That was quick. I'm sorry. That was quick. Yeah, it's great to be here with everyone. I'm Ravi Matre, founder and one of the lead partners at Lightseed Venture Partners. We recently led the largest investor in a $3.5 billion financing for Anthropic. Today we're fortunate to be joined by two people who have unique and interesting perspectives on the frontier of AI. Jack Clark is a co-founder of Anthropic and was formerly policy head at OpenAI, and is one of the world's most respected thinkers in AI governance and safety. We're really privileged to have Senator Mike Rounds of South Dakota as well. He's been a leading voice in defense and intelligence policy, a member of the Senate Armed Services Committee and several select intelligence committees. So just want to set some brief context about the current moment in AI technology. It's clear to just about everyone at this point that we are in the early innings of very major change. We've already seen AI solving incredible problems as it relates to the design of new innovative drugs and materials. We've seen the beginnings of very significant change and transformation across a variety of business services. One of the most obvious ones if you live on the West Coast is we've really arrived at the point where self-driving autonomous vehicles, we call it in San Francisco having the Waymo moment when you walk around and see that a third of cars on the streets don't have a driver. So it's very clear that AI is already at a point where major change is happening. But it is also very early innings. How quickly will AI improve from here? If it's rapid, should we be thinking about export controls as a way to manage how the technology proliferates? Or if it's going to happen more gradually, should we be erring on the side of how the US is strongly pushing AI into the world in a way that it can set the standards? That's really the key thing we're going to explore today. I'd like to start, Senator Rounds, with you at the highest level. How is the US in your mind, and how is the government thinking about planning to have the US stay competitive against a backdrop of more technology competition than AI than there really has been before?
I think perhaps the best example I can give of the way that we win in this challenging world right now with regard to AI is what you're seeing right here right now. When you talk about from the valley to the hill, you're talking about bringing in private enterprise. You're talking about the individual initiative, the need for the private sector to recognize that they need to be involved in activities on Capitol Hill in terms of helping us to actually develop the right path forward and providing your expertise and your knowledge as well. None of us here are experts on AI. What we know is that it is a race that does not have an ending to it. But we've got some really tough challengers out there. The reason why we look at this as a race is because for the folks that are ahead in AI, they develop the standards. They develop the path forward in terms of what the uses are. We are in a confrontation right now with other world leading countries. They recognize how big of a deal, how big of a tool AI is. They're going to do everything they can to expedite it. What they don't have is the private sector that we have. And when we build this together, it means then that we can stay ahead of what they can do. China right now, they can do a lot of stuff and they're very innovative, but they don't have the private sector that we've got. They're going to steal our chips. They're going to steal some of our intellectual properties and so forth, they're going to create some on their own. But if we can stay ahead of them in terms of the production of the best, if we can keep the brightest minds here, if we can keep individuals that really are the innovators to recognize that this is where their future is, we win this game. So number one, it is a matter of being inviting to other people that can actually create the future with us. Number two, it is restricting what goes to China during this time in which we try to establish dominance so that we can lay out appropriate standards for them. It means then also that we've got to address some problems that I think a lot of you are aware of. Number one, we do about 1300 gigawatts of power per year is basically where our production capability is. If we don't substantially increase the amount of power that we have, then we don't have the ability to create the data centers, operate them here. It means then that people start looking elsewhere where they can actually afford to do it. Put the data centers in and create those centers. So number one, we got to have power. Second of all though, we need the collaboration that is identified right here. The collaboration is critical.
And just to follow up, and Jack, I want to get your thoughts on this, but you mentioned the need for innovation and the need to move quickly, but on the other hand, you've been very vocal, I think, in a good way about the US's need to maintain leadership in critical technologies. So how does Congress or in your views, as a leading voice in government, focus on this balance the need for speed and innovation and driving the US hopefully being the standard setter with the need obviously for some types of controls to prevent adversaries or key competitors from obtaining the technology which could allow them to be leaders. How do you, what is the balance? What is the thinking?
Part of that's got to be delivered by the folks that are in this room as an example. It is a matter of we can either push them to develop their own or we can keep the most innovative, the brightest, the top end technologies here and allow for the sale worldwide of the medium-sized packages that are out there. The equipment which is not top end but that will still provide other people with a reason to buy from us rather than to create their own. But we've got to be in that position to actually create the best and the brightest and to stay that one notch ahead all the time. It is a very fine line and it's one that you can't just set and put into a law. This is one that you literally have to regulate and you've got to do it with industry participating in those discussions on an ongoing basis, not a one-time set. So this is about government really needing to have a way to collaborate with private sector to keep its finger on the pulse and make adjustments as needed. Absolutely. But it's more than just saying, okay, we want the biggest to come on in. This is a case of where you have to have a cross-section of the innovators, the young, the new ones on the block along with those that are the tried and true that are still developing. You need a cross-section on it. You also got to talk to the folks that have proprietary data centers and those that want the open data centers, because they're both going to be a part of it. The most dangerous stuff is going to be in those closed systems, but the innovation and the major, the fastest growing stuff is going to be in those open sets. So you have to have them both. You've got to talk about them both and you've got to have people from both sets involved in the discussion.
And Jack, from the perspective of someone who's on the West Coast in Silicon Valley and working with one of the major AI lab companies, what is your perspective? Has the US found the right balance from the objectives of where Anthropic would need to go between enabling innovation and on the other hand ensuring that AI labs are secure and that potentially technology doesn't proliferate in the wrong way?
The good news is we're a lot further than we were a few years ago. I've been coming to Washington in one form or another since 2016 when I used to talk about neural nets back then it was to a relatively small room and people were not very interested but things have changed. The US companies are really good at going really fast and where we need help is ensuring that the US industry doesn't have accidents or misuses of its technology because an accident or a misuse of our technology would lead to a regulatory blowback that would probably do to this industry what happened to the nuclear power industry. So where the US has been smartly developing things as an infrastructure that lets US companies like Anthropic or the other frontier labs move really quickly but with confidence that we're not going to create misuses of our technology that could enable geopolitical rivals. A few years ago, there was no ability for me to go to the US government and have experts on bioweapons or cyber warfare work with me to calibrate my systems. So I wasn't accidentally putting those on the internet and selling them to everyone rather than a subset of people that may be able to use them for good in America. So we've got significantly further than we were a few years ago. The most important thing now is recognizing that if you believe what the frontier companies are saying and what we truly believe is that we're in this era where over the next few years we will build the largest data centers that have ever been built on the planet to train genuinely transformative systems that will have huge economic and national security relevance. We have to ensure that we don't give the frontier infrastructure to train these things away to our rivals. I think that the work that we're currently seeing on export controls is crucial to this because otherwise we will make an unforced error of proliferating the ingredients for frontier dual use technology into rival countries that we're trapped in a geopolitical competition with. So we're doing okay and there's more to do.
It sounds like there's alignment on this ongoing need for an iterative partnership between the private sector and public sector. Yeah. I want to turn specifically to, because we don't have that much time now, to I don't know maybe it's a spicy topic but to China specifically DeepSeek and where are we relative to this view that we're in an AI arms race. First question, Jack, I'm going to ask you specifically for your perspective on DeepSeek. They clearly, as a foreign adversary in this sense in an AI competition, have taken a different approach to how they develop AI infrastructure from the labs in the US. How seriously should we be thinking about the impact that they can have? And do you think DeepSeek is a technology that could be dual use for China?
I think the DeepSeek hype is perhaps a bit overblown. I describe DeepSeek as a competent group of engineers and researchers that have read all of the same research papers all the frontier labs have made and have clean sheeted a new system off of that which has some clever algorithmic ideas. But if you actually analyze it, it looks 6 to 8 months behind where US frontier companies are. Additionally, the DeepSeek founder said the main thing limiting his company is compute. So if they had access to arbitrarily large amounts of compute, they might be a closer competitor. Okay. I just to add one layer of detail, we ran the same national security tests on DeepSeek technology that we run on our own and it was significantly behind our own. Neither our own technology or DeepSeek's technology are yet what you might call concerning to national security people. It's more like they're symptoms of progress that we should track closely.
Okay. Senator, I want to ask you specifically how does Congress today assess the risk that China could have an AI breakthrough, some type of a step function change in their abilities? And if there was a breakthrough, what is the US assessing its response could be?
I got to be careful. Look, they're very good. They're good because anything that they do in any part of their private industry is available to government. And so DeepSeek as an example, if we talk about it having dual uses, the answer to that is yes. By definition, the government in China can access it, can use it. The question is, are they progressing at a faster rate than we are? I don't think so right now. But they have the capability of continuing it and they have a mindset. Xi Jinping is not a stupid guy and he recognizes that this is an area that he cannot fall behind in any more than what he already is. He's going to do everything he can to make sure that China has access to this because they see this as the key long term. You have five domains: air, land, sea, space, and cyberspace. He wants to make sure that in those critical areas of space and cyberspace that he is our equal, and so he's going to do everything he can to utilize all of the resources that develop in China that will be made available to his military sources. There's no question about that.
Okay. And not to put you on the spot, but as the follow-up question, I mean, if China did have some type of an existential AI breakthrough, what is the US's response at this point?
Well, they're not there right now. And the whole key here is not to let that happen. Whoever controls the best, the fastest, the most capable along with the trained data sets, they will have the weapon systems that can either protect against or be offensively utilized in the future. You got to remember when we talk about AI, we're talking about an acceleration in a decision-making process that transcends what our traditional conventional military systems in all aspects, air, land, sea, and space. AI is the key today. We don't have an option. We have to win this race. There's no question. And government can't do it alone. We have to have the innovation that comes with the private sector. We don't have a choice.
And Jack, I want to ask you a question now about where Dario, Anthropic the company's stated position has been has really advocated for the importance of export controls around silicon technology because of post-training time as a way that models are going to really advance in capabilities. But what about policy responses to open source generally and premier US AI research labs that are making open-source models available? Is that another source of vulnerability in your mind? If so, what from a policy standpoint should we be thinking about?
