Sriram Krishnan (White House AI Adviser) breaks down the US-UAE AI acceleration partnership and American AI dominance strategy

with the administration. Uh, very excited to get his take on what's going on since he was there on the ground. So, let's bring him in. Shirram, welcome to the show. Wow. looking fantastic in a tie. Making us look casual. Making us look casual.

Um I I just want to say people might think that I'm dressed up because I work in the White House. No, I'm dressed up because when you're in the presence of the technology brothers, you have to come correct. Thank you so much for joining. Uh welcome to the show. Long overdue.

uh why don't you I mean it it it'd be great to start with just like a brief introduction on like how did you wind up in the position that you're in and then uh what's what's uh the latest update uh with the trip to the Middle East. Yeah, let me do that.

But before that, I just want to say John Ji, I've known you guys for a while. I have two overwhelming emotions about being here today. One is just pride at what you guys have done with uh the technology brothers and second is a fair amount of jealousy because I tried to do this I couldn't.

I fail, but you guys have succeeded and you're crushing it. So, congratulations. This is amazing. So, yeah. I mean, we're building on the shoulders of Giants, Clubhouse, Good Time. There's a little bit of that in here. Obviously, different medium and different uh different format.

There's some There's some alpha and like tall podcasters, too. Yes. Yes. Underated. You guys You guys both Yeah. The average height on this call right now is like 66. It's great. That's the bar. That's the bar. But, no, thank you. Thank you for having me.

Uh so uh and by the way I would say you know this is the first platform podcast TV channel that we're talking about this on when like when you got to talk about there's only one platform it's you guys but no thank you for having me on. A little bit of backtory I think some of your uh viewers might know me a little bit.

I've had you know I've been in the technology industry for a while uh you know working in a bunch of large tech companies. I was at Andre and Horovitz as an investor for many years.

uh you know I I I was doing a podcast uh with my wife Arthy and I kind of been around on the block and the way this happened is a couple of things.

Uh like millions of others I was convinced that artificial intelligence and what is happening is maybe the most important thing to happen maybe in our lifetimes uh and definitely in the next few years. That's number one.

Um second is I was convinced that what the previous administration was doing around AI was just the wrong direction and the wrong direction for America when it came to winning the global AI race.

So after the election I had a conversation with David Saxs who is another podcast mogul but maybe more importantly the new uh AI and cryptozar and we were discussing some ideas and David said hey listen why don't you come work at the White House. I was like, well, what does that mean?

Long story short, a couple of days later, uh, I got a call saying I was part of the team and it's been an adventure kind of ride so far. And so, you mentioned the Middle East. So, we're just coming hot off that.

And just to set some context, um, on the second day of the new administration, President Trump signed an executive order.

And that executive hard a very clear mandate for the country, for David, for all of us which said that America does and should continue to dominate the global AI race, American AI domination where basically our marching orders and we've been working towards that uh relentlessly ever since and this trip is pretty historic as a part of American AI dominance.

Um so to kind of understand this, we should kind of go back a little bit. So for the last say a little bit of time, if you were an ally in the Middle East or a lot of other parts of the world, if you wanted to get access to American AI infrastructure, you pretty much couldn't. You were the doors were closed.

You were told to pound sand. There was really no way to really get access to the technology that our friends, a lot of our friends, a lot of our networks in Silicon Valley uh go and build. And this week changes all of that.

So on Friday, Thursday, uh the administration, President Trump as part of this historic first visit of this new administration to Middle East in Abu Dhabi, we signed the first AI acceleration partnership. Okay?

And you guys probably read about in the press, but there are probably three important components that is, you know, I wanted to have the technology brothers have the alpha and have the first scoop on that. Amazing.

the the the most important part the first part is that this represents a large investment in US data centers and US AI infrastructure. So these countries will be investing in US AI infrastructure to make them as equal if not larger than the data centers and infrastructure they're building back home.

So this means obviously a large infusion of capital revenue to data centers here in America. So that's story that story was kind of lost, right? I feel like a lot that a lot of the focus was on the localized investment and infrastructure that various groups.

So, so just to break it down in in uh in a ter in language that venture capitalist could understand.

Um, this is something like uh what we're seeing with Stargate where there's a ton of capital forming and that's coming from Soft Bank, but it's also coming from uh Middle Eastern investment funds and sovereign nations investing in American infrastructure.

And then there's a whole host of companies that might come in the stack to actually build a new data center. Is that right? Exactly. Oh, you should you should be doing our talking part. That's perfect.

I would say look these countries have AI ambitions right they want to buy American AI they want to buy semiconductors they want to buy our large language model they want to use us and so as a part of this deal they're agreeing to a few things the most important thing they're going to agree to is that capital like you mentioned right like and and this is by the way net new this is not part of any existing project these net new deals will mean infrastructure being built out physically in the US so for example if they build out x you know megawatts or gawatts of capacity This will mean the same x megawatt of capacity in the US.

And this is an important point because some of the chatter has been hey how does America maintain its lead? Well, one of the ways we maintain our lead is everything that is being built up by allies we get a matching deal back home. So that's probably the number one headline.

Um the second headline would be that the vast majority of the GPUs that are you know as a part of this deal which is going to be say hosted in the UAE will be hosted run operated by American hyperscaler companies right so you know you probably know them all right like these will be large American companies so they will be running it hosting it maintaining it and this is actually important because this represents a expansion opportunity for all of our companies this means they would get to win market share away from competition from other countries ries and obviously there's a whole huge amount of revenue and ecosystem coming in and so that's the second key point the vast majority of the GPUs are going to be run by American companies often by a lot of our uh you know friends in these large uh you know hyperscaler companies and the third point and this is again something that was lost in the chatter is I'm sure you know you've heard questions about hey how do we make sure these GPUs you know don't get to somewhere they don't need to be so there are rigorous security protocols in place so every GPU you get get shipped over.

We we're going to make sure that a that they can't be physically diverted. These are really large boxes. You can't hide them under your t-shirt or your tux and kind of sneak them out the door. You know, you can't really go George Clooney Oceans 11 on them.

So, one is there's going to be a large amount of physical verification and physical security protocols. The second is remote access. We're going to make sure through these deals and through the framework that uh you know nobody who's not supposed to have access especially from countries of concern can get access.

And so these three kind of the kind of the core pillars and here's why this event right and I think everybody in your audience who's like a technology person a technology brother or in the software world here's why they'll understand it what has history thought as a software industry the company with the biggest network effect the biggest ecosystem wins right we've all grown up with Microsoft how did Microsoft win the windows and office ecosystem think about this as the American AI ecosystem we are getting these resourcerich countries who are critical allies in very interesting geopolitical cases to basically adopt the American AI stack right up and down, right?

And so this means they are going to be part of our ecosystem for years and decades to come. And it essentially forms a shield from them ever adopting or using technology, you know, or you know or working closely with some people that we don't want to work them to work with.

So so in a way I kind of think of like a software ecosystem play where we have now have them you know tied to the American AI ecosystem. Yeah.

How much have you guys looked at the the story around 5G and Huawei and and and that market dynamic as a comp and something that we as a country want to avoid happening within AI.

Uh I would say that has come up probably every single day right like which is how do we basically this is in some ways look history rhymes but doesn't repeat is the cliche. So there are definitely some um um you know comparisons.

I would say the difference here is we are in the lead uh from the get-go because all the you know the the market leaders in Jensen and Lisa or the the best models are all based out of you know probably places from a few miles away from where you guys are sitting right now. So but we want to maintain that lead.

Uh I would say China is close behind. You know deepseek has was a wakeup call. Uh what Huawei is doing is a wakeup call. So we have a lead. We want to maintain that lead. And I think one of the best ways to do that is to get these countries who want to work with us. They're like, "Hey, we want to buy American.

We want to buy these American AI companies. Great. Let's have them buy our products, tie into our ecosystem, make sure we have rigorous security protocols in place that nobody else can ever get at these GPUs, and also make sure that they're funding the building of infrastructure. " So, our lead continues to stay there.

Yeah. Um, you mentioned the hyperscalers would be responsible for buildout in America. Um, do the Neoclouds have a chance to compete? I mean, I'm looking at semi analysis, the cluster max ratings. Corewave, Crusoe, Lambda, these are some top upandcomer companies. They're not hyperscalers.

Do they have a shot at getting a piece of these deals or is that kind of conversation? You know, if you listen to uh the Secretary of Commerce, Howard Lutnik, when he talked about this, the thing we care about the most is that they are American. Sure. Okay. That's the thing we care about the most.

Second is that they can you know basically sign on to our security protocols. So if someday if me or Sax somebody someday be like hey what's going on there there somebody who's an American company who's on the line.

So absolutely and I think you know um um so when I say hypers skills I don't just mean the you know the top two three companies I think if these are American companies who are willing to live with our protocols right like and they kind of get through that bar they should be open for business.

But look once we get this ecosystem here you are talking about a few of the most uh some of the most important allies that we have uh in the region or maybe u you know in that part of the world who are going to be forever using American AI right be it semiconductors or American models or maybe agents on top they're going to be bound to the American AI ecosystem.

Okay. Uh who are the other key partners in building this AI ecosystem?

Obviously, the Middle East is in focus this week and last week, but are similar deals coming around or have they already come around with the five eyes comp countries in Australia and Canada and the UK or are there other countries that are are a little bit more of like jump balls that we're maybe uh that we're going to see uh trips or conversations emerge over the next couple months?

It's a good question. So, first is so under the Biden regime, there was something called a diffusion rule. Um, and this is only, you know, those of you who been following this story closely, you probably know this, but essentially what it did is it sliced the world up into basically the halves and the have nots.

GPU rich and the GPU poor. Yeah, pretty much. And trust me, all the hnotss were not thrilled because some of them were our really strong allies and there was no way for a hnot to ever kind of become GPU rich.

And so we've toned that up um, you know, and we've done away with the diffusion rule and I think what you're seeing the last week is the first of a kind of these partnerships. But um stay tuned. There'll probably be more and if there is I should probably come back on the show to announce it here first uh as one does.

But you know I would say think of this as a template for how do we have a deal where American companies win. We get American infrastructure with the right security protocols in place. Who's the most AGI pill uh person you met uh in the Middle East? Chic Tanoon would be my my candidate.

Uh but anybody else that was surprising? Uh I would say I would say there's a lot of enthusiasm for AI across the board. Uh his highness uh Shik Taknun uh is really really up there uh and he has some very very interesting views on AI.

He I in some ways he's like almost like a technology founder uh in the way he thinks and operates but also his royal highness uh you know Mohammed Prince Crown Prince Mohammed bin Salman is also a huge believer in AI. All these folks are like look we want American AI. They're very optimistic. They try all the products.

They are personal users of all the products. They're paying $200 a month for subscriptions. They use it themselves. They use ask their team to use it. They're extremely bullish on using American AI.

Can you walk me through how countries are thinking about the the control over the foundation model layer versus the application layer? You can imagine a scenario where a country just hypothetically any country uh invests all this money trains their own model but their citizens are still going to chat.

com and using chat GBT and so the aggregation theory uh you know leads to American AI and uh which could be good uh but um but is there is there a similar push for uh for international countries to create their own application layer products. Yeah. Uh I would say the game is all about inference in my mind.

Um I I'm not sure it's an official administration view, but it's like my view and uh you know I think Satya in a recent earnings call had talked about like tokens generated as a metric. So I think this is all about building these token factories, right?

And these token factories are critical because they give you healthare, they give you education, they give you predictive traffic patterns, they help you figure out, hey, what is happening with the hydrocarbons that you're digging out of the ground and how are you processing it, which is obviously a very important part of the region.

So it's all about tokens at scale. That's what the key part are. I'm not really sure why they think of like training their own models.

There's definitely some interesting applications of how do we encode local values and local language in there but a lot of the conversations I had was about like inferencing at scale for all these applications so they can deliver to their citizens or to their key infrastructure companies for example oil or you know uh gas refining you know the benefits of AI so and there's also a lot of interest around agent applications is obviously very very early days but I saw some very you know interesting new prototypes as well but I would say inferencing to deliver education, health care, u you know, make the critical infrastructure on energy better.

That those are kind of like the lot of the key conversations we w up having and all of them are like look we want to work with you know all these large American AI companies or these startups. They often had deals orus we want to work with these guys.

Is it still uh the conversation is probably like 80 90% large language models little bit on image diffusion maybe a little bit on robotics. Is that kind of the the the general application because of where robotics is? It's exciting but we're not quite there where we're in the roll out yet. Yeah, that's probably fair.

I think a lot of it just LLMs and creating these token factories. There's definitely some image models. So, for example, as I was talking about when you think about these hydrocarbon refining, there's definitely some imagery involved.

Robotics uh uh look I'm not the expert uh you know I know you folks have had other people who are very very close to it.

there are some early interesting like you know startups and companies and definitely a lot of energy uh but you know uh but not I would not say that's kind of the meat and potatoes of what we're talking about the meat and potatoes is large language models inference tokens at scale and let's build applications that provide value okay um we sorry you go and then you go you go so uh I want to walk through like the actual mechanics of how these deals happen to the degree that you can uh we've been talking about how Jensen has been on this whirlwind tour where he's at a conference like every day, but usually today he announced a huge deal in France and then he was in Taiwan.

Yeah. And so usually when you see a CEO on the on the on the conference circuit, you're like, "What'd you get done this week? " But he gets so much done. Every time he shakes a hand, he shows up for an hour, gives a talk, shakes a hand, and gets a deal done.

So, are these deals happening before the events and then they're just being announced at the events or or is Jensen just has the gift of Gab and does a handshake deal and it's done? Always be. Well, uh my secret theory is Jensen has a few clones, right? Like and he just kind of rotates him around, right?

I don't know whether that's like leaking some like, you know, secret information, but I think like there's four humanoid. He's got humanoid. Yeah. Yeah, he definitely has a humanoid. But uh but I I I mean to to kind of contextualize it a little bit more for viewers um is Jensen's in a unique position.

I mean it's a multi- trillion dollar company. Uh is there anything that uh startups or early stage founders can learn? Is there is it crazy for a series B startup to hop on a plane, go out to Saudi Arabia and try and sell some product there?

Or is it or is that kind of putting the cart before the horse and you have to be like a power law winner before you start striking international deals? Not at all. I would say a few things. One is how does a government do it? Like look, I'm new to government. I've been here for like 3 four months.

My past world was uh we see where I spent a lot of my time tweeting, right? That's my primary function. Uh and uh but I would say there on the government side look there's a bunch of things which came together. We did away with the Biden uh diffusion rule. We've worked on this.

David Saxs, the commerce secretary, a huge set of people. I want to make clear an amazing team which worked on this under the guidance of President Trump and the vice president. We, you know, we obviously want to get this done before we go in there and then kind of the official signing in place.

But for startups, I I would say look, these countries are energetic. They want to do business. They're very interested in having companies out there and do deals with them. My suggestion would be that, you know, just get on a plane, right?

uh just get on a plane, go out there, make a trip and go see the situation on the ground. It may not be right for everybody, but I think you'll learn a lot if you just go see have a situation on the ground and they are very very willing to take meetings.

So my suggestion for everybody get on a plane, go see the world, go see how things are out there, then make a decision for yourself. So these first few months you've been focused on basically foreign AI policy. It feels like that's been like really in focus. How are how what's going on on the domestic side?

How much time are you spending with lawmakers uh you know locally uh kind of working on kind of frameworks for maybe long-term domestic AI policy and how to kind of wrestle with the the coming changes. So I would say this quarantine deal um it's probably not the bulk of our time.

It just obviously it's a big thing which happened last week.

So um you know we and obviously we spent a lot of time in the last couple of weeks working up to this uh trip for the president but I would say the bulk of our look our marching orders are very simple dominance of American AI and we are working towards what is going to be called the American AI action plan that is supposed to come out I'm going to say roughly in a month or a month and a half that is a clock we have to get this out in the first six months of the administration and that plan and you know I'm not going to like you know I don't want to sort of spoil it too much but it's probably going to have like two large two big components One is essentially how do we basically promote domestic AI and that means everything from um you know energy to open source to you know u how do we work with regulation whole sort of things in there uh which I don't want to get in detail on but like all the sort of the usual suspects you know I think we're going to be focused all and that is going to ladder up to the idea of like how do we make sure our companies at every level of the stack from semiconductor tools to semiconductors to GPUs to models to applications to agents to robotics to you know uh yeah up and down right uh and you know that we are winning and you know they have a clean runway um and then there is a protect side which basically makes sure that uh you know when you know we are protecting our secrets that we are making sure that our um you know when we build these things that we are playing defense against uh you know our competition um globally right and so if you look at this deal that we did in the Middle released last week that is a part of that because we now have uh with this deal we are now kind of got them hooked on American AI and as they build this data center imagine for example let's say you're one of these countries you're you know shipping over a bunch of GPUs you're building these data centers you're hooked on the American AI stack it's going to be really really hard for you to say four four five years down the road I'm going to work with another team you're probably going to stick with America so that's kind of locked in that's why it's really exciting but stay tuned I think there's a lot more we have uh coming on American AI yeah what what is the actual shape of that lock in because we've heard about it uh from the CUDA ecosystem that it's very difficult to switch to AMD and AMD and I mean this is just in the domestic market.

Uh AMD is not a drop-in replacement even though on a dollar per flop basis they're becoming more and more competitive. Um but at the same time you look at open router and uh software developers are swapping out LLMs every two seconds and you go on X and you see one one day it's cloud 3. 7 sonnet and then it's uh GPT 4.

5 is the hot model and developers seem to be comfortable swapping some of these things.

So can you tell me a little bit more about the shape of the infrastructure and and I mean to what degree is it even possible to lock in a particular part of the stack if we're just talking about energy or something or going in or or cooling or air conditioning.

I mean as we talk about the the massive AI factory there's there's a world where the components they have to be somewhat swappable at a certain point right. Yeah. Yeah. It's a it's a good point.

Um so I would say by the way Jensen did a a podcast or a newsletter with Ben Thompson of Strat today which kind of guard a little bit right um and I'm going to steal one of his analogies uh which is uh the AI stack. So you're absolutely right. But look, imagine you are a country or a company.

You put up made a large investment in America. You spend over a trillion dollars. Uh you have shipped over a lot of GPUs. You now optimize them to run either Nvidia or AMD. As long as an American company, you know, we'd love for you to work with us. Then you're building the models on stack on top, right?

And those models are going to be optimized for the hardware they're running on. And there's a million different details from networking infrastructure, power, uh how the racks are laid.

there's a million different details but they all work together right like whatever be your model of choice and we don't really care as again as long as you're uh American um and then on top those applications are also then going to be tuned for those models or the infrastructure you have so you kind of have a lot of this ecosystem that is going to come on top and now there are going to be developers you know either domestic or international building on top of that and they are going to be calling these American models for inferencing or they're going to be building on top of those American models for inferencing so so I think if you count all of that.

there is a little bit of a uh network effecting happening where say four or five years sure theoretically there could be an API which lets you switch over but practically if you're happy and you spend four years with all this investment with America it's going to be really hard to change away right and you know I don't I don't want to ever compare us to like Microsoft but think about like a working with a large ecosystem like Microsoft for 20 30 years you're you're using Windows you're using office using all the APIs you're getting good value probably quite challenging for you to switch over So I think you're going to see something similar.

Given the Deepseek moment, has uh has Meta become more of a national champion and seen more as like a nationally critical company? You go back uh five or 10 years, senators were asking Mark Zuckerberg how he makes money.

uh today you know they're they're like llama for the roll out hasn't been exactly smooth but at the same time you could imagine a lot of interest in Washington to have a state-of-the-art opensource American foundation model that um America can provide and say look yeah you can fine-tune it it's open source you can see the weights or open weights um we should actually spend time rehashing an acquisition that happened 12 years ago Yeah.

Yeah. But what is the mood around Llama Llama 4 and and what Meta is doing? Yeah. By the way, fun fact, uh Deep Seek I think was a big wakeup call. Uh fun fact is Deepseek was the day before I started my job.

So I got a call saying uh I think it happened like four or five days after the inauguration and I was just about to get started at some paperwork. I had a call saying like, "Hey, there's this Chinese model which has come up.

Can you come over to the White House on this wister badge because you know a lot of people just want to know what's happening.

" So I was kind of like thrown into the deep end you know we were really off to the races and I do think deepseekers was an very important moment because look um all respect there was some great technical work which happened in there KB caches MLAs how they're kind of working around their networking uh or their hardware limitations a lot of like just like really really good work done there are some also some questions about you know what was maybe taken or not taken but some really good technical work done but I do think two things kind of became uh very very clear the first is our lead is pretty small.

So, um, and again, you know, I don't want to just kind of go back to blaming the previous administration, but you know, be, you know, there was kind of a mood that America had this huge lead on AI models, but I think for a period of time, I could be wrong. R1 was maybe outside of 01 the only reasoning model out there.

Uh, but I don't think there was any other reasoning model out there and it's kind of pretty high up on all the the leaderboards for at least for a period of month or so. I could be wrong, but somewhere in those ballpark. So, one was it was clear that our lead was not big.

The second part was I think there was a meme that opensource was somehow giving away our secrets but it turns out that open source actually a great product that people want to use. They you know there's real value uh in terms of being able to build on it, modify it and so on and so forth.

I think those two are very very um interesting. One of the things about this job is we want American companies to win. So I tell people I want a American open-source model that just crushes, right? like uh and um and I I I want Llama to do well.

I think Sam talked about uh his model uh which is coming out I think soon or you know shortly. I want them to do well. I think you know a lot of us have friends or investors who are investing in open source model.

I want them all to do well but I would love for America to have an open source model which does really really well as opposed to one that has censorship and values from the PRC. So I'm game for any one of those and I think it's going to be an exciting time. That's incredible.

Well, there's a lot more to talk about, but we'll have to have you on again when you have more news. And for reference, Poly Market has will open AAI release an open source model in 2025 at an 85% chance. So, pretty likely it happens, but uh it's just been incredible. It's been incredible to see your journey.

Uh we met not feels like forever ago, 2021, but uh to see these pictures out of last week of you, our president, and uh his real it definitely a record scratch for East Frame. Yep, that's me. How did I get your mouth? That's a cool, you know, by by Thank you. That means a lot. And look, John, we've been through a lot.

I've known your story. So, congratulations. By the way, my favorite headline out of that whole thing was there was a headline, I think, in uh I think maybe Indian newspaper saying origin man powers or world leaders, right? Like something for us mugs. Mugs. He hype mugs. That's great. Well, congratulations on the deal.

Uh we'd love to have you back soon. We'll talk to you. Yeah, this is great. Great to catch up. I will. Congrats on all your success, guys. More later. Thank you. Talk to you soon. Bye. Next up, we got Josh Steinman from Galvanic coming into the studio to break down.

A uh very, very interesting story that I think went under the radar all about um how Chinese solar panels have been discovered to have um uh communication devices on