trust each other, there is no point in continuing to argue.

Dario’s grievances with Brockman had accumulated over many years, and combined with the collapse of his trust in Altman, they made the situation impossible to salvage.

In 2021, Dario and Daniela left the company, followed in succession by more than a dozen core OpenAI researchers.

Seven of them later became co-founders of Anthropic. The group not only left a rising star in the AI industry, but also personally created what would become one of its most formidable competitors.

Tech giants come courting

After Anthropic was founded, its first task was to solve a fundamental challenge to its survival: training advanced AI models requires enormous amounts of capital and computing power.

The technical abilities of a handful of people on a startup team alone could not enable it to compete head-on with a company like OpenAI. They needed to find partners willing to provide funding, computing resources, and commercial distribution channels.

ChatGPT burst onto the scene at the end of 2022, rapidly turning OpenAI into the center of the global AI industry, and major technology companies began paying attention to its competitors as well.

Google was among Anthropic’s earliest major backers. In February 2023, Google confirmed that it had invested in Anthropic and would provide the company with cloud-computing resources. Amazon also began investing in Anthropic that September, committing a total of $8 billion by the end of 2024 and establishing AWS as an important cloud-service provider for Anthropic.

Through Amazon Bedrock, for example, businesses can use Claude directly without having to build an entire model infrastructure from scratch.

It was during that period that Anthropic gradually established its own technological and commercial ecosystem.

Although Claude was long regarded as a challenger trying to catch up with ChatGPT, the two companies gradually pursued different product strategies:

OpenAI used ChatGPT to enter the mass market, first building a huge consumer user base and then offering model services to developers and businesses.

Anthropic, by contrast, initially focused on enterprise customers, particularly professional use cases such as programming, software development, and knowledge work. Claude gradually built a strong reputation in programming and began evolving from a chat tool into an assistant capable of participating in real-world development work.

The launch of Claude Code further amplified this advantage. It can understand codebases, plan changes, write code, run tests, and debug. For businesses, these capabilities can translate directly into greater development efficiency, giving them a reason to keep paying for the service.

Anthropic gradually found a path to commercialization that suited it.

Its customers include not only consumers who occasionally use chatbots, but also a large number of businesses willing to pay a premium for advanced model capabilities. Claude has been integrated into internal workflows such as software development and document processing, gradually becoming an everyday productivity tool at some companies.

Enterprise services have created a virtuous cycle for Anthropic: the more customers use Claude, the better the company can understand the needs of real-world workplaces, enabling it to attract more customers through model and product upgrades.

After Claude Cowork launched a series of skills early this year, it sent shock waves through industries such as law and finance, directly triggering a large-scale sell-off in traditional software stocks. Shares of giants such as Thomson Reuters even suffered double-digit declines.

By this point, Anthropic's growth had far exceeded outside expectations.

A Trillion-Dollar Breakup

During the IPO, more than 1,200 institutions lined up to subscribe for Anthropic shares, yet over half received no allocation at all. The capital market's enthusiasm for the company could hardly have been more evident.

The prospectus shows that Anthropic's revenue reached $17 billion in the third quarter of this year, while its annualized revenue run rate had reached $81 billion by the end of October.

Its customer base is also expanding rapidly. As of the end of September 2026, more than 11,900 customers had each generated at least $100,000 in revenue for Anthropic over the preceding 12 months, compared with only around 950 a year earlier.

Anthropic has grown from a startup into an AI giant with a vast enterprise customer base.

To support model development and customer demand, the company has also signed numerous computing-resource purchase commitments and lease agreements. According to the prospectus, these long-term commitments total approximately $895.4 billion, of which about $447.5 billion will come due by the end of 2030.

These enormous figures show that Anthropic has entered an entirely different stage of development. It now needs not only researchers, but also long-term, stable access to capital and computing power.

Anthropic and OpenAI now have different technological approaches, organizational structures, and business models. Both are at the forefront of the AI wave, yet they have different views on how to develop the technology, manage its risks, and turn it into commercial value.

When Dario left OpenAI in 2021, he must have been filled with ambition and a burning desire to prove himself.

But even the most optimistic participants could hardly have imagined that five years later, the company would be worth more than $1 trillion and the people who built it would have fortunes in the tens or even hundreds of billions of dollars.

This breakup not only changed the destinies of many people, but also altered the trajectory of the entire technology industry—and even the world.

It turned what could have remained a contest between competing approaches within a single company into a long-term rivalry between two independent companies. It also ensured that the future of AI would not rest solely in the hands of one company that had transformed from a nonprofit organization into a commercial enterprise—OpenAI.

This story of going from zero to $1 trillion may be the most expensive breakup in Silicon Valley history.

Source

TMTPost · AGI