AI Pressures Open Source, Tests Google’s Energy Strategy and Fuels Amazon’s Satellite Ambitions

Hey everyone. Good day. Welcome to Techstrong Gang on Monday.

It's another week, another week full of AI stories probably, and stuff going on. No shortage of stories to go around. There's no shortage of events happening around the world, unfortunately, either, or maybe fortunately.

Some are unfortunate, some are fortunate. But we've got a great group of people to talk about what's happening today. Let me quickly introduce you.

We have our friend Garima Bajpai. Looks like she's still home, yes, in Ottawa? Yes.

Got that right. Joining us, looks like he's home as well, in beautiful Hudson, Ohio, the one and only Steven Foskett. And he's also back home in the High Castle- Right ...

Chief Content Officer Mike Vizard. Very happy to be home. Folks, welcome.

It's good to be home, at least for a little while. So we've got an interesting show today. We've touched on this article, on this aspect or angle before, but the impact AI is having on fill in the blank.

We could do like a "Family Feud," right? The impact AI is having on, survey says. Well, this one, it's open source.

It's having a huge impact on open source, both good and bad. There's two sides to this sword. But Mike, I don't want to give away the whole thing.

You guys, we're going to talk about it. Well, two folks with tongue firmly in cheek put together something called Project Malice, and it was basically the idea that you could create maybe a clean room service for all these open source projects, and they were going to use AI to kind of go in and clean up a lot of the software, maybe just clone them and not have any of the license restrictions because, well, the AI will just create a version of that that anybody could use. In fact, last week when I was at that MCP conference we were talking about, a fellow from Duolingo got up and said, "We have this open source AI agent for creating a Slack AI agent," which was kind of interesting.

" So, it seems like we're getting to the point, Garima, where maybe it's feasible to everybody can just create a fork of an open source project and do whatever they want with it from here because there's no barrier to swapping things out these days. And is that where we're headed, and is that a good thing? Or do we just wind up with too many instances of open source projects to support, and then everything just gets crazy again?

What do you think? I have too many thoughts on this topic, but let me get started with thanking you for actually giving me this prime spot for talking about this opportunity. Because what I say always is that the question is not what AI can do, the question is what AI should do, right?

And this is like, I would use strong words here because I feel that an open source community advocate, I cannot close my eyes to the subject. It's to me looking like license laundering. And why I'm saying that, I'll tell you.

I have done some research on these topics and both these articles which, I've seen this published this week. The first concern I have is the core tensions which we have with AI models like bundling up code, and it has many dimensions to it, like code, weight, data, and how they produce output, right? And this could be a very simple thing that you make this as a copy, and it's not a real copy, it's just like an improved version.

And then you can change the licensing from a copy left license to a permissive license, which is basically a dominant license, when AI models generate code. And this is not a new problem. 0.

I don't remember which one was it. But essentially what it was doing was basically violating or kind of undermining the licensing of this open source and probably not giving enough credit to the creators, which fundamentally arise, this is a problem with fundamentally how we are using open source- ... to train models and generate code out of it.

And I iterated in multiple sessions before as well, that to me, it looks like we need a new licensing model because this redistribution or reuse or modification, how we could stop it? I'm also seeing that people who are generating these LLM models for coding assistance will somehow watermark these codes, and then we will generate another problem in the longer run. So, I think, sometime back in 2025 also, this was a topic, and Linux Foundation came up with this OpenMDW license.

And again, coming up with a license model is one thing. How to adopt is another problem, right? So I leave it back to you, Mike.

I think we need more conversations on this topic, and to me, this licensing laundering is not acceptable in this kind of ecosystem. I get your point, but I'm also at a loss to say, well, what's going to prevent anybody from doing this? Because, well, it's the fundamental nature of open source, and we're all free to do whatever we want to do off of the fork of the code in the first place.

So, if we start limiting what people can do, are we no longer open? I don't know. Alan, jump in.

Yeah. So, my friend Ron Guler used to tell me there's two kinds of freedom with open source. There's free as in beer and free as in freedom.

And free as in puppy, I should mention. Yes, but there's no free puppies. Exactly.

And there's no free beer. And there's no free as in freedom either, really. I don't want to be put the pessimist here, but just because you fork something, Mike, doesn't mean you could do anything with it, right?

There's this notion of derivative work, right? " You know that you're still somewhat beholden to the source code that you're using from that fork. You could fork it and use a different OSI license.

You could fork it and go down a different path than the main community one is going. But I don't think you can necessarily make it your own. It is derivative works, right?

And that's what these things go after. But here's fundamentally the issue with where we're living in right now. I can use a, quote-unquote, "clean room" and have AI develop a similar code base for similar functionality.

Now, one could make the argument that that AI was in fact trained on the code in the open source, and therefore it's also derivative. But that's a weaker derivative claim than a fork. A fork, you're acknowledging you took that code.

In an AI training, well, the AI trained on it, but it's not what it's spinning back here. And so you can do that. " And though the human doesn't go back to the project and look at the repo, the AI does.

Mm-hmm. So what we're seeing is less humans involved contributing code, maintaining code, being good open source community members, that has been the backbone of the open source movement. And instead, you've got a lot of AIs just sucking in code and using it to build things.

And that to me is the danger, danger, Will Robinson moment. Stephen, to us. Yeah, it is, and it's a really weird situation, as you and Grima are both...

This is a strange situation because we have some weird conflicting facts here. So first off, yes, you can fork an open source project and change the license, but you only change the license on your contributions and your new content. Right.

So it's not like you can just go to GitHub and find something that's under, I don't know, the GPL and decide that this is going to be proprietary or this is going to be, I don't know, under the Apache license or something like that. You cannot legally do that. Full stop.

That's established. Second is the fact that it has been, not 100% established, but it's generally been accepted so far that any output from AI-based coding is not copyrightable because it is not an expression of the human mind. So that means that no content from an AI coding assistant can be subject to any copyright or license, which is pretty weird.

So let's just set that one off to the side. Then the third thing is, as Alan just pointed out, absolutely, these AI LLMs were all trained on a lot of copyrighted material, whether it is code or prose or weather reports, I don't know. It's all copyrighted, and the fact that they're able to generate something that may or may not be different from that is based on the training.

Now, there's been an argument on the AI side that that's no different from Mike going and reading everything on GitHub and learning how to do all coding and him coding something up. And I guess I can see the point of that, but the problem with that is that it has also been shown that through association, and especially when we're getting these large parameter models, the border between learning and memorization is becoming very blurry. To the extent that it has been shown that a lot of LLMs, when asked to produce lookalikes of whether it is novels or code, will actually produce the original- Sure ...

word for word because, well, maybe it learned it, maybe it just remembered it. But the point is all of this. Now take those facts together, and we get to a very weird situation where essentially we have a machine, as the Hacker News writer that has commented on here wrote, we have a machine that is capable of plowing right through the GPL and ignoring it.

And it's not... obvious that that's illegal. I don't know.

Let me ask this question. What is the difference between what we're talking about here and what AWS wound up doing with Elasticsearch when they changed their licensing terms, and then they created a fork called Open Search, which then became a cloud native computing foundation, or actually, it was its own foundation now that I think about it. And they just took that code, and they created a fork, and off they went, and no AI was involved.

And so OpenTofu is the same thing, right? Right. But they didn't change the license, and they didn't relicense the existing code because that's not possible.

Right. In other words, when you fork open source code, the code that you forked, not the code you built on top- Yeah ... but the code that you forked has to remain under an OSI license.

Which is why I can still use MinIO, even though they changed the code about their new system. Right. Well, this goes back, I mentioned Ron Gula.

Ron, of course, was one of the co-founders of Tenable, him and Renaud Deraison. 7 of Nessus was the last open source version, and then it went proprietary. It's still free, but not open source.

And now that was interesting because Ron and Renaud were actually the owners of the open source code, so they had more leeway- Mm ... right? Because that was their code.

And so when it's your code, when they forked, they had that ability, right? They actually were the copyright and IP owners of the code. That's a very different thing.

That's a kind of a corner case. You don't see that in most open source kind of forks. Like, I also feel that this is not an open source-ish problem because if you think about this, like you know- This is an everything problem, huh?

Yeah. Yeah. So, how do you maintain competitive landscape when everybody is using the same LLMs, and they can produce the same type of code, right?

So there has to be some kind of boundaries defined with the overlapping IP issue which we have right now. Because, I'll tell you, this is again, a question that if you produce a code with one LLM, you can actually debate whether you have the IP for this or not, because users often have the IP of the generated code. But when you generate this code with three different LLMs, who owns this code, right?

So, there is a lot of issues, not only with open source being kind of hacked in the sense, but also like enterprise IP. No, it's all IP. All AI-generated content IP, what is the status of that?

And right now, the courts have said that if you, in essence, worked with the AI to generate that IP, it's your IP. Because otherwise, think about it, everything generated by AI we would not be able to use for sure because there could be conflicting claims here of IP theft. I know that's how you would like it to be.

But that's actually not right. Exactly. So far, the court cases have shown the exact opposite.

Anything generated by AI, completely by AI, is not subject to copyright. Well, no, Stephen, if I use the AI to generate something on my instructions- It is not subject to copyright. I believe it is, counselor.

I believe it's not. Well, we can fix this. I'll see you in court.

Those court cases are still winding through the system. We could have Judge Judy and Mike after this. Yeah.

So it's subject to this again. Let me ask this one question just to kind of wrap this up- Question while we move on ... before we move on.

So, if I kind of play this out, does this mean like the Linux Foundation is going to have to create a police arm that's going to go investigate everybody's use of open source software? How enforceable is any of this stuff? Who's going to know what I did in the first place, and I'm just going to be doing my stuff on the side here?

Karima. So they have come up with this license, at least they have been advocating since last year, because the AI-generated code is not only code, right? It's data, it's the weights behind it, how do you train the model to code as well as the code, right?

So there are multiple factors to it. So they're advocating for this open MDW license, which makes sense to a certain extent. But again, as I mentioned, that how they will implement it, right, it's a problem.

And if the LLMs do not have a legal obligation to watermark this code in a certain way, we will never be able to do this. So the enforcement has to be a legal obligation to a certain extent. Absolutely.

I think, as usual, it's going to take a little while for the courts to catch up here, but I'm going to go do a little research on this and get back to you, Stephen. I suggest you read the US Supreme Court case of Thaler versus Perlmutter from March 2nd. I will have a look at it because I think I can distinguish there between what was done under...

Well, we'll talk more. Let us jump on- The US Copyright Office specifically says in an FAQ on their website that prompting does not trigger authorship. Yeah.

Okay. We'll come back to it because a lot of people say AI is full of gas. Well, maybe it's gas that's powering these AI data centers.

Exactly. Mike, what do we have here? Well, it's interesting.

Steven, I know we've talked about energy consumption and AI at length on this show multiple times over, but I feel like maybe this is a step in the wrong direction. But what's your sense of what's going on here? Yeah, this is a real interesting one.

I actually spent a little time researching this subject over the weekend because it does seem wrong, doesn't it? That we would be rolling out gas turbines, literally like semi-truck trailer gas turbines, to power these AI data centers. Why?

Well, there's a very, very specific reason for that, and that's what they call in the biz, time to power. In other words, right now, the most difficult aspect of rolling out a new data center is not, I am shocked by this, it is not procuring GPUs from Nvidia. It is not procuring all the rest of the infrastructure, including RAM.

That's not even the most expensive element. It is not pouring concrete and building the site and all that kind of stuff. Apparently, Liebert units are hard to get, but it's not that either.

It is utility hookups. It is getting the megawatts to turn that sucker on. And right now, that's been a huge issue because according to the research, it can take anywhere between five and 10 years right now to go from permitting to power flow for grid-connected data centers.

And that is, as you can imagine, a bit of an issue if you just invested millions or more in building out an AI data center and you can't even turn it on. Well, somebody that you might have heard of, called Elon Musk and Company, solved this issue in an amusing way, which is they literally had semi-trailers bring in temporary gas turbine units to power their Colossus data center in Memphis, and then later in the Colossus II in Mississippi. And essentially there's about 20 of these portable truck turbine generators screaming away in Memphis to power up this data center.

And they were able to get it online in months instead of years. To the point now where these gas turbines are now in short supply. They were intended to be short-term solutions or used for grid outages or temporary needs or whatever.

Well, they're now in short supply around the world because essentially the AI data centers, apparently there's 40 of them running in Mississippi right now. Google is, as you mentioned in the story, Google is also doing this as well, which is not exactly in line with their promises. But basically, Google looked at it the same way xAI and Elon looked at it and said, "Well, we need a gigawatt of power.

" And the power industry's like, "I don't know. " It is a weird situation. It is not a great situation, but it is what they're doing, because otherwise, they just have these things sitting idle.

So let me ask you a question. You don't think somewhere up there, down there, somewhere, Deming is not sitting down with Barratt from The Goal and having a beer laughing over this classic theory of constraints, just in time, lean manufacturing. Because as soon as we solve the electrical issue, who's going to solve the water issue?

This is a classic theory of constraints bottleneck. I mentioned it last week. It's like the child story we used to read when we were little.

Remember, they had mice in a town, and so they brought in the cats to get rid of the mice, and then they had a cat problem, and they had to bring in the dogs to get rid of the cats, and they had to bring in something else to get rid of the dogs. And they had to bring something else to get rid of that. And so it went, and so it went, and so it went, and then all of a sudden, they had to bring in elephants to get rid of it.

And now they had to get rid of the elephants. They brought the mice back in. It's just like them cane toads in Australia.

Yeah, exactly. What's his name? So do you think at some point, though, that a bunch of states, I don't know, arguably in the Northeast, are going to file some sort of suit against those states in the South who have allowed this to happen?

Because all that CO2 winds up in the air, and it goes somewhere, and it probably- Well, it's the pollution ... gets there. It's also the noise, though.

Could you imagine having a bunch of semi-truck-laden gas turbines firing up 24/7 anywhere near you? Yep. Well, that's why they put it where poor people live, Alan.

I know. And they're happy to have them there, right? I don't know about that.

Consider all the jobs it's creating. Well, arguably, maybe some of those areas already have oil wells and all kinds of things, so it's just one more item on a long list of items. But oil wells don't make pollution, but they don't make the noise of a screaming turbine engine.

True that. Well, you want to know the craziest irony of this story? One of the leading makers of these screaming truck-powered turbines, I kid you not, it is called Solar Turbines.

Solar Turbines- Somebody should sue them ... was founded as the Solar Aircraft Company because San Diego is so sunny. And now, rather than having fricking solar powering these things, we have Solar Turbines.

Doesn't that sound nice? Hmm? Is it feasible tofuel any of this stuff with solar, or they just can't, no longer considering that out of politics, or what's the deal here?

Oh, hell yeah. I'll just say factually, you look at what's going on in China, look at what's going on in the Middle East even, deploying solar panels and battery is absolutely a effective way to do this. Wind as well.

Wind energy also works really well for this, especially with battery powers. And battery is a very effective mechanism of bridging the gap, and the prices have come down to such an extent that even in remote and peculiar places like Texas, they are deploying massive gigawatt scale battery facilities, even without solar, without wind, without anything, just to balance demand on the grid. Because yeah, we have the technology to fix this without gas, and solar would be it, except that there's some political dimension there that I don't really understand that's keeping us from- I'll tell you what, we don't want none of that green socialist stuff in my America.

I guess we're concerned about where those panels are manufactured, I guess. I'm not quite clear, but... Well, but not concerned about where those chips are manufactured- ...

or the networking gear they run on, or anything else. Nope. It's a pretty weird situation, isn't it?

When, essentially China has installed, I kid you not, 300 gigawatts of solar- Let's go ... panels in 2025 alone, and that would solve this problem if only. Yeah.

Oh, what a tangled web we weave. And what happens, just out of curiosity, because they're not going to turn this on tomorrow, I don't think. But, let's say the next administration comes along.

Is Google just going to slide this whole agreement underneath in a drawer somewhere and say we're back to solar panels or what? Well, the good thing, these things are mounted on the back of a truck. I guess they can just drive it away.

There you go. Maybe we could sell them to Latin America or something. Maybe we use them for their intended purpose, and that's in the event of an emergency or something, we could fire it up if we need power and suddenly- Yeah ...

you know, little things. They could be in Mexico in a few hours, right? But let me just tie the bow on this one before we move on.

What a well-conceived plan as we race into the 21st century here and build out our AI infrastructure that no one thought this through in terms of a true engineering kind of problem and solution, that how many Nvidia GPUs have been purchased that will sit idle because the data centers they were bought for were never able to turn on for one reason or another? This is fire, ready, aim, and it starts with the incentives that the government put in there to race ahead because we go to the Moon because it's hard, right? And all of this stuff, and we want to be the first and the last and the best.

Where countries that maybe thought it out a little bit better, supply chain. You would think we learned supply chain lessons from COVID, but no. And what a shame, because it's not a trivial amount of money here.

We're talking trillions of dollars. You got to ask, all this AI investment stuff, is this just really, at the end of the day, a real estate scam? We're just trying to buy some property and move some stuff around, and we- Well, when you're a hammer, everything looks like a nail, and if you look at the guy up in the top rung over there, he's a real estate guy.

There you go. All right. Let's move on to space.

Space. The final frontier. There you go.

So, there is this space race going on around satellites and communications and networking, and largely between Elon Musk and Bezos, who runs Amazon. And Bezos is making a lot of plans lately, and then suddenly he turns around and he goes, "Hey, wait a minute, maybe I don't need to build all this stuff. " And so now there's an acquisition plan to get Globalstar.

I don't know if Globalstar's network is up to the task at hand, but Alan, what's your take on what's going on here? Because it does look to me like even if we do or do not put data centers in space, we do need these satellite networks, but I'm not quite clear who's going to build them and how fast they're going to get there, and will I ever see the stars again? So as a Pittsburgh Steeler fan, I look at the competition between Starlink and Jeff Bezos's Amazon much as when the Cincinnati Bengals play the Baltimore Ravens.

I don't want either team to win, and if there was a lot of players injured, I'd shrug my shoulders. Not that I ever want to see anyone get injured, but they're not two of my favorite teams. However, I was out in Napa Valley last week, first world problems, and we were staying in a guest house on a vineyard that was hooked up line of sight to a Starlink set up down the road a bit, and it sucked.

I got to be honest with you, the internet sucked. And I spoke to them about why they wouldn't put a Starlink in the guest house. There is, for those of us who sit prettily in our urban or suburban digs, and the building or network or neighborhood we're in has fiber run underground and right to the doorstep, we forget that the rest of the world doesn't work like that.

And this satellite-based internet is the difference between being a have and a have-not for a lot of people in this world. To me, this should be the Peace Corps initiatives of this decade. I don't want to sound like Al Gore, but I can't help it.

Bringing broadband to the world has to be a human right, because if you don't have broadband in this world, you are at a distinct disadvantage, and you'll never be able to catch up. So I'm all for anything that's going to allow us to bring more satellite coverage. Yes, there is the space junk issue.

Yes, it's controlled by these two Dr. Evils, and they're going to use it for their own billionaire tech bro reasons. But yes, we need to do this anyway.

It's that important. And so if this Globalstar acquisition can help Bezos have a competitive offering to Starlink so prices come down, there's more choices for people, and most importantly, more people in Africa and South America and Asia and other places have access to broadband that they can't get now, it's a good thing, and we've got to support it. Yeah.

I want to add something here because I think the cause you are referring to, Allen, is great, and if this could have happened with this cause, that would be really incredible. But coming from a telecommunication background, there are several core issues with this. The first issue is this whole technology is CapEx intensive, right?

Doesn't create a business case for mass rollout of broadband to rural communities, which it was intended to be. So if it is not subsidized, it's not government incentivized, it cannot be a partnership model between big tech giants because these tier one, tier two companies, I mean, the business model is very weak, right? So it's a very high intensive CapEx opportunity.

The second problem with this is that it could work with specific industry segments like oil, mining, agriculture, because the reliability there is not that a bigger issue. For certain use cases, you could use those, like your experience with Napa Valley, for example, right? So there is also some technical issues, so problems with spectrum allocation, how this whole commercial gigs would work, what kind of regulatory communication models we should put in place with the radio technology.

It's also something which we need to be aggressive about. And the last part, which Elon Musk is kind of advocating, the mass constellation of satellites for AI data centers. It becomes extremely ridiculous to a point that it can be a very good idea to excite some researchers, but it has no commercial viability.

What kind of reliability you have, and how would you roll out this? And there are so many cons to this. I want to nominate Garima for world emperor, because I like the world where we actually look at things like business model and physical constraints instead of just making stuff up and launching stuff into space.

Because you are 5,000X correct. I mean, everything you're saying, hallelujah. Wouldn't it be nice if we actually did basic business school stuff and said, "Is there a market for this?

" And I would say when it comes to something like a Starlink constellation, the answer is probably yes. I think that that actually could make some financial sense. But you're completely right, that data centers in space make no sense.

They don't make any business sense, but they also don't make any technical sense. And I have ranted about this before, and I will rant again if you want. But let's just stay on task here a little bit and say, as Garima said, there is no business model that justifies sending up many, many more thousands of satellites.

The business model argument for Amazon to buy Globalstar, well, that actually makes sense because they have a huge customer called Apple that's giving them money and has invested in them. And the same with Starlink. Starlink has been useful around the world.

I know people on boats and trucks and campers and stuff. Don't forget to mention the people at war. Exactly, yeah.

Great for war fighting. When they turn them off. But it makes sense.

That's actually a business model that makes sense because it's technically feasible. AI data centers in space, I just don't understand how we can have companies pitching something that is physically impossible as their business model and hoping it's going to work. I thought this is why he sent the Tesla up to space.

It was a test run to see if we could send these semis with the jet turbine engines up there too, to power these data centers. No? What a great idea.

What a great idea. I think Elon should have gone up on that convertible and seen how space works himself. Because you know what he would've noticed?

Sure. He would notice that it's not cold in space. It is a vacuum.

Yeah. Nothing. So some people, though, who should know better and are considered some of the leading lights in tech, it's not just Musk and Bezos that are saying this.

There's all kinds of people talking about data centers in space. So what's the fundamental issue, Stephen? Have they all lost their minds?

Yes. No, they haven't lost their minds. They're freaking lying to you.

Yeah. They're lying to everyone. It doesn't work.

Okay, if you put a hot thing in a thermos bottle with, not even a perfect vacuum, just sort of vacuum-ish, does it get cold? No. That heat doesn't go anywhere.

It just sits there. Space is like that, but a thousand times better. Can't we put fans on?

With what air? So in order to cool things in space, you need to use radiative cooling. A 40-megawatt data center would require 120,000 square meters of radiative cooling.

That is many tons of radiative cooling on a space bigger than the data centers that we are currently doing. " Okay, sure. We're putting that up there.

Yeah, we're going to put up a billion satellites, each one the size of a football pitch, and each one running one Grace Hopper. And that's only the cooling. Then you've got to think about radiation.

How do you shield this stuff? None of these chips are radiation shielded. They're going to get absolutely demolished in space.

That's why we use special processors on the space station. Hello, that's a thing. Then you got to ask data.

How do you transmit and receive data? Latency is massive- Yes ... up there.

How do you move data around? Then you got to talk about launch costs, and that's where the secret is. Who's pushing for this nonsense?

Two companies that own rockets. Why do you think they're pushing for this? They're the only ones who would profit.

Boy, Stephen, you're just killing us here with this sensible stuff. Come on, where's your imagination? Yeah.

It doesn't make any sense, and anyone who's put even a minimum of thought into it knows that it doesn't make any sense. And therefore, all these smart people proposing this, the only logical conclusion is that they are lying in order to make money. Fair enough.

I've got to call an end to this one today. You better, because otherwise I'm just going to explode. I'm going to end it with this, so.

There is no dark side of the moon. It's all dark. That's my ode to Pink Floyd.

We're out. Thank you for watching Techstrong Gang today. Garima, Stephen, great show.

Great stuff. Stephen, take deep breaths. Deep breaths.

You know what I'm going to do? I'm going to watch Networking Field Day this week. Fantastic.

We've got three days of great, incredible networking content. When does that start, Stephen? Wednesday through Friday, right here on Techstrong.

Every one of these companies is talking about products that exist and function and are technically feasible. It's great. I love it.

Tune in. Imagine that. All right.

Until tomorrow, this is Alan Schimmel on behalf of the gang. I hope you've enjoyed today's talk. I thought I had a lot of fun with it.

Stay well. We'll see you next time.