美国 AI 战略内幕：基础设施、监管与全球竞争

Great to see everyone, and I'm thrilled to be able to talk about the issue of the day, which is artificial intelligence and AI in our world. David, Michael, I'd love you to talk about where we are right now in terms of the pursuit to be the number-one leading AI country. How are we doing, David?

I think we're doing great. Maria, last year, President Trump gave a major AI policy speech in July, and he declared that the United States had to win the AI race. He had first of all declared that we were in one, and I think his speech was reminiscent of when President Kennedy declared that we were in a space race and had to win that race.

I think since then, what you've seen is that American companies have only innovated more. You're seeing all sorts of really incredible products being released all the time. I think that American AI models, chips, and data centers only keep getting better and better. So I feel very good about the American position in this AI race.

Certainly, we have some very competent and formidable competitors. China obviously has a lot of very smart people working in this area. But I do think that just what you see from American companies in Silicon Valley right now is really incredible.

And yet there are still so many questions about all of the spending underway to build this out with regard to data centers. Of course, the question keeps coming up: Are we spending too much? Will we get the return on investment? How do you see that?

I think that we will. I think the reason why you're seeing this huge infrastructure buildout is because the demand is ultimately there. I know a lot of people worry about whether this could be like a dot-com situation. Remember when we had the whole fiber buildout in the late '90s and then we had a dot-com crash?

The difference here is that in the late '90s and early 2000s, we had a problem known as dark fiber, where you had this fiber buildout and then it didn't get used. There's no such thing as a dark GPU right now. Every GPU that's being put in a data center is getting used, and it's being used to generate tokens to power this new generation of AI chatbots or coding assistants.

There have just been some releases in the last couple of months on the coding front that, if you're following what software developers are saying, they're saying it's mind-blowing. It's completely revolutionizing their industry. So demand for tokens just increases, and that increases the demand for the data center buildout that we're seeing.

I don't think it's going to stop anytime soon. Just last year, this infrastructure buildout added about 2% to the GDP growth rate, and that's what helped propel us to this 4% to 5% growth rate. I think you're going to see something similar this year.

Well, it is certainly leading growth. Michael, I'm so happy to be able to get this conversation going with both of you, who are really leading this. David, thank you. Michael, thank you. Same questions for you, Michael. Assess where we are right now on AI.

I think just a reminder for the group, for those who haven't been tracking as closely as we do every day: The plan really had essentially 3 pillars. It talked about, first, how the US can continue to out-innovate our competitors; second, how we can drive the infrastructure build that we need to support this AI revolution; and third, how we actually share with the world, or export, our great American technology.

For each of those 3 pillars, there have been quite a lot of actions that the federal government has taken to drive that forward. I think we're pretty proud to say that we've made pretty good progress on all 3.

Just focusing a little bit on the innovation one, which you were talking about earlier, I think the core insight that we've always had about how you drive this innovation is that you have to have a regulatory environment that allows this technology to be developed and ultimately commercialized in the United States. The US has done a great job compared to the rest of the world on setting that up and creating a framework that works, but we can always do better and improve it.

The president, in his speech in July, talked a lot about this issue of a patchwork of state regulations and how we can ensure that there aren't 50 different rules around AI. What's most important about this debate, which I think a lot of people sometimes miss, is that the patchwork is actually most detrimental to early-stage young companies and entrepreneurs.

If you want to develop a new AI technology, if you want to build something on top of one of our great frontier models, having to figure out how to navigate 50 different rules across 50 different states creates a lot of friction. Ultimately, the big guys are the ones that can succeed in that environment the best. So we're spending a lot of time trying to think about how we can create a legislative proposal that can deliver on a sensible national framework to solve this regulatory issue.

So what would you say, then, Michael, are the basic frameworks that are must-haves in that kind of federal oversight? Some states did push back in the US and say, "No, no, we want to be able to control our destiny when it comes to AI." What's most important when you look at that framework in terms of federal oversight?

I think in the executive order the president signed in December, directing us to work through this proposal, he listed a few things that the states should continue to be able to pursue individually on their own. Legislation or rules around child safety was on that list. The rules around permitting of data centers and buildouts are continuing to be something that states should look at.

There are a few things that were enumerated, but that's the kind of stuff that David and I are going to be working through. I don't know if you have any thoughts on that.

Yeah, I think the basic problem that we have is that, frankly, the states are going hog-wild right now with regulation. There are over 200 bills going through state legislatures right now. I think it's very much a knee-jerk reaction. I know there are a lot of fears and concerns about AI, but it seems like for every hypothetical concern, there are multiple state bills now to try and regulate that thing before we really know how it's going to play out.

I think it would be better to spend a little bit more time studying how AI is actually being used and what risks are actually materializing before you overregulate the thing. But in any event, that's what we're seeing right now at the state level.

I think the president's been very consistent that it would be better to have one rulebook, a single rulebook, at the federal level—a lightweight federal standard. I think this problem is only going to get more acute over time because, again, as you have 50 different states running in 50 different directions, the patchwork problem only gets more significant.

In any event, this is something that we're going to work closely together on this year: to see if we can get enough consensus on a federal framework to enact a law. Only Congress can ultimately preempt the states. We understand that. As you know, it's very difficult to get a bill through Congress. You need 60 votes in the Senate, so it has to be bipartisan to a certain degree. But we're going to try and see if we can work to get that consensus.

Yeah. And do you have any clarity on the timing of that, in terms of support in Congress for federal oversight? Or do you see pushback there as well, depending on the state you're talking about?

Well, there's pushback in Congress to the idea of preemption without a federal standard. In other words, you can't replace something with nothing. This is the thing that we heard repeatedly.

But I think there is quite a bit of interest in both the House and the Senate in having, again, some sort of lightweight federal standard. We're still in the early stages of those conversations, and we're going to see what we can try and get done this year.

Meanwhile, you've got some people pushing back after wanting to see the innovation and growth of data centers. Now they're saying, "Not in my backyard." What about that? Is that an issue?

Yeah, we got a letter recently from Bernie Sanders saying, "Stop all data centers, all data center development." If we do that, we will lose the AI race. You do need this infrastructure. Other countries are building out this infrastructure. China's building out—I think they're spinning up a new nuclear power plant or coal plant, new energy, every single week, and a lot of that is going to power their data centers.

It would fundamentally, I think, handicap the United States in the AI race if we just stopped building data centers altogether. At the same time, there are concerns about affordability, about whether consumers would have to pay a higher electrical rate because of data centers.

President Trump's been really clear that consumers should not have to pay higher rates for electricity because of data centers. You saw just last week Microsoft stepped up and made a pledge that its data centers will not cause residential rates to increase. I think you'll likely see other tech companies stepping up and making similar commitments.

In fact, when I've talked to the hyperscalers and when I've talked to the AI companies, it was never their plan to draw off the grid.

They all are also standing up their own power generation as part of their build-out. What Secretary Wright, our Secretary of Energy, has been doing is trying to reform the regulations that make it more difficult for these AI data centers to stand up their own power behind the meter.

Basically, our vision is—and I should say this is President Trump's vision, really since the beginning of the administration—let the AI companies become power companies. Let them stand up their own power generation as they build, side by side, with these new data centers. The result of that is, A, we get this infrastructure; B, residential rates don't go up.

Yeah, because, Michael, this race has fast become—it’s moved from an AI race to a power race.

I think what we're seeing is that we need to share a good story about how, ultimately, this build-out is going to be net positive for American ratepayers. If you're in a small community and someone shows up to build a data center, you have to make it clear that, ultimately, this is something that's going to lower your rates long term.

The president put out a Truth last Monday where he was, as David said, very clear that if you're going to build a data center, you have to pay your own way for it. Microsoft has stepped up, and our hope is that many others will do the same.

But some companies, because they don't have the cash right now, are borrowing money to build out the data centers. There's also a worry that the banks will be left holding the bag for some of this because, again, the spending is too much. Your thoughts on that?

Well, I think there is obviously that concern. You see Oracle making a huge investment. You see Blackstone making huge investments, along with real estate companies. Ultimately, I think these are very savvy market players, very deep companies, and they're doing this because they see an ROI at the end of the rainbow.

Can I make one other point about the data center, just on electricity? I actually think that if we allow the data centers to stand up their own power generation, it will bring down rates. Not only will it not increase residential rates, it will bring them down.

It will do that in 2 ways. One is that the data centers can give or sell power back to the meter when they have excess, so that will help bring down rates. Second, there's a lot of fixed costs involved in power generation. It's not all variable. When you're able to amortize those fixed costs over a greater supply, you bring down the meter rate for everybody.

There are huge economies of scale. The more scale you get in electricity, like most other things, the price comes down. It's actually a good thing that we have this build-out going on because it will ultimately reduce prices for consumers. But we do have to make sure that these new data centers aren't just plugging into the grid and using it; they have to be contributing back.

I think a great policy change has been made under this administration. The Biden administration had, as a matter of policy, made it such that you couldn't do this behind-the-meter energy generation. If you wanted to bring your own power, you couldn't. You had to be part of the larger grid.

I think that rule has been changed by Secretary Wright and by FERC to allow this to happen. Ultimately, I agree with David. Once you have greater scale in the power generation, you'll be contributing back into the grid in a way that benefits ratepayers.

Let's go back to the uses and how AI is changing our lives. You mentioned earlier all of the uses and the impact AI is having. What do you see as the most important use, and where is AI being deployed and implemented best right now?

Well, it's interesting. I think there's been an evolution. We started with AI chatbots like ChatGPT, and in a sense, that was kind of like better web search. It was really great for research—asking it questions and getting answers to anything.

Then we saw models add chain-of-thought, and they could start to do deeper reasoning. Then we saw coding assistance. I think over the past few months, there's been a real breakthrough. If you talk to software developers, it really seems like there's been a major shift and improvement in the quality of coding assistants.

I think where that's going next is tools for knowledge workers. The same types of assistants that have been outputting code can now output any type of format. Whether it's Excel models, PowerPoints, websites—you name it—knowledge workers are now going to be able to generate all these different types of things the same way that software developers have been using AI to generate code.

I think that's one of the big things you're going to see in 2026: just this productivity boom for knowledge workers. That's one of the things you're seeing on the ground.

Separately, there's a bunch of things happening in industry verticals. Different industries are being impacted by AI. In health care, I think there's a tremendous opportunity to reduce administrative bureaucracy and improve the processing of paperwork. There's also the use of AI in medical and scientific research to help find new cures.

You're already seeing users tell all sorts of stories about diagnosis. They've been able to put their medical records into ChatGPT or another chat engine or chatbot and get remarkable results. They've been able to finally figure out what was wrong with them and take that to a doctor. You have doctors using it, too.

Medical care is a really interesting area, but there are a whole bunch of examples of different industries that are now being impacted.

The one area I think a lot about is AI for science. Going back to David's initial point about the progress we've seen in these frontier models, the very early ones started with just general knowledge. You have to go back and understand why. The question was: What was the data available for those model builders to start training their models?

For the early models, you could just scrape the internet, cram everything into a model, and train it. That's where you had this first phase of large language models. The second one was coding. If you think about how you get a really good coding model, you have to train it on existing code, and that's something that's relatively easier to acquire than other types of data. You saw great progress and jumps in the coding models.

I think the third big shift that hasn't really been touched on yet, which the government itself is trying to push, is AI for science. The reason it's so challenging for scientific discovery to tie in with the way that LLMs are traditionally trained is that science data is extraordinarily fragmented. It isn't done or formatted in a way that can easily be applied to a large language model training run.

Scientific discovery is spread out across so many different disciplines. You have chemistry data, math data, materials science data, and all of that is in different formats.

Our effort in the administration—we launched something called the Genesis Mission—is our attempt to make these big, bold leaps in AI for scientific discovery. Our national labs at the Department of Energy have been doing incredible research over the last 50 or 60 years, and all of that is sitting there, ready to be used to train these models.

My hope is that over the next year, we're going to see a lot more work in scientific discovery to accelerate how quickly we can choose which experiments to run, run those experiments, go back and figure out what we did wrong, and run them again.

This ties in with lots of interesting ideas that people have around some of these AI labs, where you can essentially put in the thesis or the hypothesis, and ultimately these labs can do lab experiments themselves and move forward. That's the dream that I have: that, ultimately, we as a country can almost double our R&D output over the next 10 years because of AI.

So what kind of breakthroughs would you expect or would you like to see?

Yeah, I think the ones that can make a big impact are, first, the experimentation and training runs around fusion. They are extraordinarily computationally heavy. If we can have a faster feedback loop on how we do these simulations for fusion, we can move the timelines in for fusion. That could be a big step.

Materials science is also a very big area where you want to be able to test all types of different molecules and how they interact with each other. This is important for all the big things we're trying to do in space, whether it's our lunar base, getting to Mars, or bringing nuclear energy to space. Having advanced materials science is important.

The third is one that everyone always cares about: health care and therapeutics. How can you more quickly identify the best molecules to solve a particular health challenge, and how do you more quickly iterate to a point where you can move to a clinical trial?

And on an everyday level, you also have the auto sector. I think it's a big beneficiary here. That's one area that seems to be spending a lot on this as well. Do you agree with that?

Well, I mean, with self-driving, or—

I mean, self-driving for sure is going to be huge. It feels like we've hit some sort of new inflection point there, where the quality has gotten to the point where you're starting to see robotaxis now—Waymo and Tesla.

What about an AI assistant? Is that going to be something that's commonplace? Someone said to me the other day that, in China, they're doing things so much differently because they're using AI for research, as you said, but we're using it as—I have my AI assistant, and they're paying my bills, cleaning my house, buying my wife a birthday present, and doing everything for me.

I think so. I think that'll happen probably this year. The product that just came out recently that everyone's going crazy over is the latest iteration of Claude Code, which is powered by Anthropic's Claude Opus 4.5 model, which seems to be a real breakthrough in coding. The software developers are really impressed with it, but inside of Claude Code they introduced a new tab called Cowork. As a non-coder, or as someone who's looking to create output other than code, you can now use it to basically create all sorts of other kinds of outputs. Like I mentioned, you can do spreadsheets or PowerPoints, things like that.

You can point it to your file drive, and it can look at the work you've already done. If there's a particular type of PowerPoint format you like, you just point it to the work you've already done and say, “I want to do a new presentation using this style, but on this topic,” and it'll actually emulate your style and the format of the work you've already done. People are very impressed with this. You can also point it at your email and have it analyze your email and pull things out of it.

Right now, it's very task-based. You, the user, have to prompt it for each task. But you can see the beginning of a personal digital assistant where you connect it to your file drive, your email, and all of your data sources, and it can start to do tasks for you. Again, it understands the format and the style that you like to produce work in. It feels to me like we just need one more layer of abstraction on top of a tool like that, and you'll have your own personal digital assistant.

And there'll be a voice interface. Have you ever seen the movie Her, with Joaquin Phoenix and Scarlett Johansson? I think Scarlett Johansson is just the voice, but he's telling her what to do through an earpiece. We're very close to something like that. I'm not saying that the AI is going to become sentient or whatever, but I think in 2026 you could see these types of tools—again, they started as coding assistants, but now they become personal digital assistants. That could definitely happen this year.

Michael, what don't people understand about AI? What do you think is most important for us to understand about the innovation underway right now with science and AI?

I think it's easy to underestimate the long-term impact this is going to have across so many industries and domains. It's easy to quickly think about AI as just a sophisticated chatbot because that's what most people interact with every day and what they touch and feel. But to me, the long-term impacts—not to keep harping on the science—I think there is a real fundamental shift happening in the velocity and pace at which we can test, evaluate, and execute scientific discovery and endeavors. I think that's going to have huge repercussions for the way that we, as a country, innovate, broadly speaking, in the years ahead.

Which is why we're watching what China is doing. Let's talk a bit about China and where it is relative to the United States. Are we winning? Is it about chips? What's the race specifically really about?

Well, I think that in general we're ahead of China. There are different layers of the stack. You've got the models, then you've got the chips, and then you've got the chip equipment. So you go down the stack. I would say that the deeper in the stack you go, the greater the American advantage.

On models, most people would say that our models are maybe 6 months ahead, plus or minus, of the Chinese models. You look at chips, maybe 2 years ahead. You go to semiconductor manufacturing equipment, and it could be 5 years. So the US does have significant advantages there.

There's only maybe a couple of areas where I think China has an advantage. One is energy production. If you look at their grid, their grid has roughly doubled in the last 10 years, whereas ours has only grown by about 2% to 3%. Energy production in the US was a relatively sleepy industry before AI came along. A lot of that had to do with regulations and the antipathy of the previous administration toward energy production.

Obviously, President Trump had a very different view on this. I think he was prescient on this issue. You go back 10 years, and he was talking about, “We've got to drill, baby, drill.” I think he understood that energy growth was the precondition for economic growth, and it's definitely the precondition for this AI infrastructure growth. So this is an area where, again, we have to basically expand our energy production, and I think that is an area where we need to catch up.

The other area where I would say—I don't know if I would call this an advantage exactly, but you could argue that China has the edge in what's being called AI optimism. There was a poll done by Stanford across countries, and they asked the citizens of all these different countries, “Do you feel that the benefits of AI will be more beneficial or more harmful?” If you thought that it would overall be more beneficial than harmful, they called that AI optimism.

In China, AI optimism was 83%, so 83% of the population feels that it's more beneficial than harmful. That number in the United States is only 39%. For some reason, people in China are more optimistic about AI than people in the United States. You generally see this: Asian countries are very high on AI optimism, while Western countries are lower. I think it's an interesting or open question about why this is. I think there are a few possible explanations for it.

First of all, the media tends to focus on the doom-and-gloom stories with AI.

The fear.

The fears. We can talk about some of those fears and whether we think they're real. But I think the media has a lot to do with it. I think the way that Hollywood has portrayed AI over the decades, whether it's The Terminator or 2001: A Space Odyssey, has portrayed this dystopian view of the future. I think that plays into people's thinking.

Then, frankly, I would say that part of the fault lies with our tech leaders, who haven't necessarily done a great job describing the benefits of AI. In fact, when they're talking about AI eliminating 50% of knowledge workers, that doesn't sound like a very utopian scenario. That sounds dystopian to most people. So I do think that, unintentionally, some of our tech leaders have played into this AI pessimism.

The reason why I think this could be a disadvantage for the US is because, again, it's feeding into this regulatory frenzy we're seeing—again, 1,200 bills at the state level. Right now, I think we are winning this AI race. We're ahead in all the key dimensions—chips, models, and so on. But we could shoot ourselves in the foot if we end up overregulating this thing to death. We could actually cost ourselves this AI race. So I do worry about this question of AI optimism.

Right, it's a great point. What would happen if the US is not number 1 in this, Michael?

Yeah, I think we need to be, and that's why we put the plan out. When I think about the China question and the larger question of how we win the AI race, what I always like to think about is this question of adoption. I think sometimes there's this overemphasis on the leaderboard—it's like, which frontier model is number 1 on some sort of metric? The reality is we're neck and neck, and as David said, we're probably ahead 6 to 12 months on our frontier models.

But I think what we have seen over time and over history is that you don't necessarily need to have the very best model or the very best piece of technology in the world for it to proliferate globally. A lot of us who were part of the first Trump administration saw this firsthand with the telecom wars of that era and what Huawei was able to do globally. At the time when Huawei first started its global export push, it certainly was not the very best technology in the world. It was certainly subpar compared to Ericsson and Nokia, but it was good enough, and it was subsidized enough that it became the default telecom system for a lot of the world. We've learned a lot of lessons from that, and we take that very seriously.

When it comes to AI, we know there's an ambition for the Chinese to export their models and have them be the models powering all these different use cases across the Global South and across the rest of the world. That's why the president launched something called the American AI Exports Program. Our mission—and I think we're in a very lucky position here compared to what we were dealing with with Huawei—is, as David said, we are dominant in almost every part of the stack.

We have the very best models. We have the various applications. We have the very best chips. So, we are in a position of power now, and it’s up to us as a country to share that technology with the world, with all of our partners and allies.

We need to make sure that any developer anywhere in the world that wants to build a new application using AI is using and fine-tuning an American model on top of an American chip. That isn’t a hard reality to see. That is something that I think we can very easily do just because we have the very best technology.

That is a program that we launched late last year, and we’re doing a big push this year to get that out the door.

It’s an important point that you make in terms of exporting AI to the rest of the world. Is it true that China is telling its companies, “Don’t use American chips. Don’t use American AI right now”?

It seems so. China is developing its own models. Obviously, about a year ago, you had the DeepSeek moment, where you had a powerful model released by DeepSeek, and I think that kind of put Chinese AI on the map in a way.

I think people in the West didn’t realize how good China was at producing models, and there was a little bit of complacency toward our relative position. People weren’t really talking about the global competition 2 years ago. It wasn’t really discussed at all.

I remember when the Biden administration created this 100-page Biden executive order regulating AI. No one was talking about whether all this regulation might slow us down vis-à-vis China. It wasn’t even part of the conversation. Then DeepSeek launched, and I think we did realize we’re in a global competition and we have to win. That’s why we have to actually be quite careful about how we regulate this and make sure we’re not overregulating it.

I think China definitely wants to compete. There have been some stories recently—I think Bloomberg and Reuters reported that they actually are not allowing NVIDIA chips into their country. The reason for that, we think, is that they want to indigenize chip production. They want to stand up Huawei as their national champion, and effectively they’re creating a market subsidy for Huawei by keeping out the competition.

They’re protecting their market to stand up Huawei. I think their plan would be to have Huawei dominate chips in China first, use that to scale up, and then try to take over the rest of the world. Chip production is a scale-up business. So, if they can dominate the Chinese market first, that gives them a powerful platform to then proliferate to the rest of the world.

So, where are we in that, Michael? First, you all came up with the AI Action Plan, then came up with another plan in terms of exporting AI to the rest of the world. What can you tell us in terms of where we are in that?

The progress is moving on that. We closed a request for information from the Commerce Department late last year, which went out to industry and said, “Hey, if we want to export the American AI stack, what should we be thinking about? How should we be designing these packages that we share with the world?”

Commerce is now ingesting that information. There will be a request for proposals that comes out very shortly. That’s where we actually want companies to come together to form consortia and say, “Look, this is what a package looks like.”

What I always try to remind people is that the buyers of AI around the world vary quite dramatically in their level of sophistication. In the U.S., if you’re a Fortune 50 company and you want to deploy AI, you have a pretty sophisticated CIO or CTO shop. You’re thinking very carefully about which cloud you want to buy, which potential model you want to use, whether you want to fine-tune it on your own data, and whether you want to build your own application.

You can test various things. You can go to all these third parties and evaluate which is best. It’s a very complicated mix of how you end up creating something that’s optimum for your particular company.

For a lot of countries around the world that are aspiring to use AI for their people or to support services, whether it be health care or tax collection or whatever it may be, they don’t have a billion-dollar IT budget. They’re just trying to figure out what is a tool that they can use in their country to deliver the benefits of AI to their people.

So, we think very carefully around how we can craft solutions that could be turnkey, to use one phrase, or how we provide a solution that can easily be deployed in a country. What often gets caught up in this debate is the question of how many chips the U.S. is going to be sending around the world.

What I always try to remind people is that outside of the U.S., China, and maybe a few other countries, most countries around the world do not have the capital or the aspiration to do large-scale training runs or develop their own frontier models. There are very few countries around the world that are going to build Colossus-style training centers.

Most countries around the world need smaller data centers that just have inference-related chips that can drive and do the inference on the particular runs that the government wants to have. So, I think what we’re working very hard to do is create these turnkey, manageable-sized AI solutions.

Then we can partner with a lot of our export finance organizations, like the Development Finance Corporation or the Export-Import Bank, to make the export of that particular stack much more appealing and commercially viable in countries that are not extraordinarily deep-pocketed.

We’re going to be in India next month for the India AI Impact Summit. This is the largest global gathering for AI folks, and we’re going to be sharing a lot more on the progress of this program there.

You want to weigh in?

Well, just to build on that, I think people sometimes ask, “How will you know if you’ve won the AI race with China and with other countries?” I think there’s a very simple answer to that, which is market share.

If in 5 years we look around the world and we see that American chips and models are being used everywhere, that means we won. But if in 5 years we look around the world and it’s Huawei chips and DeepSeek models, then that would be very bad, right? That would be a bad sign. That means that we lost.

So, I do think that the proliferation or diffusion of American technology is really critical to winning this AI race. We know from Silicon Valley that the companies that end up becoming huge are the ones that create ecosystems.

As a technology company, you want to have the most apps in your app store. You want to have the most developers writing on top of your API. You want to be a platform company. In all these technology races, the biggest ecosystem wins.

That’s basically why I think this program is so important: We want to create the biggest ecosystem. Now, this is not only about benefiting the U.S., because in order to have a successful ecosystem, you have to create value for your partners.

As Michael was saying, not every country is going to be on the cutting edge of developing its own chips or developing its own frontier models. But they can use these tools to derive value, apply them to their businesses and their economies, extract value, and be part of this technological revolution.

So, I think we have to think with this partner mindset. I do think that this type of mindset is actually very common to Silicon Valley. Every great technology company thinks in terms of how we get the most people on top of our tech stack.

But it is a form of thinking that’s pretty alien to the bureaucracy in Washington, which has much more of a command-and-control type of mindset.

When President Trump came into office, just to give a couple of examples, the regulations that were sitting on our desk had just been handed down by our predecessors. Again, we had this 100-page Biden executive order on AI that was all this new regulation, and there was a 200-page document called the Biden AI Diffusion Rule, which was 200 pages of regulations on the export of semiconductors.

So, we were turning the AI industry—models and chips—into a highly regulated industry. That was basically the direction that Washington was going in. The first thing President Trump did in his first week in office was rescind all of those unjust regulations, which I think was absolutely critical.

The thing that really makes Silicon Valley special is this concept of permissionless innovation. Since Hewlett and Packard started building Silicon Valley 85 years ago, the idea has always been that just a couple of founders with a great idea start their company. They get some angel investors to write a check for seed capital. Those investors think they’re probably going to lose their money, but they figure there’s a shot.

It could be the 2 guys in their garage, or it could be the college dropout in the dorm room. They don’t need to go to Washington to get permission for their idea. It’s permissionless innovation. That’s what has made Silicon Valley the crown jewel of the world, and it’s why so many of the heads of state who are here are always asking, “How do we create our own Silicon Valley?”

That was not the direction we were on when President Trump came into office. The new 300 pages of regulations concerning AI that the Biden administration left us with would have changed this environment of permissionless innovation to an environment where you have to go to Washington to get approval for your idea.

I think President Trump really corrected that. Since then, we’ve been implementing his AI Action Plan, which is all about being pro-innovation, pro-infrastructure, pro-energy, and pro-export. It’s been a total change, and I think just in the past year you’ve seen the results of that.

I think one thing to add there is that part of the international agenda that we have on AI is, one, obviously, to do the export. But the other piece is trying to share with all of our partners and allies how you can actually create a regulatory environment that allows this technology to succeed.

Here we are in Europe, and I think many of us who have tried to work with technology companies in Europe have hit a lot of roadblocks and a lot of stumbles. The Draghi report came out, and it says that there are a lot of issues, but things don’t ever seem to really change.

And I think all of that—the way our regulatory structure is designed in the US and the way the entrepreneurial spirit thrives in the US—is something that we try to share with countries all around the world. I think the general knee-jerk reaction for most policymakers around the world is to move to a corner obsessed with the precautionary principle.

This is the concept that every time something new comes out, the role of the policymaker is to sit in a room and whiteboard everything that could go wrong, then design regulations to make sure those hypothetical wrong things don't happen. When in reality, what we try to do in the US is sit in a room and whiteboard what rules we can create to actually unlock innovation. What are the ones we should remove to allow more innovation to happen?

I think that mindset is something that we constantly try to share at all these international fora. There has been an A/B test on what regulatory structure works and what succeeds. We've seen how Europe has approached this over the last 20 years, and we've seen what the US has done. I think the recipe is kind of obvious, but sometimes we have to keep repeating it to our counterparts.

And I love the Draghi report because it clearly identified companies in Europe. Novo Nordisk is a $350 billion or $400 billion company, and in America we've had trillion-dollar companies, with Nvidia hitting $5 trillion.

So, what is the path to innovation?

Well, I think part of it—and I think this is the difference between maybe the American mindset and the European mindset toward this—is that ultimately, innovation in the United States comes from the private sector. It comes from the entrepreneurs, the founders, the innovators, the geniuses with an idea.

I think the government sees its role, at least when it's thinking properly about this, as being an enabler and just setting the rules of the road, and maybe putting in some guardrails. But basically, it's letting the entrepreneurs cook, and that's how you get innovation.

I don't want to bash our European hosts too much, but when the EU talks about AI leadership, they're talking about the regulators. They think their value-add is, "We're going to show the whole world the regulatory model for AI." So, it's a bad case of main-character syndrome, where the regulators think they're the main characters in this.

No, look, the regulators are the supporting players. The main characters always have to be the entrepreneurs. It's got to be the innovators. That's how you unlock innovation. When the regulators and policymakers start to see themselves as the main characters, that's not a great recipe for innovation.

And I think just a minor point on the AI situation in Europe is that the EU AI Act, which has been so detrimental to the AI ecosystem here in Europe, was passed before ChatGPT was even invented. That shows the challenge here: you're believing that you can solve some kind of problem, but at the end of the day, innovation is moving so much more quickly. Ultimately, that rule makes no sense now in a world of frontier models and large language models, and they have to edit it.

So, let me push back before we go and ask you to identify any risks, threats, or downside risks in all of this. What should we be worried about, if anything, with regard to AI usage?

Well, I think there are Orwellian scenarios with AI that we should be concerned about. Again, I tend to think those scenarios were described by George Orwell, not by James Cameron and The Terminator. Specifically, it's the misuse of AI by government.

I do think AI could be used as a tool to surveil, censor, and even potentially brainwash the population. This is why the administration has taken such a firm stance against what's called "woke AI," which I almost think trivializes the magnitude of the problem we're talking about. We're talking about AI having a political bias built into it.

The bias can be so subtle that people don't even necessarily notice it over time, but it has a huge impact on what people are allowed to learn, think, and know, and on what children learn. So, I think it's very important that we try to make sure that AI is politically unbiased.

One of the things we were so concerned about with that Biden executive order on AI, which we rescinded in the first week, is that it had 20 pages of language on DEI. It was promoting this idea that AI models need to build in a DEI layer. Well, this is how you ended up with the Black George Washington story, where the first version of Gemini came out and was basically rewriting history to serve a current political agenda of DEI.

That case of bias was so ludicrous that everyone kind of laughed at it. But it gives you a sense of what could happen if you start to build the bias into AI. That same so-called trust-and-safety apparatus that was starting to be built into social media sites as a way to censor, deplatform, and shadowban could be built into AI models as a way to control the public discourse in a very serious way.

I think President Trump just put a total halt to that; he rescinded it. President Trump also signed an executive order saying that the federal government would not procure politically biased AI.

So, look, on a First Amendment basis, if an AI company wants its AI to be biased in some direction, it probably has a First Amendment right to do that. But we, as the federal government, have the discretion not to buy that software, and we've said that we won't.

I feel very good that during President Trump's term in office, for the next 3 years, this idea of Orwellian AI is not going to be a problem. But I do worry that at some point in the future, if you had a different regime in Washington—if the federal government started to pressure AI companies to build in this political bias—that would be a very serious threat to our freedoms.

It's a great point to make. Before we wrap up, real quick on jobs, can either of you explain what Elon Musk is saying about the impact of AI? He said we're not going to need to work. AI is going to do it all. I'm trying to understand what he's saying—that we're going to go on holiday, jobs are going away, and AI is going to do everything.

Well, Elon's a friend of mine, and I'll disagree with him slightly on this. But let me just say, his comment about job loss is obviously what gets all the headlines, but at the same time, he's also saying that in this future, there's going to be so much abundance that everyone's going to have what they want and there's not going to be any money.

People leave out that part of the story and just report, "Elon says everyone's going to lose their jobs." No, we're talking about a radically different future. It could be the future described in Star Trek, where there is no money because we have everything.

I think Elon is directionally correct about the future. I think we're heading toward a world of much greater abundance, rising living standards for everybody, and greater productivity. I think that will lead to rising wages. I don't think it's going to put everyone out of work. I don't think that's going to happen.

But again, the timelines matter a lot, and getting to a world with no money is not something that's going to happen in the next 5 years.

And of course, Michael, this is helping us in terms of longevity and living longer, right? In terms of the impact on science.

Totally. I think generally the abundance story extends well into health care and everywhere else, including quality of life. So, good things ahead, I think.

We'll leave it there. Michael Kratsios and David Sacks, thanks so much.

Thank you.