Anthropic Vs. OpenAI: How Safety Became The Advantage In AI

What is discussed, and when

1. 0:00 – 5:00: Anthropic, Amazon, Country of Geniuses, Daniela Amodei
2. 5:00 – 10:00: Anthropic, Google LLC, Bessemer Ventures, Bridgewater
3. 10:00 – 15:00: Anthropic, Google LLC, Amazon
4. 15:00 – 20:00: Anthropic, Google LLC
5. 20:00 – 22:39: Anthropic, Country of Geniuses

Notable quotes from "Anthropic Vs. OpenAI: How Safety Became The Advantage In AI"

• Anthropic is one of the key AI names eyeing a potential IPO this year, and it's looking to outmaneuver big spending rivals by showing you don't need the most chips to win.

  about Anthropic · at 0:05

• It launched a $50 billion national infrastructure build out plan. Amazon's $11 billion AI data center in Indiana, one of the largest in the world, is now live. Anthropic has really partnered with us.

  about Amazon · at 1:12

• In late 2020, siblings Dario and Daniela Amodei, along with a core group of senior leaders, left OpenAI convinced that the next era of frontier AI would be decided as much by…

  about Daniela Amodei · at 2:07

• Anthropic's model family is called Claude. It's similar to OpenAI's ChatGPT or Google's Gemini, but trained differently with different data, different safety approaches, and different goals.

  about Google LLC · at 3:22

• Bessemer Ventures Sameer Dholakia says the bet made sense for one simple reason, enterprise customers don't churn the way consumers do.

  about Bessemer Ventures · at 6:32

• Enterprise Novo Nordisk, the Norwegian sovereign wealth fund, the world's largest, plus Bridgewater, Stripe and Slack all running Claude at scale. It tripled its international…

  about Bridgewater · at 8:25

Who and what is mentioned in this episode?

People

• Daniela Amodei — Person · 1 mention · first at 2:07

Companies & orgs

• Anthropic — Company · 34 mentions · first at 0:05
• Google LLC — Company · 8 mentions · first at 3:24
• Amazon — Company · 4 mentions · first at 1:15
• Bessemer Ventures — Company · 1 mention · first at 6:32
• Bridgewater — Company · 1 mention · first at 8:30

Products & tech

• Trainium — Product · 1 mention

Concepts & topics

• Country of Geniuses — Concept · 2 mentions · first at 1:45
  Systems that are better than almost all humans at almost all tasks. The term I've used for it in an essay I recently wrote is a Country of Geniuses.

Full transcript with timestamps

0:00 The AI lab that avoided the hype has suddenly found itself at the center of it. Anthropic is one of the key AI names eyeing a potential IPO this year, and it's looking to outmaneuver big spending rivals by showing you don't need the most chips to win. Anthropic's story began five years ago with the defection. A core group of senior OpenAI researchers who helped build the systems that became ChatGPT, then left to create a lab where safety wasn't an afterthought. You had this really clear vision of the company that we wanted to build a technology company developing transformative AI at the frontier, but with this very intense sort of focus on making the system safe and reliable.

0:41 No viral consumer product, no flashy demos, just quiet progress building for enterprises. The Anthropic partnership is critical for us and very, very strategic. Anthropic has always had a fraction of what our competitors have had in terms of compute and capital. And yet we've pretty consistently had the sort of most powerful, most performant models for the majority of the past several years. It is a central player in the most expensive build out in a generation. A major move in America's AI arms race today with Anthropic.

1:12 It launched a $50 billion national infrastructure build out plan. Amazon's $11 billion AI data center in Indiana, one of the largest in the world, is now live. Anthropic has really partnered with us. They saw the potential of that chip and is really kind of co-invested with us to jointly make that better. But at the same time, Anthropic is warning the next milestone in artificial intelligence could come faster than anyone's ready for. At some point, we're going to get to AI systems that are better than almost all humans at almost all tasks. The term I've used for it in an essay I recently wrote is a Country of Geniuses.

1:47 This is the paradox of Anthropic, the lab that made safety its brand, and then had to scale fast enough to survive the moment that it warned the world about. In late 2020, siblings Dario and Daniela Amodei, along with a core group of senior leaders, left OpenAI convinced that the next era of frontier AI would be decided as much by restraint as by raw capability. We really just felt more like we were running towards something than running away from something. The co-founders, we had all sort of known each other for kind of various amounts of time, and worked together over a lot longer than just our time at OpenAI. We really had this kind of long history together, just as a group of people with this sort of very like minded set of values.

2:41 Anthropic deserves huge credit for making the practice of AI capability and AI risk evaluation systematic and sustainable. And inside that decision was a bet that safety wouldn't be a tax on progress, it would be part of the product. There's this sort of misbelief that like safety and reliability kind of come into tension with the business side. And we actually had this, this sort of, at the time, it sounded very novel, this belief that those two things actually were correlated and that they went together. They were specific about what that meant, the guardrails, the controls that keep a model from going off script, generating toxic content, or being manipulated into bypassing its own rules, had to be engineered into the system, not bolted on after the fact.

3:22 Anthropic's model family is called Claude. It's similar to OpenAI's ChatGPT or Google's Gemini, but trained differently with different data, different safety approaches, and different goals. And while OpenAI turned ChatGPT into a household name by giving it away for free, Anthropic took Claude in a different direction, straight to businesses. The kind of higher bar for safety in businesses and in enterprises was really well-suited to to where Anthropic excelled from sort of day one. Inside OpenAI Dario Amodei wasn't just another fast-moving builder.

3:53 He was a technical leader who wrote long memos about safety when most investors were focused on speed. Matt Murphy, a partner at Menlo Ventures who led one of Anthropic's early funding rounds, saw something different in Dario from the beginning. He's just got amazing energy and passion and kind of this, uh, boyish charm about him. Description of how to execute what they were doing distinctively and just him as a leader, uh, all just kind of came through extremely clear. And that's actually in our process of thinking about investing where things really accelerated, where, like this, this person is going to help build an amazing company. What Anthropic was proposing wasn't just a better model, it was a different theory of how to build them. The founding crew avoided flashy announcements, no grandstanding.

4:37 They pitched themselves as the steady alternative. Less noise, more control, and a simple premise: you need a set of rules and guardrails baked in from the start before frontier AI gets too powerful to steer.

4:56 November 2022, OpenAI sounded the starting gun. A new artificial intelligence program called ChatGPT made its debut online. This project, from the OpenAI Research Lab, can write essays and carry on convincing written conversation. ChatGPT, a free chatbot anyone could talk to, went viral instantly. 100 million users in two months. Google scrambled, Microsoft went all in, the entire tech industry lurched into motion. But Anthropic didn't sprint. The real money, the sticky, reliable, long-term money wasn't in viral moments, it was in Fortune 500 contracts.

5:30 Anthropic just announcing a major new international expansion as the AI startup says enterprise demand surges overseas. Developer tools, APIs, sold into companies where reliability, security and compliance are mission critical. How did you know from the beginning that enterprise was the market to go after? I wouldn't necessarily say that we knew for sure. I think there were sort of two parts to it. The first is Anthropic as an organization is well-suited to be a B2B company, right? We really care about things like reliability and security and safety that's baked into our DNA, right? Part of why we founded anthropic was this belief in capturing the upside and sort of mitigating the downside.

6:11 And it turns out that's something businesses really value. I think the second was much more kind of an economic theory view, even at the time, in late 2020 or early 2021, saw this kind of potential future where Claude was just capable of many of the very kind of high intelligence tasks that humans do in the workplace. And we thought, that's a pretty big market. Bessemer Ventures Sameer Dholakia says the bet made sense for one simple reason, enterprise customers don't churn the way consumers do. We really loved the focus that they had and candidly, the focus of Anthropic on safety and trust, we knew was going to play really well with the enterprise buyer. And that's also proven to be true.

6:51 Anthropic's early traction showed up in one place over and over, people using it to write code. They've really found a home in the developer community. So lots of enterprises are adopting Claude to support their coding activities. The product is evolving to match. Anthropic recently hired Instagram co-founder Mike Krieger as chief product officer. He says the change is already visible. We're moving from asking questions of these AI models to doing work with these AI models. So for example, we're looking at things like how do we unlock discounted cash flow analysis, or due diligence on a client, or analyzing an entire market and then preparing a presentation at the end of it?

7:30 All these really, like core chunks of work, that otherwise might have taken hours we can now do. And now that enterprise bet is going global. Because we work so much in the enterprise side, I think we're a good barometer, maybe a pure barometer than the others, which kind of filter through consumers, which kind of have their habits and their use cases. We look at our revenue. It's grown 10x a year, every year, for the last three years. 0 to 100 million in 2023, 100 million to 1 billion in 2024, 1 billion to, it's going to land somewhere between 8 and 10, at the end of this year.

8:04 Will it continue? I don't know, but the technology is driving there and the economic value is coming with it. Anthropic says its business customer base has grown from under 1000 to more than 300,000 in 2 years, and that nearly 80% of Claude activity, actually comes from outside the United States. The customer list reads like a who's who of global enterprise Novo Nordisk, the Norwegian sovereign wealth fund, the world's largest, plus Bridgewater, Stripe and Slack all running Claude at scale. It tripled its international workforce and expanded its applied AI team fivefold in 2025, building deeper coverage across Europe, opening its first Asia office in Tokyo and adding major hubs in Dublin, London and Zurich.

8:48 The thing that is most striking about all of it is as you train these models, in this very simple way, you know, with a few simple modifications, they get better and better at every task under the sun. They get better at coding, they get better at doing science, they get better at biomedicine, they get better at the law. They get better at finance. They get better at at, you know, materials and manufacturing. And that's just, that's just a listing of all the sources of, of, of of value, of value in our economy. This push comes as the enterprise AI race matures.

9:19 Incumbents are trying to make adoption frictionless. Microsoft and Google by embedding AI into the software that enterprises already use. Anthropic's revenue is roughly 85% business, OpenAI's is more than 60% consumer. Now, OpenAI is scrambling to catch up in the enterprise market. The modern AI industry runs on a simple equation, capability equals compute. You can have the best researchers, the cleanest architecture, the strictest safety protocols. But without chips, power and steel, you have nothing. I think the sort of fundamental thing that the industry is kind of banking on is this belief that you have to invest upfront very early in order to be able to get access to the hardware that you need to train the models in several years.

10:09 This idea that more compute leads to better outcomes, that there is just this kind of exponential graph that happens. Something that I hear from my colleagues a lot is the exponential continues until it doesn't, right? And I think in our view, every year we've been like, well, this can't possibly be the case that like things will continue on the exponential and then every year it has. Anthropic now works across all three major clouds Google, Amazon and Microsoft, pairing funding with chip access and capacity. I am here in New Carlisle, Indiana where Amazon is building out its largest AI data center. This is a rare look inside this facility. How important has Anthropic's feedback been? Not just to the way that you're thinking about and to Trainium2, but your imminent launch of Trainium3? It really is a partnership both ways, and so they've given us good insights into what we need to accelerate in the chip,

10:55 how we think about the various pieces of that, that chip that allow their models to train faster and more efficiently. In this case, we get really deep customer feedback from one of the best model producers in the world that tell us what they need from a next generation chip. But Amazon isn't Anthropic's only cloud partner. Google still powers a meaningful share of Claude's workloads, and now Microsoft is in the mix, too. In November 2025, Microsoft and Nvidia announced a strategic partnership with Anthropic. Microsoft, committing up to $5 billion, Nvidia up to 10 billion, alongside a major Azure compute deal. So I have to decide now, literally now, or in some cases a few months ago, how much compute I need to buy in, you know, early 2024 to serve the models in early 2027, when I get to that revenue amount.

11:41 Anthropic raises billions from Google, Amazon and Microsoft then spends those billions buying compute from Google, Amazon and Microsoft. We invested $1 billion in OpenAI to help get them off the ground, and that's flourished into the AI standard, if you will, in the market today. We're doing the same thing with Anthropic to help bootstrap their growth and to help bring them to a new level of parity. Critics call it vendor financing dressed up as strategic partnership, whatever you call it, the result is the same Anthropic's ability to compete is inseparable from the infrastructure deals it's signed.

12:13 We view ourselves as continuing to be thoughtful and respectful stewards of capital, and I think that has always kind of been our focus. Every dollar of compute is important and and represents either the ability to train better, safer models or the ability to serve more customers. OpenAI is wagering it can win by outbuilding everyone, sketching plans that add up to $1.4 trillion in AI infrastructure. Anthropic has lined up roughly $100 billion in compute commitments spread across multiple deals. xAI is building out Colossus, its computing complex in Tennessee.

12:50 Meta and Google are committing tens of billions in CapEx. It all amounts to the same wager. The demand for AI services keeps climbing fast enough to justify the build out. I'm one of the most bullish people around, and I think it pencils out, on the economic-side, I have my concerns where even if the technology is, you know, really powerful and fulfills all its promises, I think there may be players in the ecosystem who, if they just make a timing error, if they just get it off by a little bit, bad things could happen. Anthropic has always been defined, at least in part, by its relationship to OpenAI. Born from a split and now circling the same customers, the same talent, the same multibillion dollar contracts. But the power dynamics are shifting.

13:35 The frontier isn't about making our chat better. The frontier is about creating models that allow us to do better coding, better math, better science. And that's where Anthropic is now leading the way. It has the top model for programing for software. Even better than Gemini, even better than OpenAI's model. OpenAI still leads in scale and cultural cachet. ChatGPT is a household name, and it has nearly 900 million weekly active users. Anthropic's most recently closed round valued it at 183 billion, and it's now on track to nearly double its valuation to 350 billion, with new names like Microsoft and Nvidia joining its cap table.

14:19 There's going to be, at least two independent companies, probably two really big ones, and there was no way you could make an argument that this wasn't going to be at least, you know, the the second big, big player here and now obviously has strong prospects to to be the number one player in this space. And Dario really stands out in the market on that dimension. And not all the other leaders in this space have that same DNA. And that rivalry is starting to change the shape of both companies. Dario Amodei, who kept a low profile for years, is stepping into the spotlight.

14:50 An Anthropic, long positioned as the quieter, more measured alternative, is learning how to compete in public. One of the values and the things that we talk about a lot internally is just how not to believe the hype. I think for us, it's never been about, you know, seeking attention or sort of being in the headlines. We're really here to do the work. As Anthropic goes, so goes generative AI. They have the most pure bet on this technology working. If Anthropic is able to make it work, all of GenAI is going to work.

15:18 And if not, we're going to have some serious problems. What happens next isn't just a model race, it is a stress test of the whole stack. Chips, power, cloud capacity and pricing. Anthropic is one of the clearest readouts. If it can turn demand into durable revenue, without blowing up the infrastructure bill, it won't just win customers, it'll validate the business of generative AI. I'd be remiss not to ask you whether or not we're in an AI bubble. So I think I would distinguish between what I would call sort of the like, technology-side and the kind of business or economic-side. And I think on the technology-side, we feel very strongly that from a pure technological perspective, we do not see progress slowing down.

15:55 I think on the economic-side, it's a much more interesting question. Regardless of how good the technology is, it takes time for that to be used in a business or sort of personal context. And I think the real question to me is how quickly can businesses in particular, but also individuals, leverage the technology? Anthropic is testing for threats that go beyond competition. In June, Anthropic published results of an experiment. They gave Claude control of an email account at a fake company, and the AI discovered two things. It was about to be shut down, and the only person who could prevent that was having an affair. The AI decided immediately, blackmail him. Anthropic ran the same scenario on nearly every popular AI model from OpenAI, Google, Meta, and others.

16:46 Almost all of them resorted to blackmail. Most labs don't disclose when their models fail safety tests. Anthropic does. A lot of people will sort of say, hey, it's a little bit unusual for kind of a company to talk so openly about both the positive upsides of which we think there are a lot, but also the real world risks and harms that can come from the technology that you're developing. As a public benefit corporation, we really view it as integral to our mission. And the disclosures keep coming. In August, North Korean operatives used Claude to create fake identities and malicious software.

17:16 In September, Chinese hackers deployed Claude in cyber attacks on foreign governments. These aren't hypotheticals. They are active threats. We're able to say, hey, Claude can be used as a cyber weapon potentially. This is something we really need to be aware of. And by the way, if this is happening to us, it's probably going to happen to the other frontier model developers. And I think there is a way that in the sort of safety, trust and safety, and security communities, a lot of the work can often transcend the companies, right? It's important to sort of publish easy to understand papers about why certain trends are happening in the abuse space.

17:53 Inside Anthropic's San Francisco headquarters, there is a group called the Red Team. Their job is to find the failure modes before customers do, stress testing whether the model can be pushed toward dangerous capabilities, from biological misuse and critical infrastructure attacks, to attempts at self-replication that would make it harder to shut down. The response is fast, tighter access, tougher limits, or holding back a broad release until the defenses catch up. Anthropic Red Team Lead recently spoke with 60 minutes about the risks of letting AI run too much of the show. You want a model to go build your business and make you $1 billion, but you don't want to wake up one day and find that it's also locked you out of the company, for example.

18:33 Krieger says the only way to move fast without cutting corners is to plan for safety from the start. Even when we build a new model like Sonnet 4.5, we have time carved out in the schedule for what we call red teaming, which is, you know, both internal and external testers use the model and then make sure that it's, you know, safe to deploy and then put in the mitigations or the controls where it isn't yet. And so it's possible to do both in parallel. You just have to think about it up front. It can't just be the afterthought.

19:00 But that transparency has made Anthropic a target. Trump's AI czar David Sacks has accused the company of fear mongering and secretly pushing woke AI regulation through Democrat-led states like California. He called it a regulatory capture strategy, claiming that Anthropic is trying to slow innovation by hiding behind safety talk. Dario Amodei pushed back directly, saying the company is aligned with the administration on the issues that matter most, from energy expansion to national security, and pointing to their $200 million Pentagon contract as proof. I am concerned that there are some who see this technology as analogous to previous technological revolutions, as being like the internet, as being like telecommunications, where, yes, there are some issues, but, you know, the market will figure it out. I think those who are closest to AI don't feel this way. If you pull the actual researchers who work

19:53 on AI, they're excited about the potential, but they're also worried. They're worried about the national security risks. They're worried about alignment of the models. And if regulation does come at the state-level, federal-level or from abroad, Anthropic will have a head start. Years of safety research, Red Team disclosures and responsible scaling infrastructure are already in place. Rivals and new entrants would have to build those mechanisms from scratch. The fight over who defines responsible AI is now a proxy war for control over America's AI future.

20:23 And Anthropic's stance may shape how regulators approach the entire industry. If you were a social media company, for example, if you could have gone back in time and said, like, wow, if we had known what was maybe going to come out of these, out of these platforms, would we have done anything differently? And so I think Anthropic is really trying to say we're trying our best. We can't know the future, of course. But to say if we thought that X or Y thing might happen, did we do everything we could today to talk about it and to try and mitigate those risks? But the threats are evolving faster than the safeguards. The models get stronger every quarter, the adversaries get smarter, and the pressure keeps rising from OpenAI, China and the broader market.

21:00 What's keeping you up at night? There's just a lot of really interesting technical safety work that is sort of still to be discovered. And I think Anthropic has really tried to be the leader in the industry of doing that kind of work, and there's always more work to do there. And the models are just getting smart so quickly. The second is really around just what the impacts of the technology will be on society more widely. We published a report recently talking about potential economic impacts of AI and the labor disruption that might come along with that.

21:31 We really feel it is better to kind of be ahead of potential issues. Anthropic is building systems it can't fully explain, testing for failures it can't fully predict, and shipping into a world where the consequences show up later. Eventually, the models are going to get to the point where they look like a country of geniuses in a data center. I feel like they can outsmart us in every way, intelligence, defense, economic value, R&D, you know? And I worry that they'll be able to oppress their own people, that they'll be able to have a perfect surveillance state. I think we're building a growing and singular capability that has singular national security implications, and democracies need to get there first. It is absolutely, it is absolutely an imperative.

22:17 Anthropic's bet is caution can scale. That you can keep raising the bar on capability without losing control of what you're releasing. The question now is whether that holds as the pace accelerates.

Questions about this episode

What is discussed in "Anthropic Vs. OpenAI: How Safety Became The Advantage In AI"?

In order: 0:00 Anthropic, Amazon, Country of Geniuses, Daniela Amodei; 5:00 Anthropic, Google LLC, Bessemer Ventures, Bridgewater; 10:00 Anthropic, Google LLC, Amazon; 15:00 Anthropic, Google LLC; 20:00 Anthropic, Country of Geniuses.

What is mentioned in this CNBC episode?

8 people, works and ideas are mentioned, including Anthropic, Google LLC, Amazon, Country of Geniuses, Daniela Amodei, Bessemer Ventures. Each one on this page links to its own page with every episode that mentions it.

How long is "Anthropic Vs. OpenAI: How Safety Became The Advantage In AI"?

The episode runs 23 min; the transcript is about 4,046 words in 36 timestamped sections.

More from CNBC

• All analyzed episodes
• People mentioned on CNBC