On the first day of Anthropic’s launch, Dario and Daniela Amodei gave an exclusive interview to CNBC. Faced with the questioning of US$900 billion in computing power commitments and US$250 billion in cash reserves, how do they respond to the challenges of profit models, open weight competition and security rhythm? (Preliminary summary: The stock goddess spent 500 million US dollars on the first day of IPO to buy SpaceX! ARK Invest sold shares to raise money but also bought them) (Background supplement: Huang Renxun's sudden resignation to attack Anthropic IPO? The listing went ahead, but the first financial report became a tough battle) On the first day of Anthropic's listing, the two founders Dario and Daniela Amodei accepted an exclusive interview with CNBC and faced the market's sharpest doubts: 9,000 Is the $100 million computing power commitment a long-term cost or a growth engine? Dario revealed in the interview that these expenditures have been spread over many years, and the company tends to conservatively lock in usage and then replenish it from the spot market, and revealed that it may seek investment grade ratings and debt financing in the future. This is an in-depth conversation about AI capital expenditures, business models and security rhythms. TL;DR On November 19, the day Anthropic went public, co-founders Dario and Daniela Amodei were interviewed by CNBC host Andrew Ross Sorkin. Faced with approximately $900 billion in future computing power commitments, Sorkin put two questions together and asked: How will Anthropic achieve profitability? Is a purchase of this scale a phased investment, or is it a long-term cost of this business? $900 Billion Computing Power Commitment: A Big Bet Spread Over Many Years Dario explained that these commitments are not a one-time expenditure, but are spread over a period of many years. According to him, when viewed in the context of the company's existing cash and annual revenue, the spending pressure will be less than the first impression given by the total amount. It is difficult for companies to accurately predict computing power usage in advance, so they tend to conservatively lock in demand and make additional purchases from the spot market when actual usage exceeds expectations. The company was willing to make the big commitment because it believed the returns from stronger models had proven the value of such investments many times in the past. Funding sources are another key variable. With proceeds from the offering, Anthropic could have about $250 billion in cash, Sorkin estimated. Dario said he hopes the company can obtain an investment grade credit rating in the future and use debt financing at some point. He did not give a timetable for obtaining ratings, issuing debt or achieving profitability, nor disclosed a year-by-year payment schedule for the computing power commitments. Sorkin pointed out that open weight models are becoming increasingly popular, OpenAI is trying to reduce the price of Tokens, and Microsoft is also promoting its own models while providing Anthropic model services. As long as some functions can be replaced by cheaper products, the market will ask whether Anthropic's profit margins can be maintained. $250 Billion Cash Reserves and Debt Financing Ambition Dario’s judgment is that competition among frontier models mainly depends on capabilities. The model moves from chatting to programming, and then into scenarios such as biological research and robotics. Each improvement in capabilities may open up a batch of higher-value tasks. He believes that much of the industry's value will be in cutting-edge models that can accomplish these tasks. Meanwhile, Anthropic is also lowering pricing on its lightweight models. Dario said that the technology and computing efficiency improvements accumulated when training cutting-edge models can also be used to develop lightweight models. Taking Haiku 5.5 as an example, according to him, its price is about a quarter of Haiku 4.5 and much lower than the Opus and Fable 5 series. The interview gave pricing changes, and the cost and profit margin information of the relevant models have not yet been disclosed. Daniela said that Anthropic adopts a public benefit corporation (PBC) structure and is required by law to take into account the interests of shareholders and the public interest. The founding team chose this structure because they hoped to incorporate considerations of public impact into the company's decision-making when pursuing the economic value of AI. Open weight attack: Where is the moat of the cutting-edge model? Dario used driving to explain the company's R&D pacing. Assuming a new technology could push the car from 60 mph to 200 mph, the team could choose to drive to 90 mph first, allowing time to figure out how to increase speed safely. Here, he discusses situations in which models may in turn accelerate their own development. He believes that in the long run, investment in security will help win the trust of enterprise customers and may be converted into business advantages. When asked about AI security risks and potential liability, Daniela said Anthropic has a responsibility to train models in a safe manner and set up appropriate safeguards. She said working with enterprise customers gives the company a clearer picture of what risks the average organization actually encounters when deploying Claude. Sorkin mentioned that Huida is an investor in Anthropic and also participated in the issuance. However, the two companies have different positions on China chip policies. Dario said that Anthropic has a long-term cooperation with Huida in the data center field and will continue to adhere to its own policy positions. When asked by the host whether Huang Jen-Hsun should return as CEO, he said it was up to Huida to make its own decision. Public Benefit Company Structure: Balancing Shareholder Interests and Public Responsibility Daniela also listed partners such as Amazon, Google and Broadcom, believing that being able to cooperate with multiple suppliers and jointly design chips is a strategic advantage for the company. As for whether Anthropic will reduce its use of SpaceX computing power when the contract allows, Dario did not commit to future procurement arrangements. He only said that the existing computing power is of high quality and the cooperation between the two parties is smooth. NVIDIA Partnership and Differences in China's Chip Policy The construction of the data center also affects the power grid and local communities. Daniela said that the company is committed to bearing the difference in electricity prices that may arise from the data center’s electricity consumption and will communicate with the communities where the project is located. As for cooperation with the U.S. Department of Defense, she did not disclose new developments, only saying that the company hopes to continue its cooperative relationship with the U.S. national security department. Related reports: Taiwan’s Fubon Life Insurance plans to spend “up to US$700 million” to purchase Anthropic’s IPO. Anthropic’s ARR is revised down to US$800 billion in 2030. Can the 9.2x price-to-sales ratio sustain it? Anthropic IPO’s intended subscription reaches $500 billion! It is rumored that the IPO price will be raised to above US$80, and the price will be set in Taipei in the early morning of the 19th. Anthropic’s founder was interviewed on the first day of listing: What will support the US$900 billion computing power commitment? 〉This article was first published in DongZu BlockTempo "DongZu DongTen - the most influential blockchain news media".
Interview with the founder of Anthropic on its first day of listing: What supports the US$900 billion computing power commitment?
On the first day of Anthropic’s launch, Dario and Daniela Amodei gave an exclusive interview to CNBC. Facing 900 […] Interview with Anthropic’s founder on the first day of listing: What supports the US$900 billion computing power commitment? 〉This article was first published in DongZu BlockTempo "DongZu DongTen - the most influential blockchain news media".

This article may contain an automated translation. The original source is linked for verification.



