It's the first time in the history of the world.
This is a historic moment!
Earlier this evening, Anthropic officially filed an S-1 IPO with the U.S. Securities and Exchange Commission (SEC) with Nasdaq, the stock code "ANT".
"For the first time in human history, intelligence has become something that humans can create themselves".
Anthropic put that in the first sentence of the book -- and that's when the company that built Claude's AI lab put its books, its research and its ambitions in front of the world for the first time!
- AI is creating AI.In September of this year, more than 90% of the code in the Anthropic Code Library was contributed by Claude, and much of the research and development of the next generation of Claude has been passed on to this generation.
- Making money for the first time.Revenue in the second quarter was $11.55 billion, which is 14.7 times higher than the same period last year, generating operating profit on the GAAP scale.
- The cycle is turning faster and faster.Claude's success rate in breaking software vulnerabilities rose from 40% to 91% in four months.
With industrialization, electricity, semiconductors, and the Internet, every single time, human lives are being rewritten.
Ninety percent code comes from Claude.AI is already creating AI.
Within Anthropic, much of the research and development of the next generation of Claude has been handed over to this generation.
It is responsible for running experiments for researchers, generating and processing training data, automating assessments and security audits, and writing and auditing code.
With the help of a stronger Claude, the researchers quickly found a way to improve the efficiency of computation. The higher the efficiency of computation, the stronger models can be trained, and the stronger models will go on to the next round.
Anthropic calls this a "combination loop" and describes it as one of the most important structural advantages in front-end AI.
In the engineering team at Anthropic, this is the fastest turnaround. In the past, it was only based on open engineering puzzles from senior engineers, and Claude scored about 50 points in nine months on the internal benchmark.In September, it reached about 90 percent.。
So engineers have changed their lives. They are increasingly giving Claude directions, test results, and no longer typing code one line at a time.
As a result, in the third quarter of this year, the amount of code that each engineer contributes to the code library per day is estimated to be around $4 billion.More than 45 times the average daily level in 2024By September, more than 90% of the contributions to the codebase came from Claude.
From January to September of this year, the value of Claude was used by Anthropic internally to calculate publicly quoted prices.More than 250 times the storm.。
As early as April,For the first time this month, the figure exceeded the company's monthly human resources expenditure.(With no incentive for shareholding)
From the end of 2024 to September this year, the number of employees at Anthropic increased by more than five times, but this year's third-quarter revenue was 86 times that of the fourth quarter of 2024.
So in the last 12 months,Per capita earnings reached more than $6.6 millionIt was $1.4 million a year ago, compared with $1.4 million a year ago.
Anthropic itself writes in its IPO that it is approaching the so-called"Return to self-improvement".This means that AI generations are making themselves stronger.
Over the past few years, AI companies have been competing for a more powerful model.
The first model to take on the life of the next generation of models is the one whose lead will roll the bigger and bigger like profits.
$11.5 billion in single-season franchisingAnthropic made money for the first timeIn the first quarter of 2023, when Claude started commercializing, Anthropic's revenue was zero. It earned $41 million for the full year, to $386 million in 2024, and to $455.5 million in 2025.
In the first quarter of 2026, revenue surged to $47.31 billion.The second quarter surged to $11.55 billion, 14.7 times the same period last year.。
In the second quarter, Anthropic's revenue grew by 1,368% and its computing and infrastructure spending by only 300%.
In the first quarter of 2025, Anthropic will spend $1.00 per revenue, $2.51 on optics and infrastructure.
In the second quarter of this year,That's down to $0.48.It's less than half.
So in the second quarter,Anthropic is profitable under GAAPThe company's operating profit was $974 million, and it's a result of taking all the expenses of equity incentives into account.
In the first half of the year, there was still a net loss of $80.7 billion, but $72.4 billion was a revaluation of the fair value of the financing instrument, not burned cash. During the same period, operating cash flows totaled $3929 million.
As of the end of September, Anthropic'sNet income retention rate of about 900%That is to say, corporate customers who paid a year ago now spend almost nine times as much on Claude as they did then.
Two years ago, no one would have trusted a leading model company to settle its accounts in a quarter and earn money in return.
The cycle is moving faster and faster, with a four-month breakdown rate from 40% to 91%.
On the exploitBench benchmark, Claude's performance in exploiting the vulnerabilities developed by researchers at Carnegie Mellon UniversityFrom 40% of Opus 4.8, to 78% of Claude Mythos 5, to 91% of Opus 5.5.It only took about four months.
This is an ability that even Anthropic himself would not dare to let go of.
When it trained Claude Mythos Preview, it judged the model's cybersecurity capabilities.There is a significant risk of large-scale openings that are strong enough to not be renewed.After adding a new cover, it was released to the public under the name Claude Fable 5.
In order to train and run a stronger Claude, from May to October of this year, Anthropic and SpaceX's wholly owned subsidiary, SpaceX.AI, signed and revised several arithmetic agreements to lease GPUs for up to $120.5 billion.
We'll have to wait until December for all the calculations to be delivered.Anthropic is paying about $3.1 billion a month to SpaceX.AI.
In addition to SpaceX.AI, Google, Amazon, Botnet, and Microsoft are all on Anthropic's list of calculators.More than $550 billion combined。
It's time to put decades of progress into years.In the end, a nation of geniuses
It's an anthropic judgment, and it goes like this:In a few years, "Strong"AI"It will become widespread".。
As described in the book, it reaches even higher levels than top human experts in most professional and scientific fields, can operate autonomously for hours, days or even weeks, and can directly control laboratory equipment and robots.
By that time, millions of Claude's will be working in parallel.Anthropic calls it the "state of the data center genius".
Once this kind of AI is targeted at the most difficult problems facing humanity, it can be used to solve the most difficult problems facing humanity.Compressing decades of scientific progress to a few yearsThe aim is to speed up the discovery of new medicines, disease detection and personalized treatment.
At the end of the founder's letter, the seven co-founders wrote:
"Our goal is to help humans navigate this technological evolution safely, to open up new areas of discovery, and to make the world a better place".This is the beginning of an era of miracles."Everyone who holds our shares, whether institutional or individual, will be a part of it".
Human beings have created intelligence, and intelligence has begun to produce the next generation of intelligence.
As of this evening, the route to return to ASI will be open to the public.With its own stock code, ANT。
RoboInd editorial channelResearch, robotics companies and market intelligence for cross-border sourcing.Explore the channel ↗
Share this
End of editorialExplore more from Global Robotics News ↗