Former Biden White House official Peter Harrell: the Nvidia 15% revshare is unprecedented, legally dubious, and sets a dangerous precedent

You can start for free. Well, on that note, I have Peter in the reream waiting room. Let's Here we go. Welcome to the stream, Peter. How are you doing? You look fantastic. Great. I I was told I should wear something, you know, not just a black suit for you guys. So, there we go. You look fantastic. You look fantastic.

How are you doing? I'm great. I mean, uh, you know, the East Coast, it's been a little cooler than it was the couple of weeks ago. It's the '9s are coming back, but, uh, not yet. So, no complaints.

And you know I am for better actually really for worse not an Nvidia shareholder so I don't have to directly deal with their 15% paid to the government. Did did the stock pop today? What actually happened with Nvidia? It's it's it's a normal day. It's up like 1%. Okay. No nothing crazy. Not a huge move.

Um before we dive into all the news uh would love would love to to just you know give us the highlights from the bio. you've done you've done quite a lot, but I think it'd be helpful context before we get into a discussion. Yeah. No, look, it's it's great for uh to be with you guys. Thanks for having me on.

I currently uh am a non-resident fellow at the Carnegie Endowment for International Peace, which is a think tank in DC. I also am an attorney and have a trade uh uh law practice sort of focused on trade and sanctions and export controls uh kinds of issues.

I was in the Biden administration at the White House where I dealt with the global economics file including a bunch of the export control issues and been in private practice prior to that.

Was the State Department for a while and when I was much younger I was a reporter covering campaigns and elections like uh many many years ago. That's super cool. That's awesome. Kind of been around. Well, breakdown.

So, so, so before we get into the Nvidia AMD stuff and the news from yesterday, take us through kind of your experience of this year so far because there's been a lot of different stories. Go back further. Chris Miller writes chip war. Everyone wakes up. Is that what happened?

Like when when did DC wake up to uh to maybe we need to think more critically about the flow of semiconductors and AI technology back and forth between the US and China? You know, that's a that's a great question. And I give Chris uh a lot of credit.

I also give a couple of guys who were at um at Georgetown University which has a center for science and emerging technology. And a couple of those guys like in 2018 2019 started really tracking what kinds of high-end chips were the Chinese buying and like what was going on.

You know they'd hired some Chinese language researchers and were really looking at what the Chinese were doing. And so you had these kind of scholarly researchers looking at this and saying hey there might be a problem here. Then you had Chris popularize it with chip war.

Uh and you also had back in 21 you know that was also the time not on the leading edge stuff but on the the mature node stuff semiconductors when we had that semiconductor supply chain crunch. So semiconductors were very much in the the news and kind of policy makers were focused on it. This was during that cars.

We couldn't make cars, right? During co we couldn't make cars and we couldn't make internet connected washing machines. Yeah, it was really bad for the IoT microwaves. That was particularly rough.

Well, there were like three months where you know I was working at the White House and you know Ford would call me up or GM and be like we can't get semiconductors.

And then it would be my job to like phone, you know, governments in Malaysia and Thailand and being like, could you, you know, get some of these shipped to your American customers, um, please, you know, before the Germans and the French. So, uh, you know, what was what was actually driving that?

Was that like was that like like you needed because I mean fabs are clean rooms. Like I imagine that you could still go into the semiconductor facility during COVID because you're not going to spread it because you're not spreading anything. You're not spreading dirt. It's a clean room.

But uh there's but there were there was like this immense amount of pull forward during COVID stimulus checks, people buying more cars than they thought they were expected, but then also there were shipping delays and all sorts of backups at the port.

Like how do you think about what was actually the root cause of the initial semiconductor crunch in what was this 20201 or 2020 and 2021? 2021 early 2022. Yeah, I it was actually I think the the combination you just highlighted. So you had demand going through the roof, right?

Because there was this huge pull forward for everything from uh you know laptops to uh you know everyone's home so they're buying home appliances. There's huge pull forward in demand. Um and you didn't actually see a fall in production of the mature node chips in 21 but because of just like bad luck.

There were uh problems in Texas. There were problems in Japan. Like confluence of bad luck took a little bit of capacity offline and the capacity coming offline which which would have been manageable kind of in an ordinary market condition combined with this massive pull forward just meant a mad scramble.

And then you get into the, you know, psychology of shortages where uh everybody starts hoarding, right? because people are sort of like, well, if I'm not going to be able to get them, I better hoard.

So, it was sort of like the semiconductor version of the toilet paper shortage, you know, we'd had a few months prior where it's like, you know, everyone's hoarding everything they get their hands on. Got to start Gwagons, Ford Raptors. Anything with a semiconductor in it, give it to me. I need it.

People were certainly doing that for PS5s, too. I mean, I remember even video games, like all all different sectors people were were pulling and they were pl pay paying like what $50,000 over sticker for like a new Hummer EV and stuff like it was getting crazy out there. Wild times. So, yeah, please continue. Yeah.

So, that was where we were on the uh on the on the shortage. Um but looking back I mean so if I that that got us the shortage Chris Miller's book some scholarly research and just kind of heating up competition with China you know in sort of late 21 into 22 got the then Biden administration but also folks on Congress.

It wasn't just sort of a Democrat Biden thing. folks in Congress uh focused on the strategic competition part of this and what is China uh where are they going with AI? What are the choke points that we might have to slow their development of AI?

And that really gave uh rise in October 2022 to the first big US semiconductor export controls where the commerce department essentially cut off uh most of the highestend semiconductor sales uh to China as well as various uh manufacturing equipment and inputs to try to limit uh China's ability to develop indigenous uh production uh capabilities.

And then that that initial export control regime from October 22 got expanded a couple of times. Um most recently in April of this year when the Trump administration paused uh sales of the H20 chips of the Nvidia H20 chips uh to China. What was driving the initial chip bands?

Was it more we I we have identified that large language models these huge training runs are important and once you marshall the big data center the the the 100 gigawatt data center you're good and you have a cutting edge model or is it more like look if we send a certain amount of equipment it will be reverse engineered and copied and then they will be able to catch up on the actual supply chain side and and we'll not just competition from purchasing from Nvidia, TSMC, and ASML, but we'll actually see Huawei, Smick, and SME act as a alternative supply chain that will be an entirely internal economy that will keep them on the frontier potentially forever.

Yeah. So, it was in the first instance an effort to limit compute to China.

So there was really a pretty systematic okay if we're in this era of strategic competition with China and if we think AI is going to be an absolutely uh transformative technology you know something with just a whole range of national security economic impacts you know day-to-day life impacts how do we slow down China's development and you start thinking about well you know they got a bunch of good engineering talent over there it's not like you can prevent them from having talented engineers years, right?

That's just sort of life. Um, you know, the the the model side of this, controlling the models is very very hard. Uh, because, you know, it's software, it's code, a lot of some of the models are open source, just very hard to control.

But what you could do mechanically is control their access, China's access to compute because their chips just are, you know, 3 to 5 years behind ours. And so if you choke off the chips, you slow them down by limiting their compute uh their access to compute for training. So it's just kind of how do you trip them?

This is the choke point. Let's squeeze on the choke point. Um and then you know they iterate, right? They start developing lower compute models. Like you see this kind of back and forth, but that's what happens with export controls generally like you know your adversary is going to adapt.

You're going to have to adapt and and that's kind of what we've seen play out. And how much were you tracking the GPU black market?

You know, in in once you had export controls, there's certain countries start popping up saying, "Well, we'd actually like$5 billion of uh whatever GPUs you got and uh don't worry about what we're going to use them for. We'll figure it out. " But not your problem. Yeah.

So, that's a known problem you're going to get into, right? And you've seen that in other instances of export controls and sanctions. And so, you know, commerce department does have a unit that, you know, is supposed to be looking at the trade flows. Where is that uh where is that happening? Where are you seeing that?

Um, you know, I I it is a bit of a game of whack-a-ole. It's clear there has been a certain amount of diversion uh of prohibited chips going to China.

I also think you know just if you look at kind of what Nvidia talks about the sales potential clearly the export controls have reduced China's ability to get to the chip so that they haven't taken it to zero. Mhm.

How do you think about the evolution of the US government's thinking on how critical or like the the role AI plays in nearpeer competition because there's kind of two frames of mind that I see in Silicon Valley. One is AI is super intelligent god. It's nuclear weapons.

It's it's one button you push and the AI goes off and just wins the battle for you. So, it's it's critical to have it uh and it's critical to keep it out of the hands of your opponent. Uh the other is that AI is more like Microsoft Excel and it's certainly useful in the military context.

You want to know how many like electricity. Yeah. Yeah. You you want to know how many, you know, shells you have in in inventory. You want to know how many how many soldiers are in this particular part of the battlefield.

and you want to be able to track things, but it it and and it might be advantageous to keep a technology like Microsoft Excel out of your opponent's hands, but it by itself will not win a battle decisively.

And it feels like long ago, pre-Chat GPT, pre kind of AI boom, people were definitely seeing artificial intelligence as, well, yeah, we use it to serve ads and it and it helps with spellch checkck and it's in Tik Tok and and it recommends, you know, what video you watch next and and then it became, okay, it's super intelligent god and it's nuclear weapons.

But now we're kind of maybe in an era where we're where we're moving back to it's more of an incremental technology. But I'm wondering how that thinking evolved in DC.

Uh so so I think the evolution in that debate is an important part of understanding why the Biden administration took a pretty tough line on export controls because I would say many of the relevant policy makers in the Biden administration were actually quite bullish on AI.

you know, did buy into the thesis that, you know, I'm not, you know, whether it's 2027 or 2029 or 2030. You know, we could debate the years, but they bought into the thesis that in a fairly near period of time, you were going to see AGI kind of reach that, you know, escape velocity where it's iterating on itself. Sorry.

Where you're going to see AI reaching that escape velocity where it's iterating on itself reaching AGI, you know, sometime in not too many years.

And if you believe that, if you think in a couple of years we're going to see that, you know, escape velocity to AGI, then you really want to slow the Chinese down as much as possible, right?

because you think this is going to be this enormously transformational technology and you know 2 years or even 18 months or you know maybe even 12 months of lead is worth an enormous amount from a policy level. So it was actually because of a very bullish view on the future of AI. Yeah.

That you saw the Biden administration really take kind of a pretty tough stance on slowing down China.

And so and so reading into that does that mean that you that the admin might not not if they're willing to say yeah we'll we'll we'll export H20 we don't believe in a fast takeoff scenario this is electricity it's valuable but it's going to be vended in everywhere because if you still believe in a fast takeoff scenario you would say it's not worth the incremental sure Nvidia you know shareholders are going to suffer a little bit if you and export to China, but it's not worth the the however many billions of incremental revenue that the government's going to generate on a 15% take rate.

Yeah. So, I think the Trump administration is probably doing a couple of things here. I mean, first, there are lots of different views in the Trump administration. That's no news to everybody. You see some folks who uh I think obviously would prefer the Trump administration take a harder line on export controls.

You also see, you know, some folks uh who, you know, basically have a view of, well, we shouldn't sell China our most advanced chip. The H20s are now a couple of years old. This is not really uh most advanced. You might as well um get the Nvidia revenue and a US government take here.

So, I think you have diverse views in the Trump administration.

I think it's less guided by a you know sort of slower view of the uptake and more my view the Trump administration is looking at this is like this is not state-of-the-art and then also Trump is a dealmaker like it it's hard to see any other uh politician Republican or Democrat coming up with this but you know Trump himself he's always talked about getting you know a cut of the deal like so it's it's in a way this is a you know sweet generous unique to Donald Trump concept that I'm going to let you do this, but uh by the way, we get 15%.

Well, and he said this morning that he he would potentially consider exporting black wells, but at a higher rate, which uh now that I mean, this is um you know, it's interesting like the H20s I do think are kind of debatable. I mean, I am in favor of controlling the H20s, but I see the arguments on both side.

It is a dated chip now. you know, I'm not D, but you know, it's not a cutting edge chip. Um, it is very hard for me to see how charging a percentage, you know, revshare uh addresses any national security concern, right? I mean, like if you have some national security concern, the revshare doesn't address it.

Maybe, you know, forcing a downgrade of the chip or trying to limit who it gets sold to, which is very hard. Like revshare definitely doesn't address whatever your national security concern is. Is there is there any possible steel man for that?

You take the 15% revenue share and you pour that into some sort of national security AI defense fund that that protects you from the super intelligence. I don't know.

It's very very hard to I buy the I buy the logic of like I'm going to make I I'm fine to export the H20 because Jensen's convinced me that it's okay but I still want my cut is kind of is that's how I read it. Like that's kind of how I read it. Like Jensen convinced him, you know, it's an old enough chip.

We have enough of a lead like you might as well export it. And then here's a guy who made his living in New York real estate. Like you know you want your commission, right? I mean, that's just I think that's kind of what's going on.

When you when you saw this, I maybe you had a a a feeling that this was coming down the pipeline, but did you had posted on LinkedIn, I think it was earlier today, trying to look for some type of precedent for USG revshare on in and didn't sound like you found one.

Yeah, I can't find any precedent for anything like this.

You know there have been some instances uh for example if you are some of our defense contractors you know the Boeings of the world if they want to export fighter aircraft they have to pay like a couple thousand fee for the government to review the license like it just clearly just sort of the fee to cover the staff time of like reviewing the license.

Uh but it is literally a couple thousand dollars. It's not you know revshare of a multi-billion dollar market. uh here. There's no precedent for anything like this. And and what are you hoping? Uh I'm thinking about where it goes. Like are we going to be selling F-35s?

I mean, this is like it opens up a whole kind of you know, what are we going to sell next on a revshare model? That is kind of crazy. Let's get some submarines over there. Hey, we got capacity. Let's make some Let's make a bunch of submarines. Yeah.

And and then the concern is you highlighted you said the revshare creates perverse budgetary incentives for US agencies to issue licenses that undermine national security because they want the cash.

It also sets a precedent that the US might try to generally tax exports, something the USG has long avoided for legal reasons and because we have historically wanted to promote exports. Yeah. So, so I do think this this does create a perverse incentive.

I mean, you know, in in theory, the reason you control these things is because you think there's a real national security risk. And to your point, I think in some sense, uh, Trump Jensen convinced Trump like there's not really a national security risk here.

But think about this going forward, you've created this really perverse incentive where uh, you know, company wants to export something to China, maybe Russia, like who knows who's next on this. uh that poses a national security risk.

Uh but hey, if you don't, you know, if you if you authorize the export, like, you know, you get 15% uh and you can use that to pay for whatever domestic programs you want.

I just think it creates a really weird incentive for a person at the commerce departments whose job is to consider the national security uh impacts of this. Is there is there a like an optimistic scenario where like American industry or western industry is strengthened by this move and I'm I'm trying to walk through it.

So, uh, we we've talked to a couple people who work in the solar industry and they've said that China is incredibly dominant in solar and they flood the market with very cheap solar panels and that's effectively kept American solar panel producers out of the game.

And so, America doesn't really have a strong uh solar panel production industry. And so maybe the bull case here for American dominance in AI is something like you get so many NVIDIA chips over there that Huawei, Smick, and SME can't really stay in the game. They can't really be competitive. They fall behind even more.

And then uh and then if if AI plays out in in such a way that you don't just need one massive data center, you need to be constantly upgrading your data center. you need to constantly be buying more and more Nvidia GPUs that if you don't have that flow, if that flow can be turned off, then you are always vulnerable.

So, so I actually think the policy question of whether to sell the H20 to China is a hard one for exactly uh the reason you lay out which is that there is a you know there is an argument and I think Jensen makes it in good faith as well as for his own commercial interests of we want China dependent on our chips and if they're not dependent on our chips they're just going to be you know building uh you know SMIC and and and Huawei chips and that's going to ultimately more inclined to more inclined to take Taiwan.

Yeah. Right. Um here's the question I got back though. Like charging an extra 15% on the American chips doesn't actually advance that dependence argument, right? It just um creates a cost differential that makes the American chips more expensive relative to the Chinese chips.

presumably if you really wanted to keep them dependent on American chips, you'd, you know, want to sell them as cheaply as the market would bear. So like the 15% even in the like bull scenario, and I agree.

I think it's an important debate of the the the kind of bull scenario you laid out, but 15% still doesn't make any sense. Yeah. Yeah.

And you look at other other categories like smartphones or electric vehicles, there's there's no precedent for China saying, "We'll we'll buy your versions of these products and we'll even manufacture them for you, but don't worry, we're not going to develop our own.

" You know, you look at this like of course they they they clearly care about having local industrial capacity and even the end products and brands and those are for categories that aren't around that don't have nearly as much national security.

And if you talk to like the you know the the CEO of GoPro when they were trying to compete with DJI and drones, he's not he's not going to tell you a story about DJI getting attacks on exports.

He's going to talk about, well, DJI was subsidized and had and had and had governmentbacked loans and had extra money poured in from the government that allowed them to flood the American market with drones and now we don't have a really strong drone industry here.

And so, yeah, the like we're kind of like there's two different arguments. There is some inongruity there. Like they they both can't be true. Yeah. So, the steel the steel man is is we were going to allow this. Yeah. So why not get a Uncle Sam, a few Billy out of the deal?

But the concern is that it would set a precedent that that could be, you know, destructive. So now we really got to talk Jensen's book and say we should be subsidizing exports of AC monies to China.

We should be America should be paying an extra 15% for every GVU sends to China to really screw them over on this domestic supply chain side. This is this is seven dimens dimensional chess in geopolitical chip negotiations. We we've solved it. We've solved it. Exactly. There you go.

Anyway, what what else is top of mind for you right now? Yeah. What Yeah. What's on the horizon for the rest of the year? How do you see things breaking? Yeah. I mean, so you got th this is a big development. We'll have to see like is this a precedent?

Does this become a precedent for you know HBM and other parts of the semiconductor ecosystem? Like we'll have to we'll have to see.

Uh the other one that obviously those of us here are watching uh not on the export side but on the import side Trump has said and uh commerce secretary Lutnik has said their tariff regime for imports of semiconductors is going to hit fairly soon.

Obviously we saw Tim Cook get a um a deal or announcement out of the White House last week where if you know companies are promising they're investing in America they won't pay the tariffs. Uh, and we'll have to see how that all plays out actually in practice.

But like how these potential semiconductor tariffs again on the import side plus whatever this exemption for companies that are investing here, how that actually works in practice is going to impact US semiconductor companies, companies that use a bunch of semiconductors.

It's going to be a big noticeable economic uh issue for the industry uh over the next couple of months and certainly into next year. Well, we'll have to have you back on if some big news breaks. We really appreciate you taking the time. Great. Super. Thanks for having me on. Thanks for hopping on.

We'll talk to you soon. All right. Cheers. Bye. Take care. Let me tell you that Finn. AI, the number one AI agent for customer service. They're number one in performance benchmarks, number one in competitive bakeoffs, number one ranking on G2. Go