ChainCatcher
ChainCatcher

Exclusive interview with Amodai brother and sister on their first day of listing: How come the promise of US$900 billion in computing power is not a big deal, and why speed control is a smarter business

Editor's note

On November 19, Anthropic officially listed on Nasdaq (ticker: ANT). The issue price of this IPO was US$83, raising approximately US$81.3 billion, setting a new historical record in the global capital market. The final closing price was US$97.17, an increase of approximately 17% from the issue price, and the closing fully diluted valuation was US$1.77 trillion.

This is an exclusive interview accepted by the founders, Amodai brother and sister, on the day of listing. The well-known CNBC host Andrew Ross Sorkin threw out a series of the most pointed market torture on the spot: How to fulfill the promise of up to 900 billion US dollars in computing power? Will proactive “speed control” stifle business growth? Will open source siege and low-priced tokens equalize profits? And how do they deal with Nvidia, Musk and the Pentagon?

The following is a selection of the conversation. The full text is about 4,700 words and takes about 8 minutes to read.

Wonderful views in this issue

Scaling Laws is just a smooth mathematical curve on paper, but when it is realized in physical reality into crushing economic productivity, it is such a shock that even we ourselves were not convinced at first.

Even if we had the self-evolving technology to push our cars to 200 mph, we’d rather keep them at 90 mph. Use the extra computing power to build security barriers - in the long run of business, only models that are absolutely reliable and trusted by large global companies can conquer the market.

The competition among large models is not mainly about price, but about quality. The exponential leap in cognition means that once a critical point is crossed, models can leap from writing code to conquering molecular biology and sophisticated embodied intelligence. Business value will only converge towards the Frontier model.

We signed a long-term computing power agreement of 900 billion US dollars, and our confidence lies in the cash on the account and the annual revenue; past history has repeatedly proved that focusing on models with stronger cognitive capabilities has very high returns, and it is worth it in the end.

The general public cannot pay for the surge in data centers. We have a strict rule: any additional price difference caused by the local power grid due to the data center's connection to the grid will be borne by Anthropic out of its own pocket.

I wrote a core principle in my letter to investors, which is also the core of what we do within the company: always lower market expectations, and then overdeliver (Underpromise and overdeliver).

You must not be controlled by others on the basis of computing power. Designing the world's best chips with so many partners, from Nvidia, Amazon, Google to Broadcom, is a real strategic advantage for us.

The law of scale is fulfilled: the curve on paper, the cognitive explosion in reality

Question: When you first founded Anthropic, almost no one in the industry understood Scaling Laws better than you. But today, a few years later, facing a market capitalization starting from "trillion US dollars", was all this what you expected?

Dario Amodai (Co-founder and CEO of Anthropic): Frankly speaking, it is one thing to see theoretical deductions, but it is another thing to see reality implemented. In the laboratory, the law of scale is nothing more than a coordinate line on a scratch paper, which predicts in mathematical language: as long as parameters and computing power are stacked, the model can unlock the corresponding cognitive level and economic processing power at a specific node. However, it is easy to draw a line on paper, but it is completely different to watch that line transform into a sweeping productivity shock in the real world. Theories have long deduced the scale that this technology should have, but when it was actually realized, we were still deeply shocked.

Q: Anthropic becomes the largest “public benefit company” (PBC) in the capital markets. For the secondary market investors entering the market today, what exactly does this structure of “writing social welfare into the charter” mean?

Daniela Amodai (Co-Founder and President, Anthropic): Our mission is to make things that are good for everyone. As a PBC, we are legally required to maintain an absolute balance between shareholder returns and the public interest. When our co-founders founded the company, we learned about the concept of a public interest company and immediately felt that this was exactly what we wanted to do. We must weld the safety mechanism and public interests within the organizational framework from the DNA of the company as a legal person. For sophisticated investors, this is by no means a commercial liability, but a strategic hedge that can save companies from self-destruction in the wave of disruptive technologies.

Speed control philosophy: Why do you drive to 90 miles per hour when you can obviously drive at 200 miles per hour?

Question: Investors are most concerned about growth, but you have repeatedly emphasized “pacing”. If technology evolution is actively slowed down for the sake of safety, how should the market accept this seemingly anti-commercial setting?

Dario Amodai: This contradiction exists objectively, but as the underlying technology becomes more and more powerful, the entire industry will hit this wall sooner or later, and there is no escape. The difference is that we choose to face it more honestly and earlier.

For example: you are driving a car at a speed of 60 miles per hour, and suddenly you invent a power engine that can instantly speed the car to 200 miles per hour. In the field of AI, this is when the model begins to possess the ability of self-recursive evolution (Recursive Self-Improvement). The so-called speed control means that we know that we can reach a speed of 200 miles per hour, but we actively only drive 90 miles per hour. We took out the redundant speed and computing power and concentrated on solving security alignment.

In the long run, this is a very far-sighted business strategy. The final outcome of the large model track is not a competition of who can run the craziest, but a competition of who is the most deserving of the core business entrusted to global enterprises. Whoever can maintain absolute safety at extreme speeds will be able to swallow the largest enterprise-level market cake.

Question: It was listed today. What kind of stock price increase would be considered "success" in your eyes? 10% or 20%? If the increase is too large, will it deviate from your expectations?

Dario Amodai: When it comes to stock price fluctuations in the secondary market, compliance requires us to remain silent.

Daniela Amodai: The stock price is that of the day, but Anthropic is a classic ultra-long cycle company. Today is an exciting milestone, but it's just an aid station on a long run. The sun will rise tomorrow morning as usual, and our R&D and bottom-line principles will not change at all. For true technology long-termists, the fluctuations in volume caused by listing are far less important than the evolution of the underlying technology.

Frontier Returns: What matters is quality rather than price, and most of the value is deposited in the frontier

Question: The current competition is unprecedentedly intense: open source models are under siege, OpenAI is desperately trying to lower the Token unit price, and Microsoft is selling your model while pushing its own affordable model. Will the gross profit margin of large models be completely flattened by this round of price war?

Dario Amodai: We must clearly see the two underlying business logics here.

First, traditional products compete on price, but the core battlefield of large models is always at the “frontier”. This is a market that takes all quality because the economic value released by a leap in cognitive ability is non-linear. Going up the cognitive ladder, the things that the model can solve will jump from the most basic chat questions and answers, to the writing of underlying architecture, to automated molecular biology research and development and high-precision robot control. The business volume that cutting-edge models at the top of technology can create is dozens of places ahead of mediocre models at the back. Almost all excess profits will eventually be siphoned into the hands of cutting-edge models.

Second, only by mastering the most hard-core cutting-edge technologies can you have the ability to reduce dimensions and build affordable models. For example, the price of our Haiku 5.5 has been reduced by 4 times compared to the previous generation Haiku 4.5, and it is many times cheaper than the top Opus and Fable 5 families. The "computing power multiplier" we accumulated while sprinting for cutting-edge large models can be directly reused on lightweight models. In an affordable market far away from the technological frontier, we can still build a moat with our leading low cost.

Computing power bet: With 900 billion in long-term purchases, how can you guarantee that it will not explode?

Q: The energy throughput of data centers is approaching physical limits, even triggering a violent rebound in local power grids. How do you ensure your infrastructure can continue to expand in the face of increasing energy scrutiny?

Daniela Amodai: Our underlying logic is: we must not let the cost of AI expansion fall on the general public. Many local residents are worried that the connection of data centers to the grid will push up local electricity prices. In this regard, Anthropic has made a clear commitment: Any electricity price difference that occurs at the grid level will be paid by us. When high-tech companies enter any community, they must ensure that the vital interests of the local people are not harmed. This is the only source of legitimacy for the long-term implementation of computing infrastructure.

Q: But you are saddled with a long-term computing power agreement of up to $900 billion. Will this astronomical capital expenditure drag down cash flow before reaching full profitability? Is it a phased investment or an irreversible permanent norm?

Dario Amodai: First of all, we need to clarify the cycle: these 900 billion are spread evenly over a very long period of several years in the future. Compared with the huge cash reserves on our books and the current explosive annualized revenue, this long-term commitment is completely within the healthy boundary.

In terms of computing power prediction, we would rather adopt a conservative lower limit strategy and go to the spot market for additional purchases when encountering explosive demand. More importantly, between computing power investment and technical output, we have seen time and time again in history: spending a lot of money to train models with stronger cognitive abilities is worth it in the end. So we are willing to invest such a large amount of computing power.

Q: Including the funds raised from the IPO, you probably have about $250 billion in cash on hand. How helpful will this reserve be to your subsequent efforts to obtain investment-grade credit ratings and develop low-cost bond issuance?

Dario Amodai: We still have to see what the outcome will be. We hope to achieve an investment grade rating at some point in the future, and debt is also a tool available to us that we would like to use at some point in the future.

Giant competition and geographical reflection: Nvidia is not the only trump card

Question: Regarding the deep security responsibility of the model, even Microsoft CEO Satya Nadella recently admitted that "computing power alone does not necessarily lead to security alignment", and even requires the establishment of a special mechanism to reversely discipline AI. As a company that has always regarded security as a selling point, can your enterprise-end defense line stand the test?

Daniela Amodai: The experience gained from going deep into the front line of an enterprise is priceless. Anthropic bears the ultimate responsibility from beginning to end: ensuring that the underlying pre-training process is foolproof, while installing the strictest safety guardrails on the model. By co-creating with enterprise-level customers, we know better than any ivory tower laboratory divorced from the business what real risks will be triggered by calling Claude in the real business world.

We have been a "safety first" company from day one, which is the whole reason we started our own business in the first place. The destructive potential of this technology is as real as its commercial miracles. Whoever can completely seal systemic loopholes can control the lifeline of the enterprise market.

Question: The secondary market environment has been extremely severe in the past few weeks, and Jen-Hsun Huang and Nvidia have also been at the center of the storm of public opinion. Faced with this kind of turmoil, how did you deal with investor doubts during the road show?

Dario Amodai: We are very confident. Technological breakthroughs are measured in years, while market noise is measured in weeks or even days. Those short-term violent fluctuations cannot shake the competitive barriers we are building.

I emphasized one sentence in my letter to investors. This is the core principle of what we do within the company, and it is also what we tell them when we recruit people: always lower expectations and speak with excess results (Underpromise and overdeliver).

Question: But Nvidia is your investor and also participated in this listing. Regarding the policy of export control of semiconductors to China, you and Huang Renxun have serious differences in their public opinions. How do you deal with this complex competitive tension? Should Jen-Hsun Huang return to the front line?

Dario Amodai: We have cooperated with NVIDIA for many years and jointly built a world-class data center. We are extremely close business partners. But in the mature business world, healthy collaboration does not mean being in lockstep on all public policy issues. Our policy propositions are entirely based on our independent deduction of the global security landscape and the technical path of general artificial intelligence; we stick to our position and fully respect the demands of our partners. As for Jen-Hsun Huang's personal position, we will not comment. That is NVIDIA's decision.

Daniela Amodai: In addition, in terms of the underlying chip ecology, Anthropic’s core moat is “thorough multi-source hedging”. We are not only deeply tied to Nvidia, but also customized self-developed chips with Amazon, deeply aligned with Google, and working side by side with Broadcom. Not tying our fortune and life to a single giant gives us strong strategic flexibility in our computing power architecture.

Question: Speaking of computing power diversification, Musk is also one of your computing power suppliers. There are rumors that after the contract expires, you may reduce the use of some SpaceX computing power and transfer it elsewhere. Is this true? Also, how are you and the Pentagon progressing now?

Dario Amodai: No one can say what the future will bring. The quality of the computing power delivered by SpaceX is very high, and the current engineering collaboration is extremely smooth.

Daniela Amodai: Regarding defense cooperation, there are currently no details that can be disclosed publicly. But what is clear is that since its inception, Anthropic has maintained an extremely close and efficient cooperation mechanism with national security agencies. Ensuring that Claude defends technological sovereignty and national security at the highest standards is a core red line for us, and we expect this cooperation to continue.

Question: Brother and sister teamed up to create a trillion-dollar technology giant, which is extremely rare in the entire business history. What does it mean for the two of you to move past this moment and strip away all the halo?

Daniela Amodai: Before entering the venue, both of us were still feeling: This is a miracle that belongs to the whole team, and we are extremely proud of the team's efforts. But at the end of the bell, we are still the same company and looking forward to continuing to advance our mission. If I have to talk about any small private wishes after the listing - it is that after this storm of fame and fortune subsides, we, brother and sister, can have more time just for the two of us.

Source: CNBC, host Andrew Ross Sorkin in conversation with Anthropic co-founders Dario and Daniela Amodei, launch day, November 19, 2026, https://www.cnbc.com/video/2026/11/19/watch-cnbcs-full-interview-with-dario-and-daniela-amodei.html