Semiconductors, AI Control Warnings and Data Center Pushback Collide

Hey everyone, happy Monday. What a Monday it is. They must be wild up in New York City today.

You saw our blue and orange Techstrong gang logo. Hey, go New York. Go my Knicks.

I've been waiting 50 something years, a long time for this to happen, so let's hope they do well tonight. But we've got even bigger news to announce today here on Techstrong gang. I'm going to announce it in just a minute, but first, let me introduce you to our gang members.

I know I left you hanging on that, but we've got a great gang for this fine Monday. We've got the one and only Barbara Roos joining us from out west. Barbara, good to see you.

Thank you. Also joining us, it looks from the background like I'm going to bet she's back home in Ottawa, our own Garima Bajpai. Garima, nice to see you.

Thank you. And then I have two special guests. Well, one's not so special.

He's been on a whole bunch of times. But the other one's pretty special. He's never been on before here on Techstrong gang, so we want to introduce him.

First of all, Brendan Burke, who is one of our crackerjack team of analysts at Futurum, and Brendan's beat is the semiconductor market, which should give you a little hint to what we're doing today. But Brendan, thanks for joining us here on Techstrong gang. Hey, gang.

Good to see you all. Nice to have you, Brendan. And then also joining us, he's a regular, though he's not as regular as he used to be, but our own Dan O'Brien, COO President of Futurum.

Dan, good to have you on. Nice to see you. Absolutely.

Thanks, Alan. All righty. And of course, we've got up in his home in the high castle, Mike Bizaard.

No Nick here, Mike? What's going on? No Nick here today.

Sorry, man, I used it twice already, and I didn't want to push my luck. All right. Well, now's not the time to change.

But anyway, let's jump into today's news, right? We kind of soft launched it last week. I put it out there just to make sure that the bugs were kind of worked out.

But I'm really, really happy to announce the launch of our newest Techstrong site today. This is the eighth or ninth site, and each one of them is like having a child, and it's a special thing. Our newest site is called Techstrong Semi.

com. And as the name implies, we're covering the semiconductor market. Why?

It's a good question. Let me introduce you to the godfather of Techstrong Semi, because truth be told, it was his idea. He brought it up to me.

It took us too long to get out. But Dan O'Brien said we should do a semi site. Dan, what was it, eight months ago already?

At least nine months ago? Yeah, sometime last fall. Time is flying, for sure.

But we're here. It's live. Tell us.

And you wrote an article we put up this morning. Daniel Newman put up an article as well. I've got an article.

They're all up there. But Dan, why don't you kick us off? Yeah, thanks, Alan.

Really happy to see this come to fruition. The conversation we had several months ago at this point was Techstrong is actually really well-aligned to Futurum. We really participate and play in the same markets.

We've got strengths in and around developer, cybersecurity, AI, infrastructure, et cetera. But as I looked across our portfolio at the Futurum group, kind of the one area that we played in in a lot of places within the business, but that we didn't have a real media site, was the semiconductor space. And as we've been bringing the portfolio together, building that synergy across the customer base, it made a lot of sense for us to tackle this next.

I think beyond that, though, it's just an incredible time in the industry. You can't follow the AI revolution without following Nvidia and the chip build-out and all the shortages and the geopolitics around US and China and export regulations. You've got this really consolidated market in terms of who's got manufacturing capacity, and it's just never been a more exciting or dynamic time in the semiconductor space.

And so, really happy that we're tackling this as a company. I think there's an intense interest in the semi industry that's really grown, probably starting with the pandemic, when some of those initial supply chain shortages were preventing things like automotives from rolling off the assembly line because of a 10-cent USB controller. All the way now to memory prices going astronomical, impacting consumer device demand.

I think $9 million per rack now with memory prices for the new Vera Rubin. So just a really interesting time in the semiconductor landscape, and with all of our portfolios, Six Five Media does an enormous number of CEO interviews with semiconductor leaders across the industry. We get a lot of the leading-edge chips coming through our Signal Six Five Lab.

And then the work that Brendan and his team are doing on covering kind of the new innovations that's coming, the data center models, the robotics models, all these end market models, translating all the way back from chip revenue all the way to wafer fab equipment. We've got a ton of great content on semiconductors from across the portfolio, and really happy the Techstrong team has created a single place for all of that to live, and a lot of news coming out day to day, really excited to see us keep up with the pace of the industry. Absolutely.

Brendan, as Dan said, you're the semi guy, right? Over at Futurum, the head semi guy, lead semi guy. What's your take?

Thrilled to have a new outlet to share my insights with the market and all of the forecasts as well as announcements we're bringing from the future and research side. Techstrong's created such a great community around tech. And the exciting part about the current moment we're in is that anyone in the tech industry can get into chips, and can experiment with accelerating whatever it is they do with the latest cloud services.

I think what's just been remarkable about this current wave is how it's gotten everyone back to where tech was, or how I heard it was in the '70s and '80s, when everyone learned how to program low-level languages, just experimenting with the latest PCs, as a brand new concept. It's like that today. Now that we have new chips coming to market, GPUs, XPUs, all making themselves available via easily accessible cloud services offering higher speed ups, greater access, different types of models.

I think everyone's back to student mode in the current market. And it really pays to learn about any of these new technologies because it just unlocks new capabilities, and we're seeing that new capabilities just translate into bigger productivity gains. And so that's directly made these new chip families grow rapidly from an investor perspective.

From a developer perspective, we're creating new applications. So, it's something to keep up with on a daily basis, and I think everyone's going to benefit from having a single place to do that. I love it.

Look, to put it bluntly and simply, semis are sexy again, right? For most of my career anyway, and my career might be a little longer in time span than some of you guys who are younger, semis weren't that sexy. It was what was Intel doing and who was copying them, and then the mobile chip stuff in the last 15 years.

But again, that was Qualcomm for the most part. It wasn't sexy. As a matter of fact, for a good part of my career, I think the tech industry specifically set about to do in software what was being done in hardware.

Right? You wanted to be able to, instead of using specialized hardware or putting the money in there, let's make software do it. Right?

And so the emphasis was on software and innovation there. But certainly in this AI era, innovation is in hardware and in semiconductors specifically. What I love about- I'll just jump back, Alan, for sure, but I also think it's also about software and hardware kind of co-development, right?

Yes. Really optimizing the full stack is kind of where the industry's gone. So, it feels like we've shifted back to hardware, but there's still a lot of software optimization happening on top of this latest hardware.

I would like to say that this revolution would instigate system on the chip. And when we talk about system on the chip, it's not only the hardware part, but you know how the software revolution started with community first engagements. I see that this would be the starting point for co-creation of system on the chip kind of revolution, where people can co-create and instead of buying off-the-shelf chips, they can co-create specialized capabilities on the chip.

And that would be very, very interesting for people like us who come from the software industry. Barbara? Barbara, you want to jump in here?

Yeah, first of all, I want to say semi has always been sexy because I grew up in the semiconductor industry at Intel, right? Oh, okay. So it's always been sexy for me.

But I love seeing the increased focus on this area from you guys. I have been kind of concerned for the last 15 years as cloud has risen, that companies have been pushing the easy button. Cloud has lifted the underlying compute, allowed them to not worry about what's under the hood, and that was great right up until it made compute scarce.

And now suddenly, companies need to be worrying about this. It's not just a technology conversation, it's a supply chain conversation, it's a geopolitical conversation, it's an energy conversation. And I worry that with the rise of the cloud, the knowledge and expertise to look at what's under the hood has eroded within companies, and it's time to get back to understanding the underlying technology and all of its implications.

I don't disagree at all. Brendan, you sit in an analyst seat here, right in the middle of what Barbara's talking about. Do you think we're seeing a renaissance and people kind of understanding the MIPS and BIPS and bites here of semi functionality?

Absolutely so. I think it's become the central driver of financial markets recently, which draws public awareness in a way that not many other phenomena can, just because you have the highest growth accruing to those companies that have the most pricing power. And pricing power is based on scarcity, and it's based on real physical industry.

And so that's become the direct connection between those trends Barbara was talking about, theA broader set of geopolitical news, and then the real economy that's built right on top of it. And I think that relates to everyone's life. It relates to commodities prices.

It relates to various sources of minerals and the availability of manufacturing capacity. And it's those basic economic factors that I think are influenced by just the state of markets every day. So I think it makes what can be a fairly abstract field of tech and brings it into something that is really tangible.

And so, I think no matter what area of the economy you're in, I think everyone's looking at how it's progressing, and assessing the rate at which it can continue. Absolutely. Brandon, what are you most excited about?

Because I can see that we're already moving past the two nanometer space, right? We're getting smaller and smaller, but maybe we're running out of places to get smaller. So what innovations, as you look over the horizon, are you looking at that go, "Wow, this is going to change the game"?

Oh, well, the chip industry is pushing ahead. There's both these new aspects in terms of the scale we're getting to from data centers that affect everyone on a day-to-day basis, just within the chip industry itself, that there's a drive to continue Moore's law, in terms of doubling the density of transistors. But that doesn't necessarily mean shrinking transistors anymore.

It just means stacking them more tightly together in advanced packaging. And it is that discipline that gets to engineering that certainly is, well, not conventionally sexy in any way, just in that it relates to new types of copper routing and power delivery, to be able to stack together wafers or substrates in a way that enables more transistors to get into the same package size. And, that is the new frontier.

It is an area that's delivering increased performance, within the same footprint, and enabling us to get to additional performance gains. And really underlying the entire build-out, because if you start a data center project today, you don't anticipate it to be approved and built for two to five years. And at that point, you have to bet on chips continuing to improve and it being worth investing in the next generation of chips.

And all signs from the industry is that full speed ahead, that the CapEx that we have allows us to invest in the new types of packaging. And that's ultimately where we're going to get to Angstrom class chips, down from that nanometer level you mentioned, and then ultimately to large scale XPUs, that are going to be able to run trillion parameter models within relatively smaller clusters than the 100,000 plus that we have to put in place today. I was at Cisco Live last week, and they were showing off what they call Cisco Silicon One.

Yeah. And their flagship chip there has a half a trillion, one half a trillion, just short, maybe 490 or 470 billion transistors in there in a chip maybe this big. And it just freaked me out, because I'm old school and I remember it.

I feel compelled, though, to do some publisher kind of stuff here and talk a little bit about the site and the structure of it. It's a fantastic site for semiconductor news, and today semiconductor news is tied to the cloud. It's tied to AI, it's tied to the data center.

So all of these peripherals will be on there as well. It's a beautifully designed site. The designer did a heck of a job.

I helped design it, but in any event, the important thing I wanted to make, this is the first site we've launched since we've become part of Futurum. And when I first sat down with Daniel Newman and talked about a joint future for Techstrong joining with Futurum, this was exactly the kind of stuff we spoke about, right? Bringing Futurum's expertise and allowing Techstrong to amplify it to market.

And that's what Techstrong Semi is. And the way we've organized it is look, most of the material, most of the content, you don't need to register. com.

However, because we have this deep bench of expertise, we are going to have premium content, not behind a paywall, but behind a regwall. You're going to need to register. We're going to need to know who you are, tell us a little bit about yourself, and we're going to have some really, really fine premium content that you're not going to just get on the corner newsstand of the internet around semiconductors with people like Brandon and all of the folk-- Dan is shy, but Dan's pretty deep into the semi world for his whole career as well.

So that kind of content will be in our premium content. Then on top of that, we are going to have access or at least a path to the Futurumanalyst content here in the Futurum intelligence platform that's available to subscription members of the market. And we'll have offers on there for individuals and for people who want to deep at that deep level, drink at that deep level of information.

It's not available today, but I'm telling you it's coming soon. So this for Techstrong is a very different hierarchy of information. Our pledge is we're going to give you all the information fit to print or to publish, but we're going to have some of it that is available as premium but no charge, and we're going to have some really, really valuable Futurum intelligence that is behind a paywall.

I'm telling you up front, I'm not hiding it. It was the design from the get-go. Dan, I know you and I have spoke a little bit about this.

What are your thoughts on that model? Yeah, no, I think we're really excited about it, Alan. There's a lot that we do that we want to be viewed by as many people as possible.

Our clients increasingly care about things like AEO and GEO, and the site does a great job in really getting the Futurum point of view out there to a large number of people. We are getting increasingly well trained into these frontier models. We want to be a net positive contributor of all of our knowledge and insights out to the public internet in a way.

But we also want to really develop a tight-knit relationship with our readers and our customers, and I really come back to the concept of community, Alan. We announced an acquisition just a couple days ago that we closed ETR, and that's really about community. And I think that's where Techstrong has always been really at its best, in my mind, is really building a community of practitioners, buyers, sellers, users, and you're really bringing them together to discuss what's going on in the industry, to learn from each other, to get a little bit smarter about our day job and about our passions.

So, I think that's really where we're trying to go here is to build that community where we have very timely, relevant market intelligence, kind of extend beyond the research organization, distributing this through some of our media properties as well. But really excited about the direction, Alan. So are we.

Thank you. All right. com.

We're a little over time, Mike. I apologize. I'll hand the reins back to you.

What's next? Well, we're talking about Anthropic, which late last week was talking about the need for a pause. They were saying that things are getting a little too dangerous with these AI models, and we should take a breath, and at the same time, they were also, with a couple other providers of AI models, appearing in front of Congress talking about, well, yeah, there's some biohazardous material that could be created with this stuff that might be weaponized, and that's not a good thing either.

" So what's going on here from your perspective? Does this make any sense? Yeah, it makes a ton of sense to me.

I think we're kind of in a bind. As the Anthropic co-founder put it, we've got all gas pedal, no brake. AI is approaching the point where it improves itself faster than humans can keep up.

In the same week, we've got CEOs of basically every major AI lab signing a letter, asking Congress to put guardrails around AI and biological weapons. And I think it reflects the reality that even while the very people building this technology are saying we need to slow down, they can't slow down, because if they do, a less careful competitor or a bad actor is not going to slow down, and it creates a dangerous situation that's even worse. This kind of ties to a piece that I wrote after the last AI field day about the concept of human in the loop.

I think a lot of people take comfort in the idea of a human in the loop, but I would argue that if the loop runs faster than a human can keep up, then it's a technicality that you've got a human in the loop. It's not meaningful. And what we actually need is humans at the helm, setting the direction and defining the constraints and able to say stop.

And we've got to get to the point where we stop wringing our hands about this and we do something substantive about it. Amen. I've used that human at the helm versus human in the loop metaphor as well, Barbara.

If the human is just the anchor weighing it down, that's not helping. But we've got to figure it out. So Dan O'Brien, what should be done here?

And here's my question, because part of my soul's, and I'll admit to occasionally being cynical, but it's a good thought, but at the other side of this thing is it feels like Anthropic is engaging in some sort of marketing program here where they're kind of advertising the fact that they have these advanced models by saying, and they're too dangerous for mere mortals to use, whether it's Mythos and now where we're going in the future of this stuff. And I kind of sit here and go there's all these bad movies I remember watching as a kid in the '50s and '60s, and there was always a theme that said, "Stop me before I kill again," and then the killer would run around and they would chase them and the killer wanted to be caught, but he couldn't be caught. And I feel like we're kind of having the same play here again.

So why can't Anthropic or OpenAI kind of stop themselves? Yeah, no, I think it's easy to look at the comments and feel they're a little self-serving. Anthropic trying to gear up for an IPO, essentially trying to create a little bit of a moat for itself by slowing down aggressive competitors.

I think it's interesting too, right at a time when Anthropic seems to be a little light on compute versus some of its other competitors too, which maybe is the one thing that could allow others to catch up. So I tend to be a little cynical on this one as well, Mike. I find myself agreeing with you a little bit here.

But I think they've got to look in the mirror a little bit. This is an Anthropic who wouldn't play ball with the US federal government on what kind of checks and balances were thrown into their program. That's led the federal government to go a little closer to OpenAI.

But I think you got to look at all of these companies and say, you've got to show the ability to self-govern, right? And I think there's been some positive steps there, like what happened with Mythos, kind of bringing it to the cyber agencies, bringing a lot of the cyber vendors in before these models got released to public. So, I do think that there's some good things happening there in terms of that self-governance.

But obviously a lot more work to do, given all the concerns that are being raised around biochemical weapons and that sort of thing. We don't need all knowledge to be democratized, I guess, is kind of where we're landing on this. There is a subset of what you'd call kind of need to know basis type of information that really does need to be kept in check with large industrial companies, governments, et cetera.

And I think that's the challenge. AI is democratizing intelligence so fast, we're kind of stuck asking, well, what are the few things we actually only want a few people to know and maybe not everybody to know? I want to also jump in here because I feel that there are two aspects of it.

One is the financial aspects and the aspects of why now and what needs to be done from a regulatory perspective. There's another angle to all this, is the ethical angle. And I wanted to highlight what Canadian government has done last week with this AI for all program.

They have launched this massive AI for all fund, which is also a reflection that there is one thing that AI is getting more autonomous and what is the critical threshold for autonomy? That's a question at hand, and how regulatories will act on it. But another thing is awareness, right?

And then connecting the Canadian government action on this is really commendable. I would like to also highlight that I feel that there is awareness versus capability leap gap, and that needs to be bridged in some way so that people are more aware of these kind of changes which are happening and what kind of capability these AI systems have. It can take a bad actor turn as well.

Wait, to Dan's point, though, and I'm going to throw this down, why can't we just, if you're Anthropic, so don't include the information about the hazardous bioweapons in the model and make that accessible to people and put some controls around that and prevent that from happening and take some ownership of this responsibility. What do you say? So there's two aspects of that, and I'm going to hit them both.

First of all, this comes down to a problem that we've had way before AI and AI chips and everything else, and that is the why we can't have nice things on the internet, right? Because there's bad actors in the world. They're not all pure.

There are bad people and bad actors and bad nations and not good stuff. Your question, Mike, is perfect, and it goes right to the example that Barbara brought up. AI offers us the promise to take your tumor, if you have cancer, and instantly analyze it and formulate an mRNA kind of vaccine that's tailored to your bio, to your DNA, and maybe cure things like pancreatic cancer, ovarian cancers that have proven the hardest things for us to do.

Right? Manipulating DNA to cure Down syndrome or other sort of inherited kind of genetic diseases. But it's that same technology that could be used to create the bioweapons that Barbara's talking about.

Same technology. How do you just put the good stuff out without the bad stuff? How do you give the good stuff just to Pfizer or Lilly or take your pick of pharma companies and it doesn't get out the back door, or they take the Chinese or the North Korean.

Barbara, I'm sorry, go ahead. Oh, no. I'm sorry.

I didn't mean to talk over you. But I think we need to stop thinking about controlling this at the model level because there's always going to be other models that compete. And so if Anthropic throttles their model, you'll have a bad actor model that will go after this and offer the same capability.

But where I think we need to focus, we got a hint of this with the bioweapons example, where they're asking for laws that leverage the choke point of where people order these DNA samples. So we need to look at what is the choke point in AI, and it's the hardware. That's where the constraint is.

So can the Nvidias of the world build in hardware level mechanisms on the chip to control what's being done on those chips? Also, can they control the supply chain of the chips itself? They decide who gets to buy their chips, right?

There's a lot of competition for them right now. This sounds like a Brendan question. So how about controlling the flow of who's able to buy those chips, and then with the cloud computing provider, they control who they sell their compute to so that bad actors aren't getting access to it That hasn't worked so well with the whole Super Micro thing, Brendan, has it?

No. No. Me neither.

But let me bring my second point into this, and it actually dovetails nicely. I saw a documentary on Peacock this week, and if you get a chance, check it out. " I don't know.

"I Was an AI, The World Is Ending" person. Whatever. It's a documentary by a guy who's having his first child.

Him and his wife are having their first child, and he's worried about will his child have a job? What will AI-- what will the world look like by the time his child grows up? And he got Sam Altman on there, Dario Amodei on there, a bunch of others, the guy who started DeepMind, from Google now.

And they all agree with what Darren Amodei said, that we need to reflect. I don't know if we need to pause, because the problem is, if you're Darren Amodei, and you want to go public, and you put a pause, and Sam and them go forward, and you lose market share, that's called fiduciary duties that you breached, right? You've got duties to your shareholders to make as much money as you can, to be as profitable, to do that.

So it's not easy for one actor to do it. It has to be done more on an industry basis. And there are certain things that the government is made for.

And those who watch this show know I am not a huge fan of our present administration. But when they do something right, you got to say they're doing something right. They're talking about calling in the AI industry and taking stakes in these companies.

And again, I'm not a socialist either. I'm not for government ownership of companies. But this is such a big moment.

If we're going to control this and use this for the best and not for the worst, this calls for government intervention. It calls for government involvement that crosses individual companies industry-wide, worldwide, even maybe more than one government. Maybe it's a consortium of governments.

Yeah. But if we're talking about changing fundamentally people's lives in terms of jobs, in terms of healthcare, in terms of the fabric of our society and civilization, you got to have the government involved. And I'd welcome them taking a stake in these companies and putting some industry-wide stuff in there.

Can I make a suggestion? Sure. Wouldn't it be awesome if we spent as much time trying to figure out how to make these damn weapons as maybe we should spend more time trying to figure out how to prevent them from actually going off in the first place?

And maybe AI can help us with that. And there's the whole- Maybe that, too. But it's not going to happen company by company.

You need something bigger. That's the reason why we have government, right? The interest in the government taking stakes in these companies is actually more of a way for it to fund UBI than necessarily a government saying it.

A government saying- Eventually, you know what, Dan? That's definitely one of the out processes. If they're going to be that disruptive and that much value is going to- You're going to need a UBI ...

as far as like how is the government going to fund this- Yeah ... this UBI? But it goes to fundamental like John Locke, the wolf man versus, I think it was Locke or Rousseau, this is, right?

Of the two different philosophies of why we form governments, why we form civilizations. If an individual could do it all, he doesn't need a government. But there are some things you need a government, you need a society, that kind of fabric for.

AI's one of them, regulation, not killing ourselves, all the above. Are you saying we need an AI social contract? Is that where you're going?

Yeah. All right. I'll go with that.

But I applaud, I want to see what comes out of this. It would be interesting to see what we pay for that stake and how the taxpayers get paid back. Anyway, hey, we're over time on this particular topic.

Mike, what else do we have? Well, let's shift a gear here to something simple-ish. But one of the issues that's been going on, and I don't think it's news to anybody on this panel, is that data centers are not especially popular with the general public, and they are kind of upset about what they're doing to a variety of things, including the environment and water consumption specifically.

Microsoft is out showing at least that they've been thinking about the project and the issue, and they have some new approaches to the way that water is run through the data center to cool the processors and everything that goes with that. And yes, Brendan, I noticed there's a theme here. Everything is your fault at the end of the day.

But as you kind of work this through for a minute, is there advantages and things that can be done here where we can make the data centers more environmentally friendly, and we just didn't think this through enough, but there's opportunity here. What do you say? Well, I think there's a couple aspects there on the popularity versus the environmental impact side.

I think the polls showed that people don't want data centers in their backyard, to impacts they might have on electricity rates or air quality in the case of gas turbines. So that's one feature. And then there's the water topic particularly and resource usage.

And I think both point towards more expensive data centers. The idea that we can easily and effectively build out gigawatts of local gas turbines and just create a new industrial age of smog-emitting power plants, it just isn't the way the economy works anymore. And-But that means that if we're able to build a data center for 10 million per megawatt today, that might be going up.

And I was over at Computex Taipei last week, where Jensen ran up the numbers in terms of what it's going to cost to build data centers to now 100 million per megawatt all in, which suggests that maybe instead of 10 million, just for the construction and the energy, it could cost 40 million. Why? Because we're siting it in places that maybe are farther away from people.

Maybe they're closer to large arrays of solar panels in the desert. Maybe they have done more advanced environmental permitting and reviews beforehand, so it's built in a more thermally efficient way. There's an immense creativity needed, and every site is a little bit different these days in terms of where it is, how close it is to people, what power it has access to.

But the cheap sites, the ones that are right next to a grid interconnect and are on an abandoned lot outside a city, those are the ones that might not have social approval to go ahead. Water topic is a good one as well. And I think Microsoft laid out there that their data center uses as much water as a fast food restaurant themselves, which has been a fairly common comparison in the industry.

That's an aspect where just great reuse of water within a cooling loop can be really effective. And so, I think there's a number of barbs that get levied against any rapid build-out. It's good to throw a bunch of stones and see what lands when it comes to public opposition.

And I think we're getting towards the real heart of the matter, which is that people don't want a belching smokestack down the street. And that's really going to force a reconsideration and ultimately realign how these plants get built. Dan O'Brien, is that what's really going on here?

Are we building AI smog towns, and we did that just because we wanted the cheapest way we could figure out how to build AI, and now we need to go revisit this whole conversation and say, what's the true all-in cost for developing an environmentally friendly approach to data centers that doesn't wind up having all our kids choking on smog? I think the fundamental issue here is that the costs are borne locally, but the benefits are spread kind of widely, right? Normally, you put up a factory in a town, it's actually probably going to be an economic boom for that town.

It's going to employ people for generations. I think we're finding with these data centers that there's probably a really big power impact locally. Rates are going up.

Jobs, well, maybe during construction, but after, a lot less so. These are fairly autonomous data centers. They don't need a huge labor force, so it's not a massive job creator.

And like I said, all of that cost of creating these things are really borne by the local community, whereas that compute is being kind of spread all over the world by the power of the internet. So, I think these things tend to work when there's a little bit more benefit locally, albeit maybe some costs as well. But that to me is the fundamental issue here.

The other thing I think we haven't gotten around to yet that I feel like is coming for us is going to be kind of the electronics waste recycling, right? What happens when these facilities are five, six, seven years in, and we're kind of through the depreciation phase and all of these racks need to get ripped out and recycled? Think about the rare earth elements in there.

I feel like there's actually an opportunity for us to get ahead of that problem in a big way, but I haven't seen any dialogue on it yet. I think necessity is the mother of invention, right? How important are these AI data centers to our future?

Well, if they're that important, we'll figure out a way. I don't know if we'll ever get it where it uses the same amount of water as a McDonald's. But I applaud this kind of thinking, though, of trying to get there.

When I was growing up, I never went to LA back then because we lived in New York, and we never went there. But pictures of LA, Mike, you know this, right? You couldn't see the mountains.

It was always smog town. Dude, I'll give you two words closer to home, Gowanus Canal. The Gowanus Canal, yes.

For those of us who know Gowanus, yes. Well, the Hudson River, for that matter. There were no fish in the Hudson River when I was growing up.

Right. Now, striped bass spawn there like mad, and it's a beautiful place to go boating. Even right in New York City, right in the harbor.

It's the same thing here. Yeah, it's first generation, we were just building. We will refine, we'll get better, we'll figure out how to make do with less water or other liquids.

Look, this whole thing, and this was part of this documentary, again, if you've got Peacock, go check it out. The whole AI revolution is premised on cheap energy. It's not fossil fuel or coal.

If we can't figure out the cheap energy piece of it, it's going to be a big inhibitor to doing what we can do here. Whether it's solar, fusion, fission, nuclear, or whatever, we need to crack the energy piece of it Yeah, and it connects back to the last week discussion, Alan, where we were discussing about this elemental fund, which is a nonprofit organization. They have started to look at core startup ideas on low carbon footprint material, cooling technologies.

So there is a lot of research, which is we are inviting that research. It's just that how fast we can catch up and how much funding is available and interest to kind of steer the needle in the right direction. Let me bring this full circle here.

This is a huge opportunity to do what Dan O'Brien was talking about, where if the United States government is going to play a role in this stuff, then maybe they can help fund the research and development of the technologies we need to do this cleanly. I think there's enough incentive for private industry to fund that. I don't want my tax dollars going for that.

But that being said, let me bring this full circle back to semis. What can we do by doing more of the AI compute on our devices? So we've been talking about GPUs in these data centers, but Brendan, what's happened?

I know today, it's probably right around now or just starting, the Apple conference. What can we do about AI chips on our devices? Well, it goes to the shrinking of power, performance, and area of chips that Mike asked about earlier.

That when we shrink these chips down more, we don't need to convert old clothing factories into data centers. We can have them be much more modular, closer to where the compute is needed. And so the goal is to get it closer to our phones.

It was the entire premise of Computex last week, as it was with Dell Tech World the week before, is that models want to run closer to where the users are, and we're going to run more and more of them on our workstations with whatever power we already have available. And it's going in that direction. Obviously, it's not one shot in that direction, flip a switch and it all runs on the phone.

I think we've seen Apple really struggle to get AI working as well, and we've got scaling laws in place. That's the countervailing factor here is that models get better the more energy you throw at them, and that hasn't been disproven yet. So we can't turn that boat around at a snap of a finger, and we're not going to be able to do any of the great things we want to do on our phone.

I think emojis work pretty well. You can make your own emoji and change your face into some cartoon character. That's pretty effective.

But I think the reality is that the devices that can actually run a model with any kind of capability that can do anything beyond what we're already capable of doing are probably going to cost more like 20K than one K in the current memory environment. And then we're going to have to get new chips that can fit more model weights in a smaller area until hopefully it does fit in a phone footprint. But we're not there yet, as long as you actually want to use frontier AI.

Yeah. And so maybe Apple will pitch that they have it yet again, but I think the fundamental constraints mean that we're going to need big chips still. Yeah, but this is, again, starting off another revolution where you will see that the telecommunication companies are investing on the edge when it comes to AI, right?

So if you talk about devices, that's great because it also has a monetization aspect, like who's paying for this kind of technology? Users would be paying for this. No big providers or telecom providers would be investing in large CapEx solutions, right, in the data center.

And if the users are out paying for it, it's still on the edge. So you'll also see not only device centricity, you'll also see radio centricity and the layer one centricity. So I think it's kind of exciting in terms of how this edge capability in telecommunication is building up.

All right. So I would just say, I'm hoping that we don't get to the point where we're going to have to create an AI super fund so that we can clean all this up. Ugh, always the cynic.

Anyway, we're overtime here. I got to put an end to it. But look, if you haven't picked up the thread running through all this, AI can change the fabric, the weave of our society, of our civilization.

And we're making choices right now and confronting issues that really us, our children, and maybe even their children will live with the consequences of it. So nothing too important. We'll continue talking about it.

com, as well as Techstrong AI, Techstrong IT, Digital CXL Platform Engineering, all the rest of the Techstrong sites. Watch it here on Techstrong TV. com is a great asset, too.

They actually, more than most analyst firms, do publish a lot of their research, and it's not even behind a reg wall, let alone a paywall. So if you're interested in this kind of stuff, it's also a great asset. Until tomorrow, though, Brendan Burke, thank you for coming on Techstrong Gang.

I hope it won't be your last time here. We'd love to- Great to be here ... make that a regular here.

Dan O'Brien, always a pleasure. Barbara, I haven't seen you in a while, but I know you've been here and busy with Tech Field Day. Thank you.

Garim, always. Mike, great job as always, man. Sure thing.

Until tomorrow, everyone. This has been Techstrong Gang. Thank you.