Anthropic founder's heavyweight interview: AI is at the tail end of exponential growth, and by 2026 will usher in a "genius nation in the data center," revenue is soaring at a speed of 10 times.

On the eve of the exponential explosion of AI technology, Dario Amodei, the leader of Anthropic, made a shocking prediction that we are in the "twilight of exponential growth", and by as early as 2026, humans will enter a "genius nation in a data center" composed of tens of thousands of top brains. Recently, Dario Amodei, the CEO of the large model unicorn Anthropic, in a deep interview with Dwarkesh Patel, rare disclosed the company's amazing revenue growth expectations, and elaborated on the timetable for AGI (general artificial intelligence), the financial logic of computational investment, and the geopolitical risks. Amodei believes that AI technology is at a critical point of transition from quantitative to qualitative change, and the next 2-3 years will determine the direction of human future for the next two hundred years. AI is at the end of exponential growth Dario Amodei pointed out at the beginning of the interview that we are approaching the end of the AI exponential growth curve, and the world has not fully recognized this transformation. From GPT-1 to today's professional-level models, AI has made a leap from a "smart high school student" to a "doctoral level", even surpassing in fields such as programming and mathematics, the underlying expansion law has never been invalidated, and the input of computational power and data continues to yield clear benefits. The magic of exponential growth lies in the outbreak at the end. Dario stated that the phenomenal growth of Anthropic's annual revenue by 10 times, Claude Code doubling engineer productivity, and the rapid breakthrough of model context length and generalization ability are all confirming the signs of "approaching the end". This growth is not just about stacking parameters, but an upgrade of intelligent nature - from data fitting to autonomous generalization, AI is completing the last few key pieces of the puzzle of capabilities. "A Country of Geniuses in a Datacenter": Redefining 2026 Amodei proposed a striking concept in the interview - "A Country of Geniuses in a Datacenter". He reviewed the technological evolution of the past three years, believing that AI models have evolved from "smart high school students" to "professionals". He boldly predicted that by 2026 or 2027, the intellectual level, depth of knowledge, and logical reasoning ability displayed by a single model will not only be equivalent to a Nobel laureate, but equivalent to a collection of tens of thousands of top geniuses working together. In terms of the certainty of this timetable, Amodei showed high confidence: "I am 90% confident that we will achieve this vision within 10 years, and I think there is a 50/50 chance it will happen in the next 1-2 years." He pointed out that the only variable could come from geopolitical disasters (such as interruptions in the chip supply chain) or severe social unrest. Revenue soaring: A "terrifying" curve from $100 million to $10 billion In terms of the most interesting financial data for the market, Amodei disclosed Anthropic's staggering growth curve. He revealed that the company's revenue is experiencing a "bizarre 10x per year growth". Amodei said in the interview: "In 2023, we grew from 0 to $100 million; in 2024, from $100 million to $1 billion; and in 2025, we are expected to reach $9 billion to $10 billion. This exponential growth roughly matches my expectations, and even in the first month of this year, we added billions of dollars in revenue." Amodei emphasized that although the adoption of AI by enterprises needs to go through a lengthy process of legal review, compliance checks, and the like due to the lag in economic diffusion speed, the improvement in technical capabilities is driving this crazy growth curve. Computational power gamble and bankruptcy risk: CEO's financial balancing act Faced with such a certain technological future, why not borrow trillions of dollars now to stockpile chips? Amodei provided a very realistic financial explanation: scaling up computational power must be tied to revenue growth with precision, otherwise the company will face catastrophic risks. "If I predict that there will be a trillion-dollar demand in 2027 and buy $1 trillion worth of computational power in advance, but if the demand outbreak is delayed by one year, or the growth rate drops from 10 times to 5 times, there is no hedging mechanism to prevent the company from going bankrupt." Amodei explained that this investment return based on the "logarithmic law of return" requires precise accounting. He pointed out that Anthropic's current strategy is to be "responsibly aggressive", with computational power investments large enough to capture huge upside potential, but if the market outbreak is delayed, the company can still survive with its high gross profit and cash flow from enterprise business. He expects Anthropic to achieve profitability around 2028, when AI will become one of the most profitable industries in history. The endgame of software engineering: From coding to replacing engineers In terms of specific implementation scenarios, Amodei sees programming as the fortress that AI first conquers. He divides AI's evolution in the field of software engineering into three stages: Phase One: models write 90% of the code lines (already achieved). Phase Two: models handle 90% of end-to-end tasks, such as bug fixes, cluster configuration, and documentation writing. Phase Three: models have "workplace experience", able to understand the context of complex code libraries and set technical directions. Amodei predicts that within 1-3 years, AI will be able to handle all responsibilities of a senior software engineer. "This does not mean that engineers will be unemployed, but there will be a huge surge in productivity. What models can do now is not just filling in code, but directly taking over high-difficulty tasks such as GPU kernel writing." This article is reprinted from "Wall Street SeeThis was a guest article by Dario Amodei, CEO of Anthropic, interviewed by Dwarkesh Patel.ing" by author Dong Jing and translated by GMTEight editor Chen Siyu.