Chip War author Chris Miller: China is 5–6 years behind TSMC, but shell company smuggling reached millions of chips

How you doing? What's happening? Good, thanks. How are you? Thank you so much for joining. It's great to have you. Uh, great to have you. Um, I would love I mean, we were just reading an article from the journal um about the uh the China versus the US AI race.

There's a whole bunch of modern details, but I'd love to kick it off with uh a little bit on your background and how you initially came became interested in the idea of a chip war and kind of how things emerge from there.

Yeah, I I got interested in the idea of a chipboard because I realized we've been living through multiple iterations of one since the first chips were invented in the late 1950s that you couldn't really separate the development of military power or country's strategic position on the world stage without understanding the ability to access and deploy computing power.

And so we're seeing a new version of that right now with AI, but it's it's actually replicating what we saw in the cold war competition between the US and the Soviet Union or the economic competition between the US and Japan in the 1980s.

And so then as now, uh there's this deep link between access to compute and your strategic position. Yeah. Going back in the history, um we we we were digging into some of the the efforts that the Chinese government had uh Yeah. How many how many fiveyear plans have they have they done?

They they keep stacking these fiveyear plans and it seems to be working for decades just to stay on the lagging edge.

Um what did you learn about that and what what can you tell us about like that narrative as you know maybe the chip war is heating up right now but China has been aware of this for decades and has been investing against it. Yeah. No, they they've certainly been investing uh for a a long time.

I think really the last decade or so it's kicked up a a real notch. um both because China's solved a lot of its other technological problems.

It's become self-sufficient in many other spheres of of manufacturing and making chips is the one of the really few areas and the largest by far where China still has to import manufactured products from abroad such that today China's largest import more even than it spends on oil is importing semiconductors from Taiwan.

Wow. And uh so this is something that they wanted to break free from more on semis than oil. That is wild. That's insane. Um, when you talk about that, are is is that driven heavily by uh stuff that's not sanctioned on the lagging edge, the you know the 15 16 nanometer like the bigger chips?

Um, or or are we talking about like smuggling numbers of 3nmter chips that have been repurposed through shell companies or something like that? It's it's a mix, but it's it's mostly things that are are legal to sell to China.

And you can sell both the the lagging edge like things in your dishwasher, but also a a smartphone chip at 3 nmters that you can still sell to China. And there's a large volume of those for PCs and smartphones and autos that are still being sold. Yeah. Yeah.

So, every every Huawei phone has a chip and those are not not sanctioned. We're just talking about the large accelerators, the G the big GPUs that are really under Exactly. under under lock and key. Um That's right. Interesting.

Um, then take me through kind of how you processed the initial roll out of the chip ban during the Biden era. What uh how did that play out for you? Were I assume people were calling you? I don't know. I I want to know your side of that of that era of history here. Yeah.

Well, I would actually rewind the clock back to 2018 and 19. That's that's when um ASML, the Dutch company that produces the lithography tools you need to produce cutting edge ships, was first thinking about selling a cutting edge tool to a Chinese customer.

And then the the Trump administration, then the first administration leaned very heavily on the Dutch and said, "Don't let this go through uh because if you do, China will have everything it needs to catch up in in shipmaking. " And so I think that was actually the the key turning point.

And then the 2022 restrictions that not only tightened up the rules around chipmaking equipment but also banned the GPUs that you can make with these uh tools uh in some ways naturally followed from that first 2019 move.

What was different in 2022 was that there was a real I think understanding and it came before Cat GPT that AI was going to be a big deal and that access to GPUs was going to be a critical input into building bigger and better AI systems. Yeah. Uh how how did the chip shortage uh fit into all of that during COVID?

I mean, everyone experienced uh you know, the the chip shortages from like auto prices going up, used cars, there were so many market distortions. Um h how did you process that point in time? And then I want to step forward even further. Yeah. So I think the chip shortage didn't really impact tech companies.

impacted as you say autos and dishwashers and microwaves. But it brought together a political coalition across the west that uh included the auto companies and the tech companies and and medical device companies all of whom realized chips were very very important.

We needed reliable supply chains and we needed to stay ahead technologically in uh in this and that's why whether in the US or in Japan or in Europe you had chips acts being passed to reinvigorate domestic manufacturing to try to provide some more resilience for these supply chains.

It would have been unthinkable 5 years earlier but because of this coalition it was possible. Yeah. Is it is it a misnomer to call China's strategy around semiconductor and domestic production of semiconductors and all of their semiconductor strategies communist.

It feels like they in some ways have an extremely capitalist system with not picking winners and putting out, you know, huge government incentives. is they're they're shifting the market, but from what we've heard a lot of times there's just a huge pool of money and there's vicious competition to go after it.

So, how how would you actually characterize or like what words are we using? How are we describing the Chinese political system now with specifically with regard to not just how they elect people, but specifically their chip policies? Yeah.

You know, I think you got a a strange uh mix of of policies being deployed by the Chinese and and in some ways that's why it's worked relatively well by a lot of metrics.

You you have the benefit of the market weeding out uh firms that can't compete, but you've also got the benefit of the government devoting long-term patient capital into its priority areas, which pulls in resources, pulls in entrepreneurs uh to these segments.

I think the risk for China is that unlike other sectors that they've done well like solar or EVs where they've had a thousand flowers bloom and you figure out which ones are capable which ones are not in the chip industry you don't have nearly as much of that and in fact the key players today like the most uh significant most advanced maker of processor chips today was also the most significant advanced in China 10 years ago so there's much more that would be SMIC which produces chips for Huawei Yeah.

Got it. Yeah. And so you don't have this thousand flowers bloom dynamic in the chip industry just because it's so capital intensive uh and the number of people with the knowhow is fairly limited uh that you got a small number of flowers blooming and all these flowers now have state ownership. Yeah.

Can you So it's a different dynamic. Uh it feels like they're building like a like a mirror stack. Uh can you walk us through uh the equivalent to Nvidia TSMC ASML over in China and then maybe give me a little color. Yeah. And on on the SKH side as well, are they doing stuff there?

And then I'd love to know um kind of uh if you can give me like a qualitative grade of like how serious these these competitors are. Huawei seems super advanced. It seems like Smick and SME maybe not, but I'll let you explain it. Yeah. Yes. I think you're right.

They are trying to build out the entire supply chain on on the memory side. You have YMTC, which is their leading producer of NAND memory. CXMT, leading producer of the DRAM memory, including uh eventually the DRM memory that will go into uh be paired with GPUs, the high bandwidth uh memory.

I think on the memory side, China's done really well. Um you probably wouldn't say it's at the cutting edge in terms of its ability to produce that volume, but in terms of the kind of pure technological capability, they've if they haven't cut up, they've gotten very close.

Um, yeah, I think for the the the processor chip uh side, it's been more complicated. Huawei is very capable at chip design. It's done it for smartphones at the cutting edge. Now it's, you know, catching up very rapidly in terms of GPU design.

Hard to exactly uh benchmark visav the Nvidia, but Huawi is very capable in chip design. The the challenge I've had is that SMIC um is right now about five years behind TSMC, five or six years in terms of its ability to produce processor chips at quality and at scale.

Uh and so that has been a a real limiting factor and the reasons behind is because there are no real Chinese equivalents today of companies like ASML or Applied Materials or Tokyo Electron that make the tools that make chips.

There are companies that would like to compete with them that can compete with some of the really simpler lower-end tools, but when it comes to the advanced tools, the Chinese firms just aren't there yet. And so, Smick needs to buy how far behind hard to say.

Um, I mean, I think certainly uh certainly 10 years behind today. They're now investing a ton. They're reverse engineering the usual playbook. Um but if you look at the the leading edge fabs that Smick operates, it's still mostly western tools inside.

And every single year uh for the past, uh decade or so, China's been one of the largest, if not the largest importers of chipmaking equipment in the world.

Last year, China accounted for ballpark half the revenue of all of the world's key chipm equipment vendors, which is a pretty good data point for uh where those firms are relative to their Chinese competitors.

What about on the what about the what about the scale of the shell company activity buying from from TSMC and and other players? Did have you tracked that closely? Yeah. Well, certainly a big shell company was discovered and then uh we hope shut down uh last year. Uh but the scale was vast uh uh truly enormous.

Millions several million chips uh procured from TSMC and then it appears diverted to Huawei. Um which is you know at a scale that I am frankly surprised by. Everyone knew there would be smuggling in the scale of dozens or hundreds or even thousands.

Uh but smuggling at the scale of millions uh really calls into the question the ability of the US government to properly enforce these rules. And you the only thing worse than a rule um that is is overly broad is a rule that's overly broad, not not seriously enforced. You get the worst of of of both worlds.

And so I I I think the US government has a lot of thinking of its own to do. And how do you actually make these rules uh stick in the real world? And thus far there's it's been a real struggle.

Are there any underexplored uh areas in data center buildout that where people aren't really tracking the competitive dynamic between China and America? I was listening to Dylan Patel talk about how with Huawei's cloud matrix 384.

They were able to do I think fiber optic connections between the chips and basically do NVL link at a much larger scale and Nvidia had tried to do that and maybe not been able not been as successful and it felt like you know Huawei is slightly behind or seriously behind in a number of categories but that might be one place that they're actually ahead.

Have you tracked that at all? Is there is there anything that uh people should be digging into there?

I feel like the the narrative was like everyone was getting up to speed on Nvidia and then people learned about TSMC and then they learned about ASML and then they learned about SKH Highex and I feel like we're all about to learn about a networking provider soon, but what what what's been your uh what's been your take on that industry?

Yeah. Yeah. No, I think you're you're right to focus on networking as a as a key area of rapid technological advance. You know, it's hard to benchmark who's how far ahead. um unlike in Foundry where you've got like a really clear way to measure TSMC versus SMIC um there's not such an obvious kind of benchmark.

I I think the the underexplored area in in the Chinese ecosystem is actually not at the chip level but the cloud level which is we know that China's smuggled in a large number of chips. We know that there are uh a meaningful volume of chips that Smick has produced domestically for Huawei.

Uh but it appears that a lot of these chips are in data centers that have low utilization and anecdotally all the anecdotes we hear about um utilization of AI data centers in China suggests a lot of real challenges in terms of getting utilization partly because there's some stateowned telecoms that are just mismanaging their cloud businesses.

Um but it's I think surprising that Chinese firms like DeepSeek talk about compute limitations given the number of chips we know that are actually in China. And so if you're, you know, if you're starting with a thesis that China is good at centralization, that doesn't appear to be what we're seeing here.

We appear to be seeing an inefficient um decentralization perhaps driven by some of the pathologies of having stateowned firms run data centers. Do you have any uh insight into what's happening uh that benefits China but happens outside of China?

I've heard reports that uh folks are taking uh training data on hard drives and flying it to another country, doing a training run, saving the weights to hard drives and then flying back.

Uh you know, you could imagine, you know, Deep Seek V4 or something being trained by, you know, a series of shell companies and it's just an American guy with a big AWS account, right?

Um, but at a certain point I would hope that Amazon says, "Wait, who are you and why do you want to spend $100 million this month on on cloud? " Like we we like billing everyone a few thousand dollars, but this is going to set off some alarm bells. Um, what are the risks there?

Uh, are is the government doing anything about that? What what is your take on this idea of of just, you know, hopping on to other clouds and then routing the data backwards? Yeah, I think this is an emerging area of focus for uh the government.

I think you're right that it's an area where the US government doesn't have a whole lot of visibility and I think cloud computing companies haven't they've got know your customer requirements etc but not nearly as as much investment as say a financial institution spends on making sure they're not transferring money to a sanctioned Iranian entity for example.

I wouldn't be surprised if over the next couple years we have a lot more buildout of this type of regulation precisely to deal with uh what you're talking about and I think you know a large training run is large uh and so I think we shouldn't be surprised to see uh heightened know your customer requirements as you get larger and larger in terms of the scale of compute uh that you're you're using.

The other nuance is that certainly US cloud providers are are are key here, but there are also cloud providers from other countries in other countries that are playing a role.

We read a lot about Malaysia the last couple of months as one of the key buyers of GPUs and uh a variety of Malaysian companies that are maybe working with Chinese companies um operating there. I think that's also an area where we might see uh more regulation.

There's been proposals in uh out of the House of Representatives for example to say we want more GPUs going to US cloud providers and fewer going to non- US cloud providers precisely for that type of of region.

America first for AI is sort of the um the way to think about that and and you know there are different parts of the government that would find that pretty appealing. Can you uh I mean you mentioned Malaysia.

Can you take me on a little bit of a tour geopolitically of of what's going on in some of the other countries that might might be on the horizon for uh AI relevance?

Uh I'm thinking about, you know, we've obviously everyone knows the US and China and then Taiwan's role and then Malaysia pops up and then France has has Mistral now and there's deals being done in the Middle East.

Um, I was always surprised that Russia was not on the conversation at all because they have a very fa like long lineage of incredible math prodigies and it feels like you you would I mean uh there are a number of AI scientists in America who come from Russia and so I was I'm surprised that we're not seeing oh Russia's getting in the supercomputer game or trying to build out some massive data center.

They have been busy. Uh and so what are some of the other countries? Can you take me on a tour of maybe maybe let's just start with like what's going on in the Middle East? How are you thinking about that? Is is it just like a jump ball bidding war between are they going with the west or the east?

Uh and then we can move on to some of the other regions in the world that might be relevant or not. Yeah. No, I think the Middle East uh has a real advantage in terms of the amount of capital they can deploy, which matters if you're trying to build big compute clusters.

And I think both for Saudi and especially for the UAE, there's a desire to build their next generation of industries if and when oil becomes less less uh less reliable for them. Mhm. And and that that's driven this investment in terms of building infrastructure.

And we saw after Trump visited uh a couple weeks ago now big deals announced especially in the UAE, you know, 500,000 GPUs a year uh to the UAE is a large uh number. Um, I think the question for the UAE is there's no doubt they can build data centers, no doubt they can afford the GPUs.

Can they build businesses on top of that? Um, and I think all these countries want to do more than just host Azure AWS data centers. They want to own their own businesses and and build on top of that. And that's a harder thing to do than than simply building data centers. I think the UAE has big ambitions in particular.

Um, but it's it's easier to have ambitions than it is to realize them. And it'll be a challenge. What's uh what's your take on TSMC's traction in Arizona? Nvidia's had some announcements as well. How how bullish are you on on the progress they're making?

So, I think the good news is that TSMC's publicly said uh their manufacturing yields for the share chips they produce that actually work are just as good in Arizona as they are in Taiwan. That's a a really powerful data point. I think the bad news is that on a unit cost, they're still a lot more expensive.

than in Taiwan. And that cost differential will decline over time, but it's it's a real differential and it's not going to go to zero. And so the key question, I think, is to what extent TSMC's customers are going to keep pushing TSMC to build more in the US versus be comfortable with um production in Taiwan.

I think the trend has been that they're pushing TSMC to do more in the US. Uh and certainly TSMC looking at the tariff dynamics has has registered that. They've announced plans to build six fabs beyond the the one operational one under construction right now. Unclear exactly over what time horizon that happens.

Um but so long as you've got these kind of tariff and geopolitical issues front and center, I think TSMC customers will be encouraging them to to pursue this diversification. Yeah. Who who would eat that cost? Because Nvidia has famously very high margins.

All the hyperscalers are thinking about doing in-house chip design now. Uh, I'd love to know your take on on Microsoft strategy, the TPU at Google, uh, anything else that's going on, uh, Tranium at Amazon for example.

Um, because you could see those you could see, yes, TMC TSMC in Arizona is more expensive, but that cost is kind of mitigated by being able to route around Nvidia's margin. But I don't know if that's a reasonable thesis. Yeah.

And I I think my assumption would be that the the cost is shared by across the supply chain and actually if you look at the supply chain you've got a a number of high margin companies involved. TSMC uh has a pretty healthy margin. Its key customers do as well.

And if you you know take an iPhone as an example, you know, ballpark the the share of TSM the TSMC content in an iPhone $50, let's say. if if that goes up by 30% in the context of a $1,000 iPhone, it's not dramatically going to change the economics. Uh same for um same for an Nvidia server.

Uh so I think you're right to say there's there's plenty of places where you could you could absorb uh some of that cost. What share of it goes to TSMC versus customers, I'm sure, is a a constant source of negotiation. Yeah. Yeah.

So, there's there's kind of like an interesting supply chain here in the actual delivery of AI applications. And I can't help but wonder if some of the uh some of the international players outside of America and China are maybe targeting the wrong thing.

It seems extremely valuable to be able to uh fab chips in your country and own your supply chain and be able to produce a consistent flow. And then it also seems incredibly valuable to have the front end to AI essentially like a chat GPT national champion that users go to.

But the model itself and the data center seems maybe less important there. But it feels like all of the different countries that we're talking about whether it's France or Saudi Arabia, they seem obsessed with owning a data center. But then if the local population is going to chat.

com uh instead of lech for mistral like has France really won if they just have a model and maybe may maybe they're thinking about in terms of like deploying within the government in the same sense that like the DoD is deploying llama right now but how do you think they're how do you think these international uh countries are thinking about the different places that they can try and plug into the overall delivery of artificial intelligence.

Yeah, I think that's the right question to to ask and I think you're right to suggest that just having a data center is not a guarantee of making money off of AI. Yeah.

Um I mean it it seems highly likely to be the case that the the application layer and the the the operating system uh for AI, however that develops, will be places where a lot of money is made.

uh and again not clear that that if you build a data center in in your country you're going to guarantee that you have that at all. I think when governments think about this there's probably a degree to which they're not asking the question you're asking and they ought to.

U there's also I think a degree to which they're thinking about um uh sovereignty and control over their their tech infrastructure. uh and they're worried uh that if they become wholly down wholly reliant either on US or Chinese AI providers, they're going to be in a a weakened position politically.

Um now is is the best use of their resources solving that problem building data centers? I think that's an open question. Uh but we've certainly seen a lot of countries uh turn in that direction as their first response uh for trying to build a bit more uh room for maneuver in the AI space.

It's interesting in the as the UAE was was opening up to western businesses back in the day, there was a structure that was popular where if you were CocaCola or Michelin and you wanted to sell into the UAE, from my understanding, you needed to set up basically a J like a 50/50 JV with somebody locally.

So, it was like basically owned and operated to some degree, but almost like more of a franchise model. And I wonder if we're we'll end up effectively seeing something like that on the data center side. I just wonder I just wonder how much that particular layer matters.

Like if you look at what China's done in just the delivery of conventional web services, it's like obviously knockout dragout fight to try and build data center capacity and import chips, but also develop the traditional semiconductor, the CPU supply chain and then ban Facebook and Google and develop homegrown alternatives for the application layer.

But they haven't tried to rebuild Linux from scratch.

like uh North Korea does have a fork of Linux called Redstar Linux apparently but um in general you know Chinese phone companies have been fine forking Android uh that hasn't been a place where they've said oh we got to rebuild we got to own that from start to finish it's like we want to be able to make the phones and we got want to be able to deliver the services so that you know Google can't like bring their own censorship rules and we can't and you know Samsung can't cut us off of actually delivering the phones but in terms of the actual operating system layer.

Everyone's been kind of fine with standardizing around Linux and Android. I don't know. It's interesting. Um I I wanted to talk about uh the AI trade AI trade deals. Yeah. Yeah. So, this this is what I want to get into. I I would love to to understand the talent market to the in China to the extent that you can. Yes.

It's been what what we've been covering for the last couple weeks, these sort of massive changes between uh in talent moving around between the different labs.

I'm curious if there's that level of competition for talent or it's like deep because in some ways, you know, these American, you know, uh uh researchers constantly, you know, shifting around to the highest bidder. Uh it's not necessarily the best for winning the AI race or winning the AI war.

It's the right strategic moves for the different labs and different hyperscalers to compete for this talent to to build the best businesses and products that they can. But it's not necessarily Yeah. An authoritarian might say, "Hey, let's just consolidate all of the AI talent at one lab in America.

" That's not the way America works. But how does it work in China? Yeah. I you still do have a pretty flexible labor market. I I think there's an interesting question. Would you be right to consolidate all AI talent?

I think if you're if you're a believer that we're going to kind of reach a single threshold called AGI and if you reach it two months earlier, you've got this kind of enduring strategic prize, then maybe concentration is the right strategy.

Um, but if you're a believer instead that there's going to be a bunch of different products which will themselves be pretty transformative, but there's not one threshold you want to pass, then actually you probably want a more flexible uh system that's going to allocate uh between different firms.

Then, you know, maybe it is macroefficient to have people jumping between firms uh once every year or two if they're bringing the cutting edge, spreading it to different firms. That could actually be a good thing on uh on the macro level as well as for the the individual level.

I I I would be more pessimistic about China's chances if I found them centralizing more aggressively. I thought one of the most pessimistic data points this year about uh the Chinese a ecosystem was the news that deepseek researchers were having their passports um confiscated. is there to report international travel.

Um, which to me is a strong disincentive to do it doing research. They China seems to love doing that. They were doing that to rare earth element experts as well. Rare earth minerals. Uh, it seems like it's just in the playbook when when a when an industry becomes important, the passports start disappearing. Um, yeah.

I mean the the other the other uh elephant in the room is you can imagine the amount of pressure that the CCP is putting on Chinese AI researchers that are here in the US.

Maybe they want to they're they're citizens already or they have a path to citizenship but if they have any family or any exposure at home there's still that's still a lever that they can pull. I want to talk about talent uh in the context not just of AI model development but actual chip fabrication.

Uh it feels like right now what we're seeing in the foundation model labs is an extreme power law in the value of AI researchers. Hence hund00 million offers for top talent and starting salaries in the 300k range. Like that is a widewide power law, right?

Uh there's always been this narrative around TSMC that you can't chat GPT how to make a semiconductor. There's a couple books out there, but it's it's a lot of lore. It's a lot of toutelage and mentorship. How power law distributed is talent in semiconductors? And is there a world where chips act 2.

0 gives Intel a bunch of money or something and then they start offering TSMC experts $100 million offers and that really pulls forward American uh chip production. So I don't think the mental model is is that expertise is power law distributed.

I I think that there's lots of very unique capabilities you need to make a chip and the level of expertise you need in each of the verticals is very deep and you need all of them and if you only got 90% of them your chip doesn't work and so you need to build this expertise over a long period of time and so if you were to say who are the three people you need to uh hire from TSMC to replicate TSMC I would say that's not the way to think about it you need to hire half the half the staff or more.

Um when TSMC was bringing the facility in Arizona online, uh they had hundreds of people from uh Taiwan working in Arizona and this was not even their most leading edge ship production there. This was two generations behind uh the cutting edge with all of the explicit help coming from headquarters as well.

And so I I think that helps illustrate how much of the the transfer needs to happen at different levels. But, you know, if you look at who's Mick hired over the last decade, it's been a ton of engineers from Western Taiwanese Singaporean semicunter firms.

And I think a lot of the advances that they made can be ascribed to the fact they were able to import all of that human capital uh from uh workers who had previously spent a lot of time at other companies. What are you tracking? We're halfway through the year.

what what kind of events are on the horizon or or various uh potential new legislation are you kind of following? I think from the the policy side there's going to be some action on tariffs and semiconductors.

It seems president's repeatedly said that unclear exactly what direction that's going to take but it could have pretty big implications for the cost of building AI data centers.

Uh that's that's certainly one and I think on the kind of tech controls front uh we haven't seen the last of the moves from this administration either and I wouldn't be surprised to see more there too. Can uh have you been tracking uh the leadership change at Intel Lip Bhutan is coming in.

It seems like they're going to be doing doing a lot of layoffs. I was trying to go back through Ben Thompson's analysis of Intel through the years and try and understand how the narrative I haven't been tracking Intel that long to follow whether there should be a split.

Uh what has been your take on Intel and has it changed and how do you think what what should we expect from Intel over the next few years? It seems like they're going to be tightening things down, but do you expect any significant changes? Would you recommend any specific changes?

I'm interested to hear your take on Intel. There's a lot of people who think that a split between the manufacturing design businesses is the right way forward. I think that the question if you think about a split is what does that mean concretely?

Um because the reality is you'd still need to have a fair amount of integration between the design side and the manufacturing side because the manufacturing side is making chips for the design side that will be in progress for a number of years and the manufacturing side needs business from the design side.

So, you know, split or no split is probably too simplistic a way of thinking through the option set. There's a bunch of different variables uh around that. And I think that's probably exactly what Liipu is wrestling with uh right now.

Um I think that the second challenge that Intel has is in its products business, the CPUs for PCs and data centers. I think there's a a a wide uh agreement that they've got work to do on both those fronts in terms of uh um remaining competitive with AMD and they've lost market share uh over the past couple of years.

They need to win that that back. And so I think those are the the two key issues he's got to deal with and they're of course interlin and it's hard to um hard to say you're only going to focus on one of them. Totally. This is great. Any other questions? Fascinating. No, would love to have you back on. This is fantastic.

Thank you so much for taking the time next quarter. really appreciate this. We'll talk to you soon. Cheers. Bye. And I'll be right back. Up next, we have Aaron Gin, the GPU whisperer, as uh his moniker in the information