Under Secretary Emil Michael: Pentagon gives 3 million employees Gemini access for 47 cents, bans DeepSeek

happen. Um anyway, we have our next guest in the reream waiting room. We have Emil Michael, the under secretary of defense for research and engineering of the United States. Emil, good to see you. Merry Christmas. How are you doing?

Welcome to the show.

I'm doing great.

Thanks for having me.

Thank you so much for taking the time. You look fantastic. Yeah, I would say thank you for wearing the suit, but unlike a lot of our guests, this is a daily driver, I'm sure.

Yes. Yes.

Uh,

well, we were buying a high school football team today, so I had to dress up.

Oh, nice. That [laughter] that that would be a real treat. Uh, well, you're not you're you're [clears throat] Is it Is it fair to say that you're buying Gemini? Are you buying tokens? or is this one of those deals where uh it's the partnership is structured so that it's not really uh you know a major cash infusion for Google or it's not a burden on the taxpayer uh it's a little bit more of uh just two great American institutions working together. How are you framing the deal? Um it's it's uh making to three million Department of War employees that includes military members. Yeah.

And civilians access to Google uh Gemini on the our network which is a private network, right? So it's you actually have to port it over.

Okay.

And they gave us a great deal for the first year. It's like 47 cents. So [laughter]

So they're very patriotic. I'm super excited to to have that in the hands of everyone at the Department of War. It's never happened before.

Yeah. Um were there historical examples that you were pulling from like like because I imagine that service members have been googling things for a long time. Uh I've heard stories about I mean even during the during the war in the Middle East uh the war on terror there were using Google maps just to see satellite imagery. Uh, and there are a whole host of reasons why a war fighter might want to use a Google product. Uh, is this a new structure? Are you building on the shoulders of giants? Do you feel like when you came in there, the Department of War had uh, a good a good reference point for how to work with a big tech company like Google or was it sort of new territory? Uh, how did you think about actually interfacing with Google?

I mean, it's entirely new territory, right? because you could get those apps off the app store or on a web browser. So to get AI in there's a lot of sort of fear about AI what it could do and then you have to remember when you're doing a Google search like yeah it it kind of records where you're searching but if you're putting stuff in AI

you don't want that to leak back into the model so you have to architect it in a unique way. So it's kind of new territory and we did it all in like three months.

That's amazing. Uh yeah, what did the team on the Google side look like? Is this I mean I imagine that they have a you know a large uh team just for working with the government. Um uh you you mentioned a few uh it was a few weeks of work. Uh but what what did they bring to bear to actually deliver this on their side? I mean they they had a ton of engineers

um because you got to put it you know in our networks which are not simple right and you have three thou three million people all over the world all kinds of um devices and so on. So they really brought the heat right they and they had a you know for today for launch they had a a room or they you know war room if you will no pun intended and as did we to make sure that it launched it launched correctly. I mean, we got a lot of a lot of flack on Twitter today.

Yeah. Because So, yeah, I saw these comments, but it was very obvious if you used one brain cell that you needed to be on a on a on an actual approved device on a military network if you wanted to access the service. So, uh I didn't I didn't I didn't get the flack.

Yeah. Well, people want to give flack just to give flack, right?

[laughter]

Well, I think some people wanted to they wanted to try it out themselves and it's like, well, if you want to do that, you can join uh become a service member.

Yeah.

Yeah. We put a we'll put a we put a link on the error message a few hours ago for job openings at Department of War if they want to join. So, they get access to Gemini for government. We're open for business.

Yeah. What what what else is uh have been the highlights of the year as you look back on the year? Uh obviously confirmation took time but then uh there was this big uh White House AI action plan. Uh what are some key moments that you feel like uh have been maybe under discussed or under didn't get enough appreciation at the time?

I mean you know there have been like three AI executive orders which is a big deal right and Jensen said the other day that no other administration has ever done this much for the AI companies. So we're really championing championing like these four national champions, right? Google, OpenAI, Anthropic, and XAI.

Um, and giving them like we're talking about, you know, data centers on military land. We're talking about how do we get more power to these uh these companies because it's it's a race and I think we're the first group in the last six months to to really realize that we better get great. We better [clears throat] get better and stay great and ahead of everyone else on this because it's really the next generation of tech.

On the subject of data centers uh on military bases or land, what kind of comps have you looked at on that side? I know there's like a historically in in weapons manufacturing you have like go which seem like you know they would provide some framework. Uh but how how have you thought about that?

We haven't thought about that yet. we're we haven't broken ground yet or to the business models we just had the executive order said we're going to do it

and then we got to figure out um how do you prioritize who gets what right and we may reserve some of that data center capacity just for the government to use for our own purposes but the point was to send a message that we're all in on AI companies all in on the data center power chip needs of these companies in a way that the last administration we think was just trying to constrain everything with these orders to to really uh you know have one company win and that's it so they can control sort of the outputs. We're not we're going the other way. We're going to have all the national champions. We're going to support them all the way. You know, David Saxs is out there making sure the American tech stacks and all other, you know, as many other countries as it can be.

And that's a big deal, right? It's, you know, if anyone's going to be using AI, we want them to be on our stack.

Yeah. How do you think about the the the the fact that uh all of the big tech companies uh seem to be working with the United States government effectively for free? 47 cents. I've heard of deals of of a dollar as as high as a dollar. Um but then I think back to, you know, a few decades ago where Microsoft had a deal to develop virtual reality headsets, augmented reality headsets for the US government and I think it was like a10 billion deal. And I was really excited when that project went over to Anderl because it's a lot, you know, younger faster moving company. Now I love Microsoft. I love them as a infrastructure provider and the and the fact that they have OpenAI's IP is amazing. There's so there's a ton of great things that that uh that Microsoft does. But developing an AR headset that always felt like it was going to be a rough go. It was such a big such a big burden on the taxpayer at10 billion. uh much better to see it in a different uh in a different uh land in a different company. How are you thinking about when the how the government is working with big tech, when to open up the pocketbook and write a billion dollar deal or when to ask for, you know, 47 cents?

I Yeah, I think, you know, we want these companies to be successful, so we're going to pay them market rates eventually. It's just

how do you get out out of the gate? you get out of the gate by giving 3 million people access and not having to worry about like a token count every every day, right? And then that's exciting to them because

now they get 3 million more customers, right? And they get to learn use cases and so on. So,

um it kind of works for them, works for us, but we need to pay them fair fair rates, but they're going to be commercial.

Yeah. And I imagine that uh there like when a new technology comes up like you know Gemini uh there's just a lot of service members that are just you know not uh by no fault of their own just like accidentally using it a little bit too cavalierly and this is an initiative to actually refocus on security refocus on privacy but can you take me through some of the you know I'm not particularly worried about about uh Google having you information from the American war fighter, but I am worried about some of the international uh chat bots having access to the American war fighters prompts. Uh can you take me through how the Department of War is thinking about uh LLMs like Deep Seek, Alibaba, these other AI models? Um are those just banned everywhere? How do you think about actually um you know banning different uh different you know vectors of attacks from a security perspective?

Yeah. Well, I mean first of all we're going to train you everyone is going to train if they want to use it for like you know higher risk use cases, right? But this is going to be on their devices just like the web browser is and they could still search Google. Um, but when it comes to the foreign companies like DeepSeek and Chinese, like in the bill that was just released by Congress, I think yesterday, we're going to ban, you know, Deepseek and all these uh foreign models from use by do members and contractors and anyone who touches it because last thing we want is for those models to get data on how we're using uh AI, right? That would be that would be a a tragic mistake. So, uh, I think we're we're already there, but we're going to it's going to be in law pretty soon.

Yeah.

Uh, what's your what's your pitch to, uh, talent to come work with the Department of War, uh, specifically in AI? It's like the craziest talent war, uh, of our lifetimes, most likely. you know, you're trying to recruit people that can go get uh I don't know, anything from a million dollars a year to $100 million a year to a billion even [laughter] up to some people literally a billion dollars.

And so and so I imagine you have to have a, you know, pretty dialed in pitch to recruit people and it's and and maybe that pitch is you're going to be able to work on I mean the H200 news today says that, you know, we aren't the you know, the H200 is not like a F-35. It's not military technology yet, but I'm sure there's a pitch

to work with you and your team on applied AI in a military context, but what do those how do those conversations go?

Yeah, so I I call them recruiting Tuesdays, right? I spend all afternoon Tuesdays calling, dialing for dollars, emailing, referencing, interviewing, and my whole team does that. and uh like the pitches like you know okay you have to have some patriotic instinct obviously but then maybe you're in between companies or you're just motivated I mean Elon has five companies and he still worked at the government for like seven months and brought a ton of Doge people who are motivated by the mission so I'm trying to motivate people by hey this is the biggest technology deployment in the world there is no bigger organization than department of war there's no more exquisite crazy interesting use cases is whether they're intelligence, war fighting, you know, even corporate use cases and you get to be part of creating that. It's never been done before. It's a pretty good pitch. And then if you when you leave the Department of War and you've done that, uh I think you get more valuable in the private sector, right? All these big AI companies now have big federal businesses. they all have sort of um cooperating with like the government in different ways and you're even you have another notch on your belt of good stuff that you've done and been an innovator on which is pretty rare in government to like literally innovate while you're sitting in the job on something that's never been done before that's that big.

Yeah. How how do you uh

how do you think about the um the different levels of AI diffusion, AI integration where AI can actually help the war fighter in the American context. I mean a lot of people uh kind of watched the Palunteer story evolve over the last few decades. this idea of of analytics, just putting dots on a map, sometimes that being enough to identify where an enemy threat might be, for example. How do you think about integrating AI from the most mundane use cases of just speeding up a little bit of paperwork here and there to some of the bigger questions that will be ultimately face the the the war fighters of today?

Yeah. I mean, imagine on like the most simple use cases just like any big organ like an employee in any big organization, right? You're it's writing PowerPoints for you, writing job descriptions, like making spreadsheets, the basics. Yeah.

Then there's like the cool intelligence use cases. Imagine that we've got decades of satellite imagery, decades of that or sensors that we've had or, you know, all all kinds of things. And now you could instead of one human analyst having to go like I think I see that there you you can go back through 50 years train a model and say look for things you've never seen before right

um and then on war fighting logistics planning you know all kinds of uh simulations if you want to simulate a war game

and a really incredible way with all the data and all the stuff in there. So it's like pretty pretty compelling, you know, and exquisite to mundane that you could do with this stuff when you deploy it the right way.

Yeah. How do you think about the I mean this partnership with Google? Uh it it feels very much like giving you know the almost the consumer product of Gemini to everyone in the military. Uh how how are you thinking about larger projects that might require bring together custom code, multiple systems, multiple pieces, you know? I mean just as a consumer we're running into things where anthropic might be better for coding and one day and the nano banana is better for this and chatd is good for deep research and if you want to build a business we see a lot of startups pull three different models off the shelf uh is that are is that something you're starting to look at as like a phase two of AI integration into the military? Yeah, I mean we're so ideally we'll have all all four models

and all four of the newest versions like uh all the time at every classification level. Wow.

And the reason you want to go up through the classification levels is so you can do more exquisite work, right? More complicated stuff. Um and then the reason you want all models is just what you said. If you're trying to write code, maybe you want to use claude if you're trying to do these other things. So as they compete, we want to benefit from it. And we've never had that here before. So it's it's trying to consumerize it, give choice, but then give more and more capabilities as you go up the chain of like uh of security classification.

Yeah. Who who who owns the data? Does the taxpayer own the data? Does the government own the data? I see a lot of stuff where it's like, oh, this this company now owns Google's going to have your your government data. Google's going to have the the data. It seems like they might be storing the data in some points, but they don't have authority over it. How do you frame to like the American who's maybe worried about government overreach or or or wants to understand where important critical military data is living? Who has control? Who has final say?

Yeah. So taxpayers own the government. Government owns the data on behalf of the taxpayers. Google does not own it. We control it. Yeah.

Uh it doesn't leave our control.

Um and that's that's why it was complicated to actually launch, right? because you like AI models are built so that they learn as you continue to query them, right? And and to make it such that the model we got the latest model and it took in queries but then the learnings from that don't go back into the general model y

is a form of making sure that the American taxpayers data doesn't leak anywhere else and just use for the purpose it was intended.

Yeah. How many other organizations globally, you know, allies, things like that need to take on pro like a project like this? Like it feels like you clear clearly it's very important to move quickly because if you if 3 million people can't use the latest AI models uh in a in a secure way uh you know that that it's not it's not super sustainable. They'll eventually you know go elsewhere to get these products. uh do you think a lot of other kind of groups need to you going to pursue a project like this and are you guys trying to help create kind of a framework so others can can benefit?

Yeah, I mean any well so we all know and you guys know any large organization needs to be using AI just for efficiency purposes right and and just you know dollars and cents if your competitors you know showing more profit in in Q1 relative to you because they have too much overhead that could be solved with AI you're going to be in trouble. So when you talk about governmental organizations, should every government be using it in some degree? Yeah. I mean it's it's sort of like an economic imperative, but it's like a strength imperative too. Um it's sort of like the way I describe it to people is you're opening the human context window. So we're not replacing a war fighter, but we're allowing the war fighter to be to do more with the same amount of time, right? To analyze more intelligence, to do more scenario planning and all that. So, I can't imagine any other organization like ours that wouldn't be thinking about this. Um, but hopefully I think we we are setting the pace here. We're going to be the pace setter here for the US government and for governments around the world.

I love it. Well, thank you for setting the pace. Thank you also for coming on the show today. This is fantastic.

It's fantastic to have you and thank you uh thank you for all the work that you're doing.

Yes.

On behalf of you guys. Thanks for having me on.

Every citizen. Come back on anytime.

Anytime. Always.

All right. Thanks a lot, guys. Have a great meal. Goodbye. Bye. Let me tell you about profound. Get your brand mentioned in chat GBT. Reach millions of consumers who use AI to discover new products and brands. Um Patrick Collison had a post that hit the timeline rocked the timeline with 1.3 million views. He says, "Two conversations this weekend make me think that there's a vibe shift a foot in Silicon Valley around what one should work on and what is worthwhile. Culturally, it feels like the moment is ripe for new frameworks. One Davos expert morality is stale and

discredited.

Let's give it up for the experts. Joe Rogan, [applause] Andrew.

Yeah, Lex.

Those guys don't podcast Davos. They don't. They're not Davos.

It's very possible the next Davos could just be a round table between Joe Rogan, Andrew Huberman.

Yes,

the real expert Chris Williams Freedman. We'd love that. That would be fantastic. Uh, it's also apparent that the quote just be superbased counter enlightenment is not really an answer. Yes, woke went too far, but simply inverting it does not work. Good point. Uh, effective altruism is no longer the automatic default for smart people. They pushed it too far. They tried to save too many shrimp from the their farming fates and now they are cooked.

Sorry shrimp. There is increasing skepticism of slot and slot machine dynamics. People don't want to work on a slot machine. They don't want to work on a slot machine. They want to work on something worthy and valuable. The question is overall, what is worthy and valuable? It feels like this question is becoming more central. What do you think is worthy and valuable?

I think uh it's worthy and val valuable that OpenAI hired Slack CEO Denise Dresser as chief revenue officer. Hm.

It's the scoop. It's happening right now.

Wait, what? What?

OpenAI has hired Slack CEO Denise Dresser, chief revenue officer.

That's that that that really flipped me for a second because they have a CFO. They have a CEO of applications. Uh I don't know where CRO fits in, but they got someone. And that's very exciting. Um is that gongworthy? Let's