Anthropic CEO's Latest Speech Analysis: How Did They Achieve Over 4.5 Billion Yuan in Revenue in Three Years?
Recently, Anthropic CEO Dario Amodei responded directly in an interview: "I am not a doomsday believer; I am one of the people who understands the benefits of AI technology best." Dario Amodei holds a PhD in Biophysics from Princeton University, and during his doctoral studies...
Recently, Anthropic CEO Dario Amodei responded directly in an interview: I am not a "doomsday theorist," I am the one who understands the most. AI One of the people who benefit from technology.
Dario Amodei holds a PhD in biophysics from Princeton University. During his doctoral studies, he earned the Hertz Fellowship (one of the most prestigious research scholarships in the United States) and then spent several years as a postdoctoral researcher at Stanford Medical School.
He previously worked at Baidu, Google, and Open.AI.
In OpenAI During this period, he took the lead GPT-2 and GPTThe development of -3 and the introduction of RLHF technology, allowingLarge ModelFor the first time, I learned to communicate according to human intentions.
Until 2021,Dario He founded Anthropic with his sister Daniela and several core members. In three years, they built it into one of the world's most competitive companies.Large ModelOne of Claude series.
In the interview, which lasted over an hour, he bluntly stated that AGI is a pseudo-concept, and discussed scaling principles, profit logic, and more.open sourceBubble… I listened carefully and summarized 11 key points, which may change your perspective on it as well. AI My vision for the future.
Not a doomsday theorist, but one of the people who understands the benefits of technology best.
AI The speed of evolution is far faster than imagined. A few years ago, models couldn't even utter a single sentence; now they can perform tasks at the doctoral level.AI It is gradually infiltrating the real economic system.
Dario has always believed in the power of scaling laws. He understands that no one can truly predict the future, but some things must be said in advance.
He did not deny AI He has enormous potential, and may be more optimistic about the changes it can bring than anyone else. It is precisely because he sees those benefits clearly that he feels a greater responsibility to remind the world not to ignore the underlying risks.
AGI is a dopamine decoy, not worth mentioning.
Dario believes that AGI (Generic Git) is a general-purpose API.artificialintelligent)and ASI (superartificialintelligent"(")" is a vague and marketing-sounding term. He refused to use these terms and instead focused on improving the model's real-world capabilities.
Although he doesn't use these terms, he is still one of the few in the industry who... AI One of the people with the shortest expected timeframe for skill advancement, and also the most optimistic.
The real explosion may happen in the next two years.
Dario Amodei said,Large ModelIts capabilities are entering the second stage—reinforcement learning, reasoning, and computational abilities have been greatly enhanced, especially in mathematical and coding tasks, where it is approaching professional levels.
Most people don't realize that this is a...Exponential curve.
Assuming performance doubles every six months, the changes may not be noticeable in the early stages, but once the critical point is approached, the growth will defy intuition.
For example, Anthropic went from zero revenue in 2023 to over $4.5 billion in the first half of 2025, repeating the trajectory of the internet in the 1990s.
AI The real explosion may only be two years away.
AI Coding capabilities are leaping exponentially.
Encoding is Claude One of the fastest-improving capabilities for models.
Anthropic is not a company focused on developing development tools, but it has made great strides in this area: 18 months ago, its model only scored 3% in programming benchmarks, but now it has soared to 72%-80%.
Internally, Anthropic's code is mostly composed of... Claude The model is written directly, or at least the model is involved in the writing process. Other companies also exhibit similar practices.
In terms of profitability, progress is also accelerating. The model is entering a "self-development" phase.
The secret to model evolution lies in talent density.
Each generation Claude The release of a model will involve improvements in architecture, data, and training methods, all of which are part of new technologies.
Anthropic doesn't often discuss details publicly, but continuous technological innovation is key to improving model performance. To this end, Anthropic strives to maintain a high density of talent, as this is essential for inventing new technologies.
No matter how high the offer, you can't buy our team.
Anthropic's core team has an extremely low turnover rate; it's not that no one is trying to poach them, but rather that no one can actually poach them.
Dario made it clear that they would not rely on "premium pay" to retain staff, nor would they engage in individual negotiations. Systematic job levels and uniform compensation are part of Anthropic's long-standing culture.
“We don’t retain people with money, but with a shared vision,” he said. Passion, a sense of mission, and long-term commitment—these are things that cannot be priced. Even if the other party were Zuckerberg, they might not be able to buy him.
Losing 3 billion isn't about losing money; it's about betting on the next generation. AI
Anthropic expects to lose $3 billion this year, which sounds alarming, but is actually quite normal.
Dario likens model development to an investment project: Suppose that in 2023, training a model costs $100 million, and after deployment, it earns $200 million, resulting in a profit of $100 million that year. However, in 2024, another $1 billion is spent training a new model. Even if the old model continues to generate profits, the company will still show a loss on paper. Each model, viewed individually, is profitable, but the company as a whole does not appear profitable due to continuous investment in the next generation of R&D.
This model is prevalent in the industry: as long as the model continues to improve, companies will increase investment to drive business growth.
If the model's performance stagnates at some point, costs will decrease and profitability will steadily improve; otherwise, both input and revenue will continue to grow exponentially.
open source≠freeIt is not the key to determining the outcome.
open sourceWhether or not it is not important. Compared to the superficial freedom of "open weights," Dario values the model's performance on the task more: Is the model useful? Can it run fast? Is it cost-effective? Is it suitable for the business scenario?
Even if you obtain the parameters, deploying them to the cloud for inference still requires significant investment, optimization, and persistent effort. Dario focuses on who performs better on the task; that's where the wins. It prioritizes results, not just labels.
In other words: whoever is easier to use wins.
We're using 20 billion to compete with others who have 100 billion.
Anthropic's revenue growth over three years has been astonishing, from 0 to 4.5 billion. Dario believes this is due to "talent density" and "capital efficiency," rather than simply throwing money around. He said, "What others can do with 1 billion, we can do with 100 million."
API revenue makes up the majority; we're betting on the enterprise market.
Anthropic derives 60%-75% of its revenue from APIs, with the remainder coming from... Claude Applications. Dario has clearly stated that enterprise-level applications are the future, especially in high-value scenarios such as law, finance, and pharmaceuticals.
Technical measures were implemented prior to ensuring safety.
He recalled his time at OpenAI leading GPTThe experiences of -3 and RLHF indicate that capability and alignment cannot be separated from research. (Leaving Open)AIThe reason was a difference in philosophy regarding governance and pace. Anthropic was founded to advance security in a way he trusted. AI.
AI Safety should not be reduced to a corporate slogan or a bargaining chip, but rather a responsibility that the entire industry must share.
Dario pursues a "responsible expansion strategy," proactively disclosing its hazard capacity assessments and constitutionally mandated disclosures. AIInterpretability research is not for monopolizing security benefits, but for putting the entire industry on a more stable track.
He disagrees with the pessimism of cybernetics and rejects the fanaticism of accelerationists. He is more concerned with whether the industry has sufficient testing and technical reserves as model capabilities continue to approach high-risk zones. If the risk of loss of control exceeds the controllable threshold, he will be the first to call on the entire industry to slow down.
Amidst the exponential surge of technological advancements, he chose a less-than-ideal path—pushing the entire industry to "compete for higher positions."
In this interview, Dario Amodei spoke truths that many people avoid discussing. He didn't talk about AGI, nor did he mention "super...intelligentInstead, they focused on improving the specific capabilities of the model, discussing reasoning, computation, and reinforcement learning.
It wasn't because he was pessimistic, but because he foresaw the potentially enormous consequences of this exponential curve. His concern wasn't that the model wouldn't succeed, but that it would succeed too quickly.
He wasn't trying to stop others from doing it. AIThe goal is not for everyone to compete at a higher level, but for everyone to be the first to launch, and for everyone to do the best without making mistakes.
I quite agree with his views on talent. Passion, responsibility, and long-term commitment are things money can't buy. No matter how high the offer, you can't retain people who truly share your vision. What do you think?
Get theDario's 1-hour interview video (original video)Scan the QR code to follow and reply: 20250811