Anthropic CEO Advocates Strategic Caution in OpenAI Debate  ## Metadata - Author: [[Hard Fork AI]] - Full Title: Anthropic CEO Advocates Strategic Caution in OpenAI Debate - Category: #podcasts - URL: https://share.snipd.com/episode/4d9a301f-2753-4622-8b4e-788d4790706a ## Document Notes ## Highlights - **Timing Uncertainty Creates Financial Risk** - Dario Amodei warns that long‑term AI fundamentals are strong but timing of value is highly uncertain. - That uncertainty drives operational and financial risks as firms race to outspend rivals. Transcript: Jaeden Schafer This all happened at the New York Times Dealbook Summit. The CEO of Anthropic, Dario Amadei, he gave one of the most candid assessments of kind of this financial AI bubble. He was talking about what are some timing errors in the industry? What are some unwise risks? I think in the process of all of this, he threw a lot of shade at an unnamed rival that obviously is OpenAI, but he never actually said their name. So the big question that he was asked is AI in a bubble. And he says that the answer is a lot more complicated rather than just saying, you know, the whole sector is overvalued or underpriced. He made, I think, what I would call a much subtler argument. He said that he remains fundamentally bullish on the long term economic potential of AI. But he also stressed that the timing of when value materializes is not very certain. He said that the uncertainty is, you know, creating kind of these operational and financial dangers. Here's a direct quote. He said, quote, there is an inherent risk when the timing of the economic value is uncertain. Companies essentially are taking these really aggressive bets, not just to compete in the market, but to stay ahead of these quote-unquote authoritarian adversaries like out of China. This is a reference because China's really accelerating their AI development, and the challenge according to Amadei is that some firms are not quote-unquote managing that risk well. I think this sets up a tension that is kind of has this classic bubble look to it there's these strong long-term uh you know fundamentals but they're also combined with these short-term Incentives that push companies to overextend i think all of the growth they're seeing is very rapid and the trajectory of ai kind of hides the fact that the industry has not yet proven The return on investment curve for frontier model development, or also for a lot of the compute expansion, like we see OpenAI doing these multi-billion dollar compute, you know, building Out these huge data center projects. And so I thought, I think a lot of this is kind of a bit of a timing trap. Amadei spent a lot of his answer discussing the lag time between investing in data centers. ([Time 0:01:13](https://share.snipd.com/snip/c4500151-a0d5-471e-9bac-e20a2657aa9a)) - **Geopolitics Fuels Aggressive Bets** - Amodei frames aggressive spending as a response to geopolitical competition, notably China's acceleration. - This pressure pushes some companies to overextend and 'YOLO' risk profiles, he says. Transcript: Jaeden Schafer He said, quote, there is an inherent risk when the timing of the economic value is uncertain. Companies essentially are taking these really aggressive bets, not just to compete in the market, but to stay ahead of these quote-unquote authoritarian adversaries like out of China. This is a reference because China's really accelerating their AI development, and the challenge according to Amadei is that some firms are not quote-unquote managing that risk well. I think this sets up a tension that is kind of has this classic bubble look to it there's these strong long-term uh you know fundamentals but they're also combined with these short-term Incentives that push companies to overextend i think all of the growth they're seeing is very rapid and the trajectory of ai kind of hides the fact that the industry has not yet proven The return on investment curve for frontier model development, or also for a lot of the compute expansion, like we see OpenAI doing these multi-billion dollar compute, you know, building Out these huge data center projects. And so I thought, I think a lot of this is kind of a bit of a timing trap. ([Time 0:02:10](https://share.snipd.com/snip/fab75f2d-a87e-40df-a6e4-755826d00947)) - **Plan Conservatively For Hardware Cycles** - Model conservatively for hardware depreciation and weak demand scenarios to avoid being upside down. - Amodei says Anthropic uses cautious chip assumptions and slower growth planning to manage risk. Transcript: Jaeden Schafer And so I thought, I think a lot of this is kind of a bit of a timing trap. Amadei spent a lot of his answer discussing the lag time between investing in data centers. And then actually when you see the returns on that investment i think the economic value of ai might look really abrupt or gradual but the data center construction timelines are fixed Right they're very capital intensive um they're very unforgiving uh he said quote there is a genuine dilemma that we try to manage as responsibly as we can then there are some players Who are yoloing who put the risk dial too far and i'm very concerned so i think with all of this uh you know saying that these certain companies are yoloing he's obviously referring to Open ai um i think he's critiquing them in a large way because of these massive infrastructure deals that they're building uh they don't know how much demand they will necessarily have. They're kind of forecasting. And maybe in the future, we'll be like, wow, they were geniuses for spending all that money back then. And we really need it now. But, you know, it's definitely seems like a massive risk. But also, like, if I was to give a bullish spin on that as well, for my personal opinion, it is that they're able to raise a ton of money right now because of the way the market is and investing That into data centers, which are, you know, I mean, these are essentially an asset that is not, if it's not useful to them, it's useful to others. So I think by investing in data centers, like if they're just building out their portfolio, their assets, their valuation, because those are just tangible things that are worth a lot Of money. So I do think those are data centers are very worthwhile to invest in today. If they're trying to build them faster and if they're overspending in order to try to speed up the process beyond what the market would typically compensate them for, then I think that's Obviously not sound business practices. But assuming they're doing that in a responsible way, I think those might be good investments. But this is what Amadeo said. He also kind of talked about the industry's anxiety. Everyone's kind of worried about this chip depreciation timeline. GPUs continue to work for a lot of years after they're created. But I think the rate that new models are coming out is making the old ones economically obsolete, which is definitely an issue. He said, the issue is not the lifetime of the chips. The issue is new chips come out that are faster and cheaper. And so the value of old chips can go down somewhat. That's interesting. We talk about the depreciation of chips over at Google or over at Meta. There was a lot of controversy that Meta was cooking the books and, you know, depreciating their chips on too long of a timeline, saying that they were useful for longer than they were. Well, the chips themselves, he makes a great point, are quite useful. It doesn't mean that like the new chips are so much faster and cheaper that it essentially makes the old ones useless, even if they're technically capable of continuing to compute things. I think if a company over invests in hardware, they lose the competitive value before it's fully monetized. And I think the economics of these frontier models can unravel very quickly if that happens. So, um, Anthropic, Dario says, is taking a really conservative approach to chip assumptions. They're modeling much weaker scenarios to avoid being caught upside down by rapid hardware cycles. ([Time 0:03:15](https://share.snipd.com/snip/da87a0f2-431d-483a-b965-247c1e6f5f29)) - **Hardware Can Become Economically Obsolete Fast** - Rapid model and chip improvements can make existing hardware economically obsolete before it's fully monetized. - Overinvestment in hardware risks unraveling the economics of frontier models quickly. Transcript: Jaeden Schafer He also kind of talked about the industry's anxiety. Everyone's kind of worried about this chip depreciation timeline. GPUs continue to work for a lot of years after they're created. But I think the rate that new models are coming out is making the old ones economically obsolete, which is definitely an issue. He said, the issue is not the lifetime of the chips. The issue is new chips come out that are faster and cheaper. And so the value of old chips can go down somewhat. That's interesting. We talk about the depreciation of chips over at Google or over at Meta. There was a lot of controversy that Meta was cooking the books and, you know, depreciating their chips on too long of a timeline, saying that they were useful for longer than they were. Well, the chips themselves, he makes a great point, are quite useful. It doesn't mean that like the new chips are so much faster and cheaper that it essentially makes the old ones useless, even if they're technically capable of continuing to compute things. I think if a company over invests in hardware, they lose the competitive value before it's fully monetized. ([Time 0:05:03](https://share.snipd.com/snip/4ab886fe-01a4-4764-ba01-9fae2e4591d0)) - **Anthropic's Rapid Revenue Trajectory** - Jaeden Schafer recounts Anthropic's revenue surge from $0 to $1B then to a projected $8–10B in 2025. - He highlights that despite hypergrowth, Amodei still plans for conservative outcomes. Transcript: Jaeden Schafer Also, they're growing much slower than OpenAI and they have less users than OpenAI. So to be fair, you know, taking a more conservative approach is also they have a more conservative growth rate of their company. What I will say is that Anthropics revenue has grown at a pace that not many tech companies have seen. So when you compare it to OpenAI, you're like, oh, they're smaller than OpenAI, but they're still crushing basically everybody else. They went from $0 to $100 million in 2023. They then went from $100 million to $1 billion in 2024. And they're projected to hit $8 to $10 billion by the end of 2025. ([Time 0:06:24](https://share.snipd.com/snip/c301ea51-063b-4eb1-be23-962562097bc4)) - **Even Leaders Can't Predict Demand** - Leading AI firms lack clear visibility on demand curves, enterprise adoption speed, and revenue durability. - Amodei admits even top companies cannot reliably forecast near‑term growth trajectories. Transcript: Jaeden Schafer Like, it's so interesting to see someone that just went from zero to 100 million, 100 million to a billion and a billion to 10 billion be like, I don't know if we're going to be at 20 billion Or 50 billion in a year. Like, it's so crazy, but also it's true. It just goes to show that even the people leading these top companies don't know. I actually appreciate how candid he was about that because I feel like with opening eyes, Sam Altman almost like tries to appear like he's like, you know, like this wizard, this fortune Teller. He's like, yeah, if you saw the things I saw, like things are going to be crazy. That's why we got to spend all this money on all these data centers. But at the end of the day, they don't really know. They're just like hoping and guessing and, uh, you know, making, making bets. So I do appreciate how, um, candid the CEO of Anthropic has been. I think all of this, uh, all the conversation that he, he recently had inevitably, inevitably drifted towards OpenAI. He never named them specifically, but last month, OpenAI had this big public relations mess where the CFO suggested that the US government should backstop the company's infrastructure Loans. A lot of people were very critical of that. They have to walk back the statements. He, you know, when asked about a lot of this, said that he said competitors who, quote, like big numbers or are constitutionally inclined to take enormous risks landed heavily. He said, quote, we think we're going to be OK in basically almost all worlds. ([Time 0:07:52](https://share.snipd.com/snip/e90257a7-95ae-49cc-9539-d5ab3c7a5d05)) - **Competitors' Risk Appetite Is Questioned** - Amodei implicitly critiques competitors that take enormous risks and may not be resilient. - He says Anthropic expects to be OK in most scenarios but won't vouch for others. Transcript: Jaeden Schafer He never named them specifically, but last month, OpenAI had this big public relations mess where the CFO suggested that the US government should backstop the company's infrastructure Loans. A lot of people were very critical of that. They have to walk back the statements. He, you know, when asked about a lot of this, said that he said competitors who, quote, like big numbers or are constitutionally inclined to take enormous risks landed heavily. He said, quote, we think we're going to be OK in basically almost all worlds. I cannot speak for other companies. So it seems like he is taking shots at opening eye, doesn't know if they're going to be OK. Obviously, there's his main competitors who's talking his own book. ([Time 0:08:44](https://share.snipd.com/snip/c5440fdc-5d70-4c3a-891b-2a7585645c90))