Quantum Race, China Supercomputing and Open-Source AI Independence

Hey everybody, happy hump day and welcome to the Techstrong gang. We are going to have an awesome lineup. There's just an amazing number of things happening in the world, and they're all pretty profound, but let me introduce our panelists for the day.

JP Morgenthal, good to see you again. How are things? Going very well.

Good to see you all. Good to see you. Wicki Wang, always a pleasure.

How are things with you? Good, good. Hi, everyone.

All right. Terry Robinson, I believe that brick wall signals that you are in New York. I am in New York, still riding the glory of the Knicks' victory.

All right. Awesome. There we go.

Andy Mann, how are you doing? You look like you're in your usual haunt as well, but I don't know, man. Yeah, no, I am back in the hometown of Boulder, Colorado.

All right. And then my favorite colleague, John Schwartz, how you doing out there on the West Coast? I'm doing well.

It's good to see everybody, and we're raring to go. All right. Well, I swear, if I'm ever bored, I just have to wait for the president to do something or say something, and we'll have something to talk about.

But this time though, I find myself maybe agreeing with the president. So he's issued a directive talking about how we need to invest more in quantum computing. It needs to become a higher priority.

JP, what's your read on this? Do we need the president to weigh in here, or will market forces just lead us where we need to be anyway? This is not unusual.

We've seen these activities from multiple presidents who support the advancements of particular technologies, especially ones that have a direct influence on the future of national security. And clearly quantum computing will, due to its capabilities. The number one use case that they keep throwing at quantum computing is breaking encryption.

And they claim that quantum computing can break blockchain encryption in under seven hours. So I mean, that's a pretty significant threat, I think, that we need to have support for that outside of just commercial entities so that we can ensure the safety and the forward advancement at an appropriate rate. And I assume at some point somebody will figure out how to leverage quantum computing for encryption of blockchain, and we'll end up having to re-encrypt the entire chain, but will it improve the security?

I don't know. Well, Terry, I'm going to come to you on that since you're our resident security person. But I feel like the rest of the world is sitting here cheering on the advances in quantum computing, and we'll unveil the wonders of science and nature, and we'll build all kinds of exciting things.

" Yeah. It's sort of saying, "Oh, the total collapse of global public key cryptography on its way. " Yeah, I think there's some right to be both.

To be sort of overjoyed about quantum and what it brings, but to be really worried about the security implications. I don't know. There's so much there, like the infrastructure overhaul that has to come along with this.

That's a huge nightmare for, I think, the security people. And we've been over the retroactive espionage with the harvest now and decrypt later. I think everybody's sort of well aware of that.

But yeah, I think that the security people have every right to kind of run screaming from the room and then come back in. But I also think that there's a little time here to strategize and start making some real plans. Right.

Wicki, I'll get your sense of this, but some folks are saying Q-Day, which is the day when quantum computers can break this encryption, I think it's supposed to arrive sometime in 2029, is the best guess. But I think it may be sooner, especially if we start pouring more money into this thing. And ripping out and replacing the encryption schemes is non-trivial, and I don't even think you can use AI to do it.

So, do we have a challenge here in terms of actually making our systems comply with something that is a whole new encryption standard? Yeah. I think it's kind of like every time we have an emerging technology, everyone's trying to find out what's the timeline, what's going on.

Do I need to have additional budget, right? I remember three years ago when I talked to a startup, they are doing the quantum security. " And then after three years, we heard the news, right?

So there are a lot of people working on this area already, but never got the real timeline. Now we're saying, okay, it's a real time. We say it's already the present tense.

It's not like something in the future. I would imagine everything will follow the cadence. People will start to ask how do we, in taking the quantum-related item, and then how to evaluate and see where's the law or compliance requirements around it.

It's just like when the AI or- privacy law just rolled out. Yeah. I'm waiting for some more information from some- ...

law or a governance institution to help. I'll definitely talk with ISACA as well. Yeah.

Mm-hmm. I'm really wondering- I think it- ... the, I'm sorry, just the timeline for Q day being accelerated.

Yeah. How that's going to really affect the timeline that the defenders- I don't- Can I just mention, like do- Yeah, I think it's important to realize also, unless you're in a John Hughes movie, you're not building a quantum computer in your bedroom. So there's only a handful of institutions right now that have the capability, and have built out quantum computing.

So, as it starts to emerge, clearly these organizations are going to have a leg up over everybody else. So, hopefully that money will be used to go to learning institutions and allow them to start to build out some of this quantum infrastructure so that it's not in the hands of IBM, Microsoft, Google, and a small handful of others around the world. But isn't that the current plan?

Isn't that- Isn't that the current plan? That is the current plan. IBM is building a big thing here in Poughkeepsie here, outside.

IBM's been doing it for, I'd say one of the longest, right? The most advanced research, you're right, on quantum has been in the IBM lab. So if they can monetize it, which monetizing stuff out of the lab has always been a problem for them but if they're able to, they would certainly be in the front of the pack.

Yeah. I remember like- And IBM's challenge... Sorry, go ahead.

No, go ahead. Okay. Yeah, I remember last year in the NVIDIA GTC conference, right?

There was a quantum day, right? People already prepare for that, I guess. I guess.

I think so. John, you wanted to weigh in on something, so let me just bring you in here. Okay.

Thank you. Thank you. So- ...

am I the only person who is skeptical about this accelerated timeline? So they want to push this dead- they want to move this up to 2028. Broader industry consensus was 2030.

IBM's ambition is, what, 2029? Does that- Google says it ... by trying to achieve this, does this force the scientific community to dramatically accelerate breakthroughs in error correction, hardware resilience?

And there's this idea, notion of the Department of Energy mandated to set up a national benchmarking center within 180 days. I don't know. I just, I- Well- Okay, but this is- Given this- It's kind of a mess ...

given where the mandate is coming from, I'm pretty skeptical. All right. Andy's agreeing with you, so let's hear from Andy.

I hate to agree with you, John. I always do. No, I really do agree with you on this one.

I hate that I'm a poli sci major. Look, even a blind squirrel can find a nut sometimes, right? This is really important, JP, you're absolutely correct.

It's really important stuff. We should be doing this as a nation. But look, the executive order directs the assistant to the president for the science and technology, a guy by the name of Michael Kratsios.

If you remember him, he might be familiar as a former principal at, oh, I don't know, Peter Thiel Capital. That's true. Look, is it just me or is this crazy?

A, the timeline, John. Yes. No one thinks that this is achievable.

B, what is this actually doing? This executive order is engaging in magical thinking to say we're doing quantum. It's almost like a scene from "The Matrix" where our chief executive is getting plugged into some scene and is like, "Ooh, I know crypto," right?

I mean- ... this is just nuts to me. It's not going to happen on time.

He gives no resources to make it happen, and he puts one of his oligarch cronies in charge of the process. I don't know. Call me skeptical.

Well- Oh, yeah. One other thing, sorry. You're so right.

I so agree with you, Andy, because given the context of what's happening today, oh, we can't clean up a reflection pool, and that seems to be the topic of the day, right? We can't even do that. How are we going to establish these uniform precision standards to keep pace with this technology and these goals that other companies that are far better equipped to do so?

How are we going to leapfrog that by a year? Well, that brings a good question, John. Is anybody keeping up with all of these EOs, and decrees, and whatever, and actual progress made on any of them?

Because I don't see it. That's a great question. No.

So let me just tell you, the first rule of Irish politics, if you want to get elected, you find a parade, and you get in front of it. This ... So, but to that point, let's just say that the timeline's not accelerated.

So we're looking at, let's say it's 2029, and we'll split the difference between 2028 and 2030. I counted on my fingers and toes here, but I think that there's roughly 1,000 working days between now and 2030 or the end of 2029. If I add in holidays and weekends, it's even less time.

So, and Wiki, are we already behind the eight ball here, and there's just no way that we're going to get all these encryption schemes updated, so now we're going to have to pick and choose the ones we can get to? Well, I want to give you an analogy that for AI law, right? If you look at the AI law for Corado AI law, right?

They just got reset. It's supposed to ask some questions ahead of time. Now they change their position to take responsibility after the facts, right?

And also you can see the timeline for several different laws or regulations. They extended the timeline to meet people's needs. So I think maybe there's always some room there.

People try to set up the best and give, but during the reality happens, there's some adjustment, I should say. But give people some pressure, maybe let people understand the importance, and try to prepare. Start from now, it's a way to go, I guess.

Yeah. JP, and the other thing that comes to mind here is it's one thing to build a machine, it's quite another thing to build software that runs on the machine. So if the machine is kind of floating around there in 2029, 2030, how long before we might see some actual applications?

Because last time I checked, there wasn't a quantum compiler anywhere. That's just three hours with Claude. I have one when we're done with the call.

You want me to start it now? The problem is I have nothing to test it on. That's the issue, right?

So again, back to the John Hughes movie, I can't exactly test the quantum compiler in my bedroom. So, yeah, I really think now you're getting into what would've taken years will take months to start producing software once there is platforms for testing and build-out, right? That is, I would say, the major bottleneck probably to the speed of advancement is a working, available compute power for people to use.

Mm-hmm. And it gets better. Andy, you live in this world of observability.

So now I got this quantum thing where I got two things exist in two places at the same time, and I'm supposed to observe this somehow or other and actually monitor and manage this. And so what would IT operations look like in a quantum computing world? Oh, look, this is going to be really good fun.

Honestly, this is actually what keeps me going, because I look at this and I see an entirely new platform. And I get to renovate all my old ideas from monolithic platforms. Mainframe is back, baby.

But for real, this is going to be completely new forms of management. You've all said it for 100 years, first comes the revolution, then comes the management. We will need a whole bunch of new technologies to look inside these computers, understand what they're doing, have them emit data, monitor them for their own security.

Let's face it, if I'm building a quantum computer that can crack all other encryption, I need the mother of encryption on that machine, too. So we're going to have to develop a whole new way of operating this environment, much more akin to supercomputer operations- Mm-hmm ... than it is to mainframe, quite frankly, because you are talking about immense volumes, immense amounts of data, immense amounts of throughput.

So this is something that, and we're talking about giga scale, peta scale, whatever is processor combinations. This is just a nuts environment to try and monitor in real time. But the upside is we're developing AI agents for SRE and observability.

Ask me about that sometime. And that's what we're going to need to monitor these computers. We're going to need AI to monitor the quantum, because otherwise, we're going to get lost.

All right. And it gets better. John.

Yes. So how much, to your point, is this a little bit of a, I don't know, I don't want to use the word scam, but- ... at some point, am I going to see a bunch of startup companies and IBM and everybody and his brother wrapping themselves in American flags, touting the fact that quantum computing is going to change the world and you should part with your investment dollars in this bucket here because, well, this is where the big return on the Wall Street is going to be?

Yeah. It's like a Wall Street play. Everything is through the prism of Wall Street.

Everything. Yes. According to one individual.

And this just kind of reminds me, to a lesser degree and a different degree than what happened with the Apple/Intel announcement that trumpeted a couple of days ago, a few days ago. We're going to bring manufacturing back to the US, it's going to be a game changer. Apple's going to spend $600 million.

They're going to be working with Intel. It's a lot of bluster. It's primarily done to affect and influence the stock market, like peace treaty talks are.

This is all just an end game about inflating the 50,000 plus number. It's, I mean- It's all window dressing. It's window dressing ...

a circle jerk. That's, well, yeah. That's it.

Sorry. I shouldn't say that. That's- That's fine.

No, that's okay ... maybe a little excited. I get excited ...

a less polite way to put it, but that's exactly right. A lot of window dressing. So JP, we may be in actually a different John Hughes movie than the one you have been thinking about.

Right. You know? But also, JP, help me with this, but as I understand it, the goal line keeps moving.

And that goal line is that the fact is that classical computers, they keep getting better as well. So the difference between what the quantum can do and what the classical can do, and maybe the classical can do at a less expensive price point, keeps changing? It's still copper to fiber comparison.

Okay. There's only maximum bandwidth you can push through copper. And you can add algorithms, you can do all these things, you can do splicing, you can run multiple path, and you can get improvements, but you can't touch fiber.

Fiber is always going to have that improved performance way above ... what the cap of what copper can do. And it is a similar thing here, right?

You have regular CPUs that are operating in a binary system, right? And quantum isn't a binary system. I think it's octal.

So, you're talking about the ability to move the amount of bits, is logarithmically, that's a hard word. It is. Right?

It's logarithmic. It's not linear in scale. All right.

I'm going to move on here, but I will just leave this one last observation is, I think the way we're going to talk to these quantum machines is through our AI agents anyway, and natural language, and I might not know anything about quantum. I'm just going to tell the AI agent to go do something interesting with quantum, and that's how I'll express some level of interaction. JP?

You still need a quantum compiler. You would think this is so- You still need a quantum compiler. And somewhere along the line, so the AI agent writes the code that the quantum compiler is going to compile, right?

Right. The question is, if it were a tool, what would that tool be doing? Right?

Why do I need quantum computing? I think the use cases are really interesting. We've mentioned one of them.

I haven't done the research on what the underlying use cases are that require a quantum computer over a binary computer. And I'm sure they're out there, I just haven't read them all, or I'm not studied on use cases. Right?

I think it's just cool. One of the ones I imagined was going to be when they first started talking about quantum entanglement, was that we would actually be able to have data in, let's say, US and Singapore simultaneously, right, with no delay. That would be a significant advancement in real-time computing.

Right. Well, I don't know the answer to that, but it does occur to me that if I hack one qubit, does the other one get hacked too, or how does that kind of work? Well, yeah.

What happens in one, happens in... That's the goal, but right now the gates are linked together with, what is it called? Cryo-level materials.

Right? It's deep, deep cold in order to operate. " And it's two meters.

So it's like, okay, well, now I'm in the glass house. Now I'm just building, like Andy said, a new mainframe, right? And it's like- And then the- ...

okay, what am I doing with it? And that's why IBM has a big hole in the ground in Poughkeepsie where they used to make the mainframes, so go figure. All right.

Let's shift the gear here a little bit, but there's just a lot of theater in the world these days, so let's continue on to the next topic. All right. There's reports out of China now that says that they have managed to build the world's fastest supercomputer, and it's CPU-only driven, no GPUs, because, well, they might have access to the latest and greatest GPUs yet.

But John, you wrote about this. Is this a leapfrog thing we're going to continue to see back and forth here? And the fact that, I'm just going to say it out front, I think technology is resistant to any kind of sanctions or any ability to control.

Mm-hmm. Yeah. Sanctions.

Yeah, the thing about this story is more about geopolitical and US export controls, and yeah, technology will find a way, so to speak, to kind of channel from "Jurassic Park," "Jurassic World," whatever. Yeah. So this unlisted Chinese supercomputer, which is called Linshan, has emerged, and it displaced the US's premier system, which is El Capitan, by 20% in terms of sustained performance.

It's interesting, and you mentioned this, Mike, what's interesting to me is that while most modern-day high-end supercomputers rely heavily on specialized GPUs from people like NVIDIA and AMD in America, Linshan has built one entirely without them. Instead, they've used this architecture relying solely on CPUs, including some customized core processors from Arm. And in a sense, what they've done is they've done an end run around US export controls through utilizing standard microprocessors rather than having to depend on restricted AI chips.

So they've kind of bypassed these embargoes. And it's interesting, and it kind of, to me, begs this question, does this kind of milestone prove that China's neutralized US export controls? Or on a flip side, to the US's benefit, have we successfully bottlenecked China's capacity to build specialized GPU-driven hardware that's necessary for next generation AI?

It'll be interesting to see where it goes. So I think this opens a can of proverbial worms, not just on technology, but in terms of geopolitics. I think there's companies in China that already have signaled that they intend over the next five years build world-class GPUs, or what we might call AI accelerators, because I'm not sure GPUs will be the right answer five years from now, but that's a whole other subject.

I mean, do we need all that graphics overhead, or we just need something that runs algorithms really, really fast? So, well, that's part of the next conversation that goes with that. But Andy- Am I looking at a scenario here where it's not just China, but the rest of the world is going to start investing in tech and doing all kinds of interesting innovations?

Because maybe it was easy to be dependent upon the US because that's where the dollars and the money were, but now that it's become a political hot potato, maybe we'll see all kinds of exciting, interesting innovations from all over the world, and that would be kind of cool. Yeah, look, I actually think you're right. And John, you're spot on.

This is absolutely a geopolitical play. And as you point out in your article, excellent article by the way, great interviews, there's five of these in the world now. Five.

Count them. On one hand, I can count them. Let's talk about this idea of moving away from US technology, though.

Because if I start to think about alternatives to Starlink, I start to think about alternatives to Anthropic. I start to think about alternatives to, obviously, Windows. Also all of the application software.

And this is not theoretical. This is already happening. We see in Europe specifically, looking to build their own alternative to Starlink, not trusting the oligarchs who run the current systems for satellites.

Look what happened in the war with Ukraine and Russia, where all of a sudden Starlink just got turned off. So as partners, I don't think our geopolitical allies can actually rely on us that way anymore, and I think we've proven. And I do think that this is fundamentally just a poke in the eye.

When your geopolitical rival is down on the ground wrestling three or four different things, including an algal bloom- ... " I just think this is fantastic timing in terms of the geopolitical leverage. But you're absolutely right, Mike.

This is signaling to the world that US technology supremacy, for want of a better word, is not everlasting, and there will be alternatives. And okay, so this is one of five supercomputers, but this absolutely does signal that there's technology coming from places where we maybe won't expect it, and it is going to threaten the US hegemony on technology leadership. As well it should, though.

There's no guarantee that we should lead in everything, and our leadership is eroding across the board anyway. Accelerated, I'm going to say, by this administration. So, I'm interested in what's coming next.

I think there's some exciting technology on the horizon. But, you know. All I could think of as Andy was talking about it was, so is there going to be a supercomputer cage match in front of the White House?

Is that where we're going? That might be the only way it gets the administration's attention. JP, let me ask you this.

Do developers care about where the tech is made and where it comes from, or are they just purely interested in the what can I do with this thing, and they're just going to go and write software regardless of where it comes from? Because they, in their minds at least, live beyond the borders of any given country and they're just interested in the software and doing cool things. I don't think you can generically say developers.

I think there are people coming from different backgrounds that understand risk and implications. There are plenty of developers who could give a crap less, and then there are developers who understand the implications of leveraging and using some of the software and the potential risk. And would not want to enable or take that risk.

So there you go. Wiki, let me ask you this. Is this ultimately a good thing?

Because at the end of the day, we spur each other on, it's competition and the next person wants to up the whatever the other one did, and this is just part of the natural cycle of tech and it's all good. Well, so I'm not a big fan for the geopolitics, so I will leave that for the geopolitics folks. But from my side, on the emergent technology side, what actually grabs me is the engineering story under these.

So you can see at current stage, we all talk about how AI and the fundamental infrastructure will go. When you can't get the chips you want, how far can the good architecture and software carry you? So turn out, pretty far .

So they build this on the CPU instead of GPO, trade the power efficiency for the raw scale and make it work. I think, to me, it's a reminder that computer was never just about who gets the fattest chip. It's architecture, software, networking, power, and sometimes even the adoption piece.

All at once, that's the same hard problem for every one of us to think about it. If we can't go one way to achieve the ultimate goal to optimize something, maybe someone else or some company will work it out in the other way, no matter where they're located. This is true.

And I think that's actually a really important point, Wiki. Really smart to think about... And by the way, really smart to bring us back on track, thank you.

But also, really smart to think about what this actually means in terms of the technology that's been invented here. Using alternative technologies to get past the US export ban has actually shown that there's a lot more room to move in this space generally If you can do all the other things that the Chinese government people have done to create the architecture change, and by the way, I'm totally with you on that, Wiki. Mainframe for life.

It's the architecture, it's not the chips, right? And you can get that level of incremental improvement without even using GPUs. Wow, what is possible if you could use GPUs and all of that innovation as well?

So I think that's actually a really interesting point, Wiki. What happens when the GPUs catch up and they're able to do that, all of a sudden, they're able to put those in the same architecture? That's going to be pretty revolutionary.

Right. John, stay with me on this line of reasoning here and then tell me if I'm crazy. You're crazy, Mike.

Go ahead. Well, thank you. But it's interesting to me that, I'm going to quote Eisenhower here, right?

Eisenhower warned about the military industrial complex, and we wound up spending an inordinate amount of money to create some of the most sophisticated weapons in the world. And is that just playing out here again, all over again using AI? " And whether it's AI or tanks or airplanes or even machine guns, at the end of the day, cheap and easy seems to win out a lot.

You know what? That's actually really a fascinating... I'm not kissing up to you, Mike, but that's really a fascinating comparison though.

I think you're exactly right. It's like if there's another way to do it, an alternative way, a more efficient way, and in a sense, this is kind of a consequence of this country kind of going it alone or just not being a leader anymore. And also, kind of buried within the story, and I don't mean to go off on a tangent, but I see this kind of movement even among companies.

We'll talk about this later in another show, but less reliance on NVIDIA. Like for instance today, OpenAI and Broadcom announced this AI chip they're developing together, so they're less reliant on NVIDIA. It's just if you can go a cheaper, more efficient way, why not?

And you know what? I welcome the rest of the world moving on and kind of working around us if we're going to make things as difficult as possible for them through tariffs and other types of controls. Yeah.

Who knows? At the end, maybe NVIDIA pick up by themself on their side. Get like- Hmm ...

either way. You never know, right? Yeah.

This is true. So the good news is it's a global community, and with that I'm going to shift gears to our next topic because it fits right in. There's a UN conference that happens every year, and people come and they talk during this whole thing, and it's been an ongoing thing.

And not a lot of people pay attention to it, but Stephen Vaughan Nichols, who writes for us, was there and he goes every year. And the major theme of the conference was we need to invest more in open source AI models. And what's really at the core of this thing is that if we're dependent upon proprietary AI models, they're black boxes.

We don't have enough visibility into how these things work, and they might be used for all kinds of malicious purposes. And open source is all about transparency, and we all get to see what's happening, and we all get to see reviews, how things are done. Wiki, can we all just kind of join hands on this and say, "Let's move to open source AI models, and maybe the whole world will be a better place"?

Yeah. I should say, I have a lot of respect to LeCun, and he helped build up this field and his case for open source AI at UN was really one of the most thoughtful- Mm-hmm ... I heard.

The part I really like in this news is what he mentioned, we need such a high diversity of AI systems for the same reason we need a high diversity of the press or news, right? From this aspect, I think he's right, and honestly, I agree with most of what he said. Recently, I do talk with some of my engineer friends.

In their company, they're evaluating the open source AI model as well. For the business, sometimes they don't care how fast the AI is or something, how advanced the technology is. What driven them is actually the token cost, right?

So it's like business choice, I guess. Mm-hmm. And also, nobody want to have two to three companies deciding what the world is and how it reasons, right?

So, I think because I take him seriously, so I want to be honest about this piece. Some part, I can't follow him by the point where he argues many current safety and complex efforts, risks showing progress. I think the security and the risk is still important, but the diverse press never mean a press with no standards, right?

We still need those accountabilities, and we need some guidance to let people know what made the first trustworthy. So- Mm-hmm ... that's my point.

All right. JP, what's your thoughts here? Because on the one hand, we're torn.

Out in the tech world, we're capitalists at the end of the day, and there's kind of this notion that if we invest more, we'll get advances faster and we'll generate more profits and society will benefit. And then there's another thought that says, well, you know what? This stuff is open source.

It's software and we share it, and it is what it is, and it's not really a place where you can make those kinds of investments long term, and it's just shared intellectual property. Well, now you're back to Cathedral Bazaar. Yeah.

Sort of 2000. There's been two arguments forever on this. One is we open source so that we can gain a much larger community looking at the source code to ensure its viability, its security, and to understand its implications on use in your environment.

Never intended to be free. And then there's the open source community model where it's free to me because I contribute to it. As long as I'm contributing and enhancing, then the value to me, the benefit to me, is that I don't have a license fee.

Right? And so now what you're talking about is I think if you extend that model or that approach to AI and frontier model development, I think that both are fair. In fact, the author of that piece or article made the point about how we need more people looking into the frontier model development, and the models themselves, and that comes from one of the benefits of AI.

Sorry, one of the benefits of open source. So I think if you stick to one of those two models, it makes sense. But there's going to be plenty of people who are just leveraging it because they don't have a license to feed a bay or they're not paying token costs.

And I think that model started to fall apart in software and applications to some degree. You see a lot of abandoned communities, a lot of abandoned software, a lot of abandoned projects. In fact, if you go and you build anything with React, you get about 30 or 40 messages about this one's looking for funding, this one's been deprecated because no one's supporting it.

So that's not the value of a model, if it doesn't get advanced, if it doesn't get updated and trained. Mm-hmm. But for all those projects that are suffering, there's other projects that replace them, and that's just part of the ecosystem of the thing, right?

So, if you're using React, I think you're supposed to be using some other framework anyway these days, but what do I know? Andy, as you kind of look at all this stuff, I guess it's fair to say, we have seen Wall Street kind of bounce up and down a lot. I think it's bouncing back today, but there was a big conversation about the investors are starting to figure out, hey, this AI game ain't exactly what we think it is.

So are we starting to get a broader appreciation of all these issues that we're talking about here, and is that going to factor into the amount of capital that's available to invest in all this stuff anyway? Because the investors are going to be looking over their shoulder at all these open source initiatives anyway. Yeah, look, it's a really good point.

And I love Steven's writing, and I'm just going to let him know that I'm looking forward to 2027, the year of open source AI on the desktop. Yes, let's make it happen. To be fair, Steven's been calling for Linux on the desktop now for two decades.

I know. Yes. Love his writing.

Look, yeah, it's going to be interesting. I think there's always room for two ice cream stands on the beach, right? So we've got the proprietary vendors who are building their own models.

And yet, to your point, Mike, it's really hard to move past a profit motive when it comes to driving innovation and technology advancement. It's really positive. So much of what we have built has come from really strong profit motives.

The other thing is that open source is an opportunity to develop for smaller communities, and I think that's actually a really important part of this idea. Not just that we can build bigger frontier models and better frontier models, but when you open up to open source, then you get more people, you get more diverse views, you get more ideas, and that all drives more innovation. And if there are two ice cream stands, then one of them is free to make as much money as they want, but the other one can find avenues to develop specific ice creams for specific people who come up to buy them.

So where Anthropic's or OpenAI's, they're going to go for general purpose, sell as much as they can, open source AI allows to develop for specific use cases for nonprofits or for healthcare, for law enforcement, for education purposes. I actually think it could be really good. My big concern is more to, I think JP's point, who directs the open source?

How long does it stay around? Who funds it? Because open source doesn't really make money.

It's our dirty little secret here in tech. So who's going to maintain it? I look back to some of the open source foundations in the past that have atrophied badly because they've been owned by the same vendors that they are ostensibly competing against.

So cloud open standards, virtualization open standards were big there. So I worry about that. Who's going to own it, who's going to maintain it, who's going to be the gatekeeper on commits, all of this sort of stuff.

But if we can get it right, it really opens up a lot more opportunity. It doesn't have to compete directly. Mm-hmm.

Yeah, I think going back to JP's point about the bazaar, not all tech needs to make money. I know that's heresy. But a lot of it is fundamentally tech we use to build something else that makes money that's more interesting, and that's probably where the opportunity is.

So maybe we should stop thinking about how to monetize the infrastructure that's a means to an end, and start making money off the things that actually provide the real value to people, and the businesses, and the stuff that we're trying to accomplish. I'm just saying. John, last word?

Well, a great story by Steven. Open source AI is maybe the only viable path to ensure global AI sovereignty. So I'm all for that.

Well, I'll give you two choices. We need to review these AI models to make sure that they're safe and they're secure. So who do you want to rely on?

You want to rely on the open source community, or you want to rely on the United States government to do that for you? Oh, God. That's a tough choice.

That's a ugh. I don't even want to... That's like a- I'm going King Solomon ...

Sophie's choice. Yeah. The Sophie's choice of tech.

Sophie's choice. There you go. All right.

I'm going to leave it there, everybody. You can vote for whichever one of those two outcomes you think is best, but I'm pretty sure I know which one of those is going to wind up being the majority. Want to thank everybody for sharing their insights again.

Happy Wednesday. We will be at Platform Con tomorrow, so we'll be having a whole show about platform engineering here in New York. Allen will be there, I'll be there, Mitch will be there, so please tune in.

And we'll see you all again either Friday if you don't catch tomorrow's show. But once again, thanks for spending time with us. Take care.