// transcript — 1209 segments
0:00 Bestie intros!
0:01 All right, everybody. Welcome back to the number one podcast in the world. Of
0:05 course, that's the All-In podcast. I'm your host, Jason Caliganis. With me
0:09 again, your chairman, dictator, Chimath Polyhapatia, and the Sultan of Science,
0:14 David Freeberg, David Saxs, will be calling in from the skiff.
0:18 >> He's in some deep negotiations uh for the United States of America. From the
0:21 skiff. >> It's not there's no tea at the end. It's just
0:23 >> tea. No tea. >> Oh, skip to my l Okay. He's in a skiff
0:28 >> doing something with his Blackberry and a bunch of generals. Nobody knows what's
0:31 going on in Sax's life, but he he'll he'll crack in from the skiff any moment
0:35 now. But we'll start with >> You guys see that Pete Hex have
0:39 announced a PT and fitness test for the generals? Could you imagine if Sachs had
0:42 to pass a PT? >> Oh my god. They should totally make it
0:46 for the administration. Sax, we need you to do a one push-up. Sax,
0:50 >> what do they do if you don't pass? They remove you from your You probably get a
0:53 cure period. >> We should do a push-up contest. That would be great. Winner take all. How
0:58 many push-ups can you do? Free birth. You have to adjust for people's heights.
1:01 I'm I'm the tallest of all of you. I have a much longer limb system.
1:03 >> What does that mean? >> So 20 for me is much harder than 20 for
1:07 you, Jason. >> I mean, 20 is easy for me at this point.
1:11 >> Yeah. You're like Bill Bull Baggins. It'll take you like 8 seconds to put
1:15 >> a tank, man. Bill Baggins. How about Thor? I'm like Thor at this. I'm going
1:18 to my Daniel Craig era. >> Kell has a elbow to my Daniel Craig
1:23 >> weight to weight ratio. That's highly advantaged. >> All right, let's get started. Enough
1:28 shenanigans. EA is what is your arm length, Jake? Do you have a good arm
1:29 length? >> My wingspan, my wingspan uh technically is enough to kick your ass with one hand
1:38 and tie behind my back. That's actually let your winners ride.
1:46 and >> we open sourced it to the fans and they've just gone crazy with it. Love
1:53 EA acquired for $55B in biggest LBO ever, why PE is in trouble
1:53 [Music] Okay, EA is being taken private in the largest takeprivate deal in history. $55
2:02 billion. Man, that just stacks up to, let's see, Texas Power Company in 2007, HCA
2:10 Healthcare, $ 33 billion. This is a large deal. Investors in the take
2:17 private include Saudis PIF, Silverlake, and Friend of the pod, Jared Kushner's
2:22 Affinity Partners. 210 bucks a share, 25 premium on the stock. Kushner's largest
2:28 LP at affinity as you know, SAP PIF as well. The PIF has invested over $900
2:33 billion when you know many of the things Lucid Motors, Live Golf, the Softback Vision
2:39 Fund, Uber back in the day, Newcastle, the Premier League, Electronic Arts
2:45 obviously is in the video game business. They were founded at uh Sequoa's office
2:50 in 1982 in San Mateo. Shout out to our guy Rudolph Botha who joined us for the
2:54 all-in summit. Their headquarters still in Redwood City Madden NFL
2:59 the Sims. Oh, that's why you have the background of the Sims this week. Need
3:03 for Speed. Pretty insane deal here. Chimoth and this is a high
3:08 watermark for private equity. Anyway, you look at it and the Pif loves games.
3:13 They are the biggest shareholder in Nintendo, Savvy Games, Scopely. I mean,
3:17 they just keep buying games. Uh, what are your thoughts here on this deal
3:22 happening right now? >> I really like it. Let me give you the
3:26 bull case and then let me give you what the bare case would have to believe. The
3:31 thing to remember is that video games is the anchor pillar of usage across the
3:36 entire internet. Last week at our poker game, we had Matt Bramberg
3:43 join in just for dinner, who's the CEO of Unity, and Alex Blum, who's the COO
3:48 of Unity. And one of the stats that they shared with us at dinner was it's about
3:53 3 billion DAO play games >> which is just an inc exactly it's an
3:59 incredible incredible stat. So in many ways it's much bigger than social
4:03 networking and social media or as big and in that EA is sort of this 800 lb
4:08 gorilla but I think the problem is is that they've always been these
4:10 gatekeepers and I think that there's a risk and a chance that these gatekeepers
4:15 get eroded away. Specifically who I'm talking about are folks like Microsoft
4:19 and Xbox. And at the point that this company is going private, there's some
4:22 really interesting things that are happening. So Xbox, I think the day
4:27 after the EA deal got announced, decided to hike prices 50% to their subscription
4:32 service. And what happened over the subsequent few days is that so many
4:36 people tried to cancel that the site went down. So what are you seeing
4:40 happening? You have distribution gatekeepers trying to raise prices and take share.
4:46 And then you have the original IP owners who have not had a a wellfunded way of
4:52 fighting back in a category that is basically as important and frankly more
4:57 important than social media. So I think if you take an asset like this private,
5:02 it allows you to take your time to clean up the opex model, right? figure out who
5:08 does what, be able to use the best of all these nextgen tools,
5:13 and then be able to find ways of finding distribution outside the scope of Xbox
5:17 and PlayStation so that you can take more of your share. If you do those
5:22 things, this is a multiundred billion dollar asset. And in that, I think it
5:26 could be just an enormous win. So, I think it's very smart. What's the bare
5:31 case? I think the bear case is extending a theme that I've talked about here a few
5:40 times which is I think the value of patents and by extension IP and
5:46 copyrights are going to go away and in that there's going to be a
5:52 spectrum where certain content IP holders lose and other ones win. I think
5:56 gaming is on the winning side to be honest and I think content studios in
6:00 general like traditional content the Disney's the Hulus the Netflixes are on
6:06 the losing side but the bare case would be that these tool chains
6:13 allow the number of games to be built to increase by two three four orders of
6:19 magnitude and that they are distributed by other places like the social media
6:24 sites. I just think that that's a pretty low probability. So on balance, I think
6:30 that Jared and Egon did a killer deal. I really like it. And for people who don't
6:35 know, Unity makes the 3D software that people build games in. It's a public
6:39 company, 16 billion, also backed by Ruof and Sequoia back in the day. Incredible
6:44 company. Freeberg, what are your thoughts on the gaming industry versus say social media
6:54 versus traditional media? We're seeing massive amounts of money being put into
6:58 each of these, but this is time and for this next generation, let's say
7:01 millennials and younger, we're seeing a big mix. Obviously, they don't have
7:06 cable TV, so that's been plummeting, but they do play games. They do like the
7:11 YouTube, the Tik Tok, etc. And they do love social social media. What what's
7:15 the future here as you see it? >> One way to answer that question is to
7:19 think about how people spend their time. Do you spend more minutes on social
7:26 media or on traditional media or playing games and how is that trending? But
7:30 importantly, which of those will acrue more benefit and as a result drive more hours spent
7:38 from AI? Is AI going to create more social media engagement? Is AI going to
7:42 create more traditional media engagement or is AI going to create more video game
7:46 engagement? And I think that one way to kind of think about this thesis is that
7:49 AI is going to ultimately acrew to video game entertainment far more than social
7:54 media entertainment or traditional. >> Why is that? Why? Explain to me.
7:57 >> Because I think you can create dynamic, more engaging experiences that will
8:03 benefit from kind of a back and forth sort of relationship than you can with
8:06 traditional content or with social media. And what we see now in a lot of
8:11 gaming systems that didn't exist, call it 12 years ago, is AIdriven players
8:17 embedded in the games that act and feel a lot more like real human engagement.
8:23 That is very hard to kind of mimic from traditional programming methods that
8:27 were used in gaming. And so that makes a a big difference. Like for example, if
8:30 you're playing Fortnite, I don't know if you guys play Fortnite or have played
8:33 Fortnite, but if you're a noob in Fortnite, like you're early player in
8:37 Fortnite, you're mostly playing, even though you go online and play against
8:40 what are supposed to be kind of other players, you're mostly playing against
8:45 AI because what they do is they tune the AI to be easier to beat so that you can
8:49 slowly develop your skills. Because what was happening early was they were seeing
8:52 a high degree of churn in Fortnite because kids would go on and play for
8:56 the first time and they'd get paired up with kids that were better than them and
8:59 so they would never win and they would get frustrated and they would quit the
9:02 game and stop. So the churn rate was high. So AI unlocked higher engagement
9:06 and higher retention on the Fortnite platform. And I think we're seeing that
9:11 in a lot of different gaming platforms now. So AI can be used for example to
9:16 maximally increase time, engagement, satisfaction, happiness. I think the the
9:20 Saudis saw this and if they're trying to diversify away from their oil holdings,
9:25 entertainment, and how people spend their free time, which by the way, I
9:29 think is a general macro bet that everyone should consider making because
9:32 if you believe in AI and you believe in the improvements in productivity,
9:36 generally speaking, people in the industrialized world will generally have
9:39 more free time on their hands and be able to support themselves with the
9:44 deflationary effects of AI over time. So if there's more time on people's hands,
9:47 the general market for entertainment is growing. And if the general market for
9:51 entertainment is growing, gaming is the future of entertainment. And the future
9:55 of gaming is AI. Now the the Saudis own 10% of this prior this company prior to
10:00 the deal. And I don't know if you guys have tracked the investments they've
10:03 made, but they've been extremely aggressive with gaming. So they have
10:06 this like investment division called Savvy Games. And within Savvy Games,
10:11 they bought Scopely for 4.9 billion in 2023. And then earlier this year, they
10:16 spent 3.5 billion to buy Niantic, the company that makes Pokémon Go. And then
10:21 they also own 4% of Nintendo. They own 6% of Take 2. They own a sizable percent
10:27 of Activision, Blizzard. So they've put quite a bit of capital in small
10:31 investments in other gaming platforms. They own a few gaming platforms. So this
10:35 is clearly like a big thesis and a big investment that they see as the future
10:38 of entertainment over time. Jared's firm, Affinity, is going to own about 5%
10:42 of the company post transaction. the Saudis are going to be the majority
10:44 owners. So, I think that this is going to end up being the next big platform
10:51 play for them and and it allows them to make the important long-term investment
10:55 in furthering the transition to AI and not have to worry about quartertoquarter
10:58 earnings, but really making a 10-year bet and they do talk a lot about this
11:02 2030 vision. >> And if you look at across those three categories we've been discussing here,
11:08 video game usage about 60% of US adults do it every week. Social media about 75%
11:13 of Americans use it every week and uh streaming, traditional media, the
11:17 Netflixes, Disney Pluses of the world, that's still 83%. So these are the three
11:22 buckets of of people's time. Uh books and going to the movies, those are
11:25 obviously the big losers. >> You know mix the market was totally
11:30 getting this wrong because the Tik Tok of the deal is super interesting. When
11:33 they were looking for the debt financing, it was about 36 billion of
11:38 equity, 20 billion of debt. They called Jamie Diamond and Jaime basically ripped
11:46 the 20 billion in on the same day just because I think I think he also
11:50 could underwrite this pretty fast. I mean, some of the biggest deals are
11:54 frankly so obvious that it just takes the courage to put it together and then
11:57 everybody's like, "Oh, this just makes so much sense." And then Andrew Wilson,
12:01 who's the CEO, is going to stay on. He's a great guy. Super super compelling.
12:06 It's worth talking a little bit about the impact I think of private equity. If
12:10 um you spend any time in the region, I'm going to be in Saudi and Dubai in the
12:15 first week of November doing my founder university and I'm I'm been out there
12:18 twice a year maybe for the last three years. They will tell you whether you're
12:24 in DOA, Abu Dhabi or Riad, we've got six or seven industries we really care
12:28 about. Technolog is at the top of the list. Private equity is at the top of
12:32 the list. Live entertainment and sports at the top of the list. And then
12:36 actually hospitality also at the top of the list. Real estate building new
12:40 places for people to go. And if you look at private equity, pull up that chart I
12:45 had there. This is just stunning how big this industry is getting. You know, $5
12:51 trillion is what we're up to here. And it just keeps growing.
12:54 >> I I think private equity is totally screwed. I I don't think Silverlake or
12:59 Infinity or this deal are screwed, but I think private equity
13:02 in general is totally owed. >> All right. Right. Well, it's it's gotten
13:08 huge just since 2015 and tripling in size. So, why is this I guess my
13:13 question for the gentleman here and for the audience, why is private equity
13:18 becoming so large and what impact does that have on society? If people can't
13:25 put EA into their retirement account, they can't put Stripe into their
13:28 retirement account. If we take all the great companies and we start to
13:31 privatize them, SpaceX, let's say never goes public. What impact does that have
13:34 on people's retirement accounts? >> Okay, look, I think I think the history
13:38 of this is important. There was a long-standing belief that the best way
13:45 to generate the best risk adjusted return, what does that mean? That means
13:50 to manage through periods where the stock markets go down and to manage
13:54 through periods of volatility. The best way to do that was to have what's called
14:00 a 60/40 allocation. 60% to bonds and 40% to equities. Over many years, especially
14:06 when we artificially suppressed rates at zero through Obama, a lot of people
14:11 started to move their allocations away from 6040 and they started to make more
14:16 and more investments further out on the risk curve. The biggest beneficiaries of
14:20 that were venture capital, private equity, and hedge funds. The thing with private equity is that
14:28 because rates were zero, they had an infinite amount of borrowing capacity,
14:33 had very little downside to them, and so they were able to manufacture returns
14:37 much faster than venture capital and hedge funds could. So, as a result, you
14:41 had an initial group of people that were defining the asset class, making a ton
14:45 of money, and then you had all these fast followers that said, "Well, if
14:49 they're doing it, I can do it, too. So far, so good." But then always what
14:54 happens is then you have this flood of lagards that just flood the zone. And
14:59 it's these lagards that make it very difficult to generate returns because
15:04 they start overpaying for assets. They start mismanaging and undermanaging the
15:10 assets that they do own. And so where we are is that private equity has seen a
15:14 very consistent way of returning money to help improve that 60/40 portfolio. as
15:20 a result they got a lot of money but then that created a lot of competition
15:24 and so that's why you see this hockey stick graph Jason and when you see that
15:27 kind of graph >> it doesn't matter what asset class it is
15:31 the returns go to zero >> and so we've seen this in venture
15:33 capital >> we've seen this in hedge funds >> and we're now going to see this in
15:38 private equity >> too much money going in to be clear what
15:44 you're saying means you kind of exit it right there's there's no returns and so
15:47 again I've said in any of these alternative asset classes, there's only
15:50 one thing you should always ask if you had to have one critical question.
15:57 What are your distributions? Don't show me your IRRa. What is your
16:02 DPI? >> The distributions on your paidin capital. And if the answer is zero,
16:09 then it is a very challenged asset class. And what I will tell you in
16:13 private equity is that over the last four or five years distributions have been few and far
16:20 between. So I think what's going to happen is that the money is going to come out of
16:24 private equity and it's going to get concentrated into the few companies that
16:29 know what they're doing of which Silverlake has generated over you know
16:35 the last 15 20 years tens and tens of billions of dollars of
16:38 distributions. They are just an exceptionally well-run organization.
16:44 They've done these huge buyout deals successfully before. So, we need to go
16:49 through that in PE. Where does the money go? The money's already leaked into
16:53 private credit, which is the next big bubble that's building. It looks like
16:57 this chart that you just showed, which is loaning businesses money. You
17:02 know, it's super interesting because you make such a good point. What we're
17:05 seeing in private equity is these continuation funds. Now continuation
17:10 funds are coming chimoth to venture. So I've been getting pitched on these
17:12 continuation funds where like hey take all your assets sell it to a new group
17:17 of people and then reset the clock and then there's never an exit. The good
17:22 news is I will say the last year we've seen a lot more activity for shares of
17:28 our companies that are still private. So the secondary market Freeberg is coming
17:33 back in a major way. But I do get worried about these continuation funds
17:36 because now you're just moving an asset from one class to the other and we need
17:41 to have a functioning IPO market. How functioning is the IPO market today?
17:42 IPO market, SPAC 2.0
17:45 Would we say it's completely dysfunctional? >> How dysfunctional is the IPO market? Let
17:51 me say it another way. And and how do we correct that? And this leads into your
17:54 new spec. >> Look, there are three ways to go public. There's the traditional way IPO,
18:01 there's the direct listing, and then there's the reverse merger or the spa.
18:08 Up until I floated IPO A in 2018, I think it was the first way was really
18:13 the only way. I was involved in two direct listings, Slack and Coinbase.
18:21 And in both of those, what I learned is that, you know, it has the same vagaries
18:26 as the traditional IPO. So in the traditional IPO, you go to a bank, they
18:29 underwrite you, they act as a gatekeeper, and they take six, seven, 8%
18:35 fees as a result, and then they allocate what is essentially underpriced stock to
18:41 their best customers. Then you see a one-day pop, maybe a two or three day
18:45 pop. All of those customers tend to unload and then the stock tends to drift
18:51 down. So the IPO is expensive and it typically is mispriced.
18:55 The direct listing you have a different dynamic which is the first trade is always the highest
19:00 trade and then it just goes straight down. That happened with Slack and it
19:04 happened with Coinbase. So >> Spotify would be in that group as well.
19:05 Yeah. >> Yeah. With Slack I remember like I I was like offside a billion dollars and I was
19:11 like well I'm never letting this happen again. And so when I had the Coinbase
19:14 thing, I sold it the first day. And I texted Brian. I said, "This is not a
19:17 directional indication of your company. It's the dynamics of the direct listing
19:21 because I learned it the hard way that the time to sell is on day one." So
19:27 where does the spat come in, you know, especially now in version two? Version
19:33 two being the the thing that I have been tinkering and refining with and am
19:39 trying to push in in this new version. I think that it's creating an incredibly
19:44 competitive vehicle where you can have a ton of money go into these private
19:48 companies, take them public at a very, very low cost of capital. And I think
19:54 that that's should be very enticing. >> So, you closed your financing. Can you
19:57 just tell us what the capital raise was like as you went out and met with folks?
20:01 What do you hear? >> Yes. You know, Nick, maybe you can find
20:05 it. You know that image of the Raptor engines? >> Yes. super complex to being elegantly
20:09 simple. >> Yeah. Nick, can you can you maybe just throw that up? What I would say is like
20:15 Spack 1.0, of which I was, you know, right in the front of the parade, had a
20:19 bunch of misfires and it was complicated, but it worked. There were
20:22 some hot fires that worked, but then there were some clear misfires. And the
20:27 whole point was to prove that you could create a competitive alternative to the
20:30 IPO. The thing that I'm the most proud of quite honestly is for all intents and purposes I started
20:38 a normalization of this vehicle that's now raised more than 1502 200 billion
20:43 dollars for American companies. I am very proud of them. That's an important
20:47 thing for the American capital markets. I think what we did in American
20:53 exceptionalism is Raptor 2. It's not yet perfect, but I do think it tries to
20:58 improve on the things that I noticed was not working in Raptor 1. And in that is
21:02 a lot of the compensation and incentives. And so when I showed that to
21:08 investors, they were quite excited. I think that they want a competitive IPO
21:14 market that brings many, many American businesses to the public market so that
21:19 they can be owned by everybody. the transparency they like and the fact that
21:23 the incentives are such now where there's absolutely no compensation
21:27 unless this thing really works. >> And historically they received warrants
21:32 in the company typically with a strike price of 1150. So 15% above the issue
21:35 price of the stock >> and founders shares that were basically
21:38 >> and there was founder shares but like did you have a reaction from them saying
21:44 hey we want some warrants we we need a little extra kicker here like there's
21:47 some sort of desire for that? No, in fact it was the opposite. I think that
21:52 the institutional investors and you know my investors in this 98.7 of the capital
21:59 was allocated to these guys are the best of the best. You you know who they are.
22:04 So they're every single blue chip A+ institutional investor. And what they
22:11 wanted was great companies. They want great companies to be public. And the
22:14 reason is the thing that Freeberg I think you mentioned this before. When a
22:18 good company gets public, the amount of money that they can raise in the publics
22:22 and then the amount of growth that they have in the publics far outclasses what
22:27 they'll ever do as a private company. And so they want the simplest and
22:34 cheapest way of great businesses to get out. Jamat, do you think that the
22:38 transaction when you find a merger partner, the traditional spa has been
22:45 announced as a merger concurrent with a pipe being done where new investors are
22:49 underwriting the valuation of the deal and saying we like this company at this
22:53 price cuz we are now going to write money in in the form of a pipe and
22:58 historically the pipe was for common shares. So it kind of was like this is a
23:01 good price and everyone felt good about it. Number one, do you anticipate that
23:06 there'll still be a pipe being done in concurrent with the merger in this
23:10 transaction? And then number two is do you think it'll look like a common pipe?
23:15 Because after the spa frenzy died down, in order to get deals done, the pipe
23:18 started to get done with convertible preferred securities. So they were
23:22 senior to common and they almost were like dead. How do you think this is
23:26 going to play out? because a clean deal has not happened in quite some time
23:31 where a spa has announced a merger and simply raised money via common in the
23:35 form of a pipe. It's a great question. I think it comes down to the underlying
23:39 asset. But there are some incredible companies that are private
23:44 that if they go public will be able to demand common pipe capital. I think that the
23:51 future maybe just prognosticating and guessing what does Raptor 3 look like in
23:56 this back. I think the Raptor 3 will look like where somebody a sponsor like
24:02 me rolls everything up into one thing so that it's already pre-wired from the
24:08 beginning where I'll just speak to a billion, two billion, three billion,
24:12 whatever it is, flexible capital that can come in as common so that it's a
24:15 totally pre-baked IPO >> at a very fair price. I think that I
24:19 think that that's what the Raptor 3 version of a spa will look like.
24:22 >> So more capital and then they they put their full trust and faith in the
24:25 sponsor to run the deal. >> Well, then meaning then there's no
24:29 conversion risk that all the money comes over right from >> it comes over, right? And so then you
24:33 have to fully commit in >> you set your compensation to be a bit
24:38 Elonike in terms of your compensation as the sponsor comes if I read it correctly
24:44 Chimath when it hits certain milestones in terms of share price.
24:47 >> Yeah. Nothing can be earned unless the stock is up 50%.
24:51 >> And then there's a tunch at 50. Then when the stock is up 75% there's another
24:54 tunch and when the stock is there's no founder warrants in the deal or there
24:58 are found there's no founder warrants. >> Nothing. >> I think this is great. You know I I was
25:02 asked by >> way the reason the reason why this is important is all of those things that
25:08 you guys mentioned increases the cost of capital to the founder and to the
25:11 private company board and to the employees. All that's unnecessary
25:15 dilution. So now we take it all off the table. >> Yeah. Smart. The thing I, you know, the
25:19 observation I had at the time, not just for your collection of spaxs in the 1.0
25:23 era, but just all of them in general, and I tried to explain this to our
25:29 syndicate members and investors as well as the CEOs because a lot of my CEOs
25:32 were like, should we do a spack? And one of them, Desktop Metal, did
25:37 this felt like venture investing. And you know, if you look at Open Door,
25:41 Virgin Galactic, um, Joby, which I don't think was one of yours, Sofi, MP
25:46 Materials, all of these companies, you h you have to look at it if it is a
25:51 venture type investment, 80% of venture goes to zero, 20% pays up for the other
25:55 80%. I think people were looking at this like it was Netflix and they were not
26:01 thinking of these companies and the stages they were at. >> Well, can I just ask a question? Yeah.
26:07 And then I'll drop it to a question cuz SoFi and MP Materials they did
26:12 extraordinary. So in this class of companies you're going to be taking out
26:15 is it going to be the same early stage or are you thinking more robust more
26:21 predictable revenue let's call it um resilient revenue maybe rugged revenue?
26:25 >> I think it's the latter but I think it's also important to note that this time
26:29 around I've tried to really minimize retail exposure to this. I don't think
26:33 that retail is well suited right now >> to have these things and what my my
26:37 honest advice is >> avoid maybe not all spaxs but definitely my
26:43 spack just avoid it. I think that there is more than enough liquidity on the
26:47 institutional side for us to do an interesting deal, but it fits in our
26:51 portfolio and our construction which is a very different risk model. And so I
26:55 would hate that, you know, people are out on the risk curve without really
26:59 understanding the risks because Jason, you can't predict the market. You don't
27:02 know where these things are going to go. >> Yeah. I mean desktop metal 3D printing
27:07 this is like a very cutting edge nason technology company should have stayed
27:11 private a couple more years or people investing it need to understand you're
27:13 you're now acting like a venture capitalist which means the return
27:17 profile and how the portfolio management works is distinctly different than doing
27:22 Netflix and Nvidia and whatever other publicly traded companies
27:25 >> I would just say do do not invest in these things don't at least you know
27:27 just >> I think you just inspire people to do it I know that's not your intent but would
27:35 when You say don't do it. Stupid. I I'm being very honest. Don't do it.
27:39 >> No, no, I know. Don't buy spaxs unless it's like less than 1% of your portfolio
27:41 The AI rollup opportunity
27:42 would be my advice. >> Before we move on, can I just make one
27:46 comment and I'd like your guys >> know about the private equity stuff
27:48 because Chimoth made a comment that private equity is baked, but I think one
27:52 of the things to take note of in this take private of EA and we talked about
27:58 it is the theme of AI empowering EA to kind of transform the business. And
28:04 Jared's brother Josh has at Thrive been executing a rollup of CPA accounting
28:09 firms that he's been applying AI to to reinvent that business.
28:13 >> Oh, is he really? >> Yeah. >> Oh, I should talk to him because we have
28:16 an investment in a company called taxjpt.com that is basically like
28:20 co-pilots with AI for accountants that's doing spectacular. So what he's done is
28:24 he's bought these kind of traditional accounting firms at some multiple of
28:27 IBITA and then he can transform the business with AI and really create a new
28:30 opportunity. And I've said like I think this is one of those few moments in
28:34 history where there really is an opportunity to beat the market and make
28:38 money in the public markets if you can be thoughtful and selective about the
28:43 companies that stand to benefit from an AI execution strategy. Because in all of
28:48 these traditional kind of markets where you have competition, everything's
28:52 commoditized and the market is mature. It's very hard for any of these players
28:56 to differentiate product service and obviously you know unit economics. But
29:00 with AI, it's completely transformative and has transformative potential in
29:05 nearly every industry. So as a public market investor, if you can identify
29:08 those opportunities, select them where the management team has the right
29:11 leadership in place to execute against this, you could make real money. The
29:15 problem is most of these companies are not led by folks that understand AI or
29:18 software first. >> And so I think there's an opportunity
29:22 for more buyouts. They're not going to be of the $55 billion scale. It's worse
29:25 than that. >> In what sense? So we at 8090 have done the dance with
29:36 all the big major private equity firms. And here's how it goes. It always goes
29:41 the same way. The partners love it because they're looking at minimal
29:47 distributions, companies that are like good but not great in many cases
29:52 and they want to see improvements to EBIT and performance so that they can
29:56 either sell them or move them out. >> And you're sorry you're saying you've
29:58 looked at this you've looked at this with their portfolio. >> All of them. Yeah. All of them
30:02 >> with with their existing portfolio companies. So the GPS are like this is
30:07 genius. We should do it. Then they're like here's a handful of companies to go
30:10 talk to. And I'll be really honest with you, what you find in most private equity
30:17 portfolios are B and C companies run by C and D folks. >> Yes.
30:21 >> And so the ability for them to go and embrace this is basically next to none.
30:25 So if I look at my customer distribution and concentration at 8090, okay, run rating into nine
30:36 figures already working on a three $400 million deal. Okay, about a single
30:39 dollar comes from a private equity firm. Although we spent initially a lot of
30:43 time trying to sell it, trying to sell our software factory and trying to sell
30:48 work into them. It's really hard and it's what you said before Freeberg,
30:52 which is the people incentives at these businesses are misaligned to the AI
30:55 outcome, >> right? >> And you can't fire these people and I
31:00 don't think the right answer is to fire them. So I don't know what the right
31:03 answer is. This is why I think private equity is very challenging. Do you think
31:06 there's a do you think there's a power loss situation where perhaps a handful
31:10 of investors in the public markets and perhaps a handful of investors in the
31:14 private markets can identify and then put the right people in place and
31:17 execute against these strategies like Josh is trying to do with his
31:21 >> I think Josh is smart so I think Josh will figure it out no matter what. What
31:26 I'm saying is if I can show you 20, 30 customers, a ton of revenue, all
31:31 these white papers that show upside, and I still can't get it done inside one of
31:35 these companies, I think it's not us, it's them, right? So, it's not inherent
31:39 in traditional a private equity to do this either, which maybe begs the
31:43 question, is there a new kind of private equity that can execute this? Maybe
31:46 that's an opportunity like like Josh is showing, right? like he's he's a venture
31:50 investor that's executing a private equity strategy and maybe that becomes
31:52 the play. >> I think if this works well, two of our biggest customers are individual deca
32:00 billionaires who own businesses and they're like you're doing this.
32:02 >> Mhm. >> So to the extent that Josh looks more
32:06 like that, which is an owner of 100% of the business where it's like you're
32:09 going to do it, >> then I think it can work. So I think the
32:13 Saudis I think the owner operated model is the only way the AI transformation
32:18 really works and then the the other end of the spectrum it's the public market
32:24 CEO who realizes that they have to do something real because they'll otherwise
32:27 lose their job or they'll be disrupted. Those are the two cohorts that I feel
32:31 today >> are on their forward foot. Everybody else is like sticking their head in the
32:36 sand. Just on the EA front, I forgot to ask you, Sir Demis, my Greek brother,
32:41 >> didn't he show a >> It's just all the always the Greeks who
32:44 get these things done. >> Yeah. >> Didn't he show like the uh 3D engine
32:49 that would make like infinite games? >> Yeah. So, it's not actually a 3D engine.
32:55 It's a class of these AI models that can render what ex what the experience is
32:59 looks like and feels like a 3D world, but it doesn't have an underlying kind
33:04 of traditional object um rendering engine. It doesn't have a traditional 3D
33:08 physics engine. So, it's a new way of experiencing these kind of world
33:12 interaction systems. And there's several startups. I think um Fay is her name,
33:18 the Stanford AI one. Yeah. >> And she has one of these. That's a
33:20 virtual worlds company that has the same principle. >> I asked Bramberg and Alex about exactly
33:25 this >> at dinner. >> What was their take? >> Yeah.
33:29 >> He said it's just really, really hard to get these things to actually be
33:33 legitimate engines at the scale of what Unity offers for the quality of game
33:38 that needs to be made for it to work. The interim step is going to be the
33:43 assets in it are created by AI. That's what I've seen a lot of startups doing.
33:46 So you want to make a character you know you dropping characters and they would
33:49 be done in real time. >> I think I think your whole the whole
33:54 thing is Unify and Unity as the rendering engine and the AI sits on top
33:59 and the AI basically can render objects can render concepts can render structure
34:03 can render the direction that you as an engineer would typically provide to the
34:07 to the Unity or Unified 3D engine and that's going to unlock not just in video
34:10 games but also in film. >> You're 100% right. Can I tell you an
34:14 example? Yesterday there was um you know in our group chat a bunch of people sent
34:17 around the Sora >> the sloth app. Yeah. >> And I downloaded it just to play with
34:22 Sorl yesterday >> and the first video that came up was
34:26 exactly this. It was like a ATP tennis match. >> Yeah.
34:31 >> Where it was a guy's face the guy like imagine you and then playing against
34:36 like a federer. And then I thought well what if he was playing against his
34:39 friend and that was the actual video game. to your point, you you get away
34:43 from all this IP licensing, gatekeeping stuff, and you can just get to good
34:47 games faster, good content faster. I think >> they're adaptive in terms of the
34:51 competition, so you're not playing somebody who's going to just dominate
34:55 you. It just get 5% better every time you play it. You'll get 4% better and
34:59 it'll just make it perfectly challenging so you don't quit and you'll learn as
35:02 you go. It's it's really going to be an interesting >> and the same the same will exist in like
35:06 content J how like you'll make shorts and films and then the ones that have
35:11 the most engagement the AI prompting system will get better and
35:16 better and ultimately it will yield like uh you know bits of content that people
35:19 >> see that happening with Star Wars or Marvel. If all of a sudden Silver Surfer
35:23 is an interesting character to you or Ashokano is interesting to you, it'll
35:28 sort of make that world or enhance that character and tell you more of their
35:32 backstory. And that can be very interesting as a >> how you can sit in your seat and like
35:37 make fun of me, call me a nerd, and you actually know the name of this Star Wars
35:41 character. I don't even know who you are. >> Very important character. Ashoka is
35:45 Anakin Skywalker's Padawan. She is a very important character. If you watch
35:49 the Clone Wars, you would know this. the Actually, >> oh, look who dropped in. Oh, David Sax
36:01 Sacks joins the show!
36:04 is here. Did you get out of your uh Were you in a skiff or something? What's
36:08 going on? Zar, >> I was in some meetings, but actually,
36:12 no, I was just uh buying some domain names. >> Oh, you are? Did you get mahalo.com?
36:17 >> I got I got mahalo for the bargain price of $1 million. That's what it's worth.
36:21 Go to mahalo.com. I'm selling it for a million. I mean, it's it's in the
36:23 dictionary. >> Yeah, I have some old assets. Somebody else should use them. I just I have
36:28 Begin.com and I'm going to be working on that >> in partnership probably with one of the
36:34 large. I might give you an equity squad for that. I'll give you a
36:37 >> mahalo is the second most important name in the second most important word after
36:44 aloha in the um Hawaiian language. I'm surprised Beni off hasn't tried to
36:49 ask you for I was just texting with Beni off. >> Give it to him as a gift, dude. He's a
36:52 great guy. Just give it to >> I will give him the I will give benny
36:58 off mahalo.com if he gives me four weeks in one of his Hawaii resorts per year.
37:02 >> He would do that. >> Oh, for the next 20. Oh my god. Imagine
37:07 Jake for 80 weeks. Oh my god. As a house for 80 weeks as a house guest. He could
37:11 be there. He could be there. >> It doesn't matter. I'll give him the
37:14 money so he buys it. Don't worry, donate it to his nonprofit foundation. Then you
37:18 can take a tax write off. >> Look at everybody's When I have
37:21 something to sell, the guy with the lowest net worth on the program when I'm
37:26 trying to pay off my jet, you guys all have criticism. How come I can't wet my
37:30 beak? I got >> Let me ask you a serious question. So you had investors in Mahalo, right?
37:34 >> Yes. >> And I assume >> this is their domain. This is their
37:37 domain. It will go to them. >> Oh, so it will. Oh, okay.
37:39 >> It will go to those investors. You're paying off liquidation preference,
37:42 >> correct? >> Okay. Just sitting there. >> So now instead of losing 100% I'll lose
37:48 99. >> Something like that. >> Uh it's just startups are hard folks.
37:53 >> But I have the begin.com and I've been talking to folks. I you know I mahalo
37:57 was originally a human powered search engine like Wikipedia which we're about
38:01 to get to and my concept was to do comprehensive search like neighbor.com
38:05 or dam in Korea had seen those services. Yeah. And it turned out to be exactly
38:09 like perplexity, but at the time we we tested machine learning, which is what
38:12 everybody called AI back then, and it just didn't work. So, we were trying to
38:16 hand roll search results and then back them up with, you know, computerenerated
38:20 ones, algorithmically generated ones, but the tech wasn't there now. Um, but I
38:23 want to do something again with begin.com. I'm really excited about that
38:27 OpenAI and Meta launch short-form video apps: "AI Slop" or the future of content?
38:27 domain name. All right, listen. We brought up Slop. Let's get into it. Two
38:33 slop apps in a Fortnite here. Uh, no pun intended. Zuck and Sammy the Bull have
38:38 both released uh >> the bull pull. What a deep pull. Sammy the bull.
38:42 Gravana. >> There it is. And uh here's a look at Sora. It's objectively extremely
38:49 impressive. Here's Sam Alman. People don't know this. Early in his career
38:52 when he was starting OpenAI didn't have the money from Elon.
39:00 And here's Sam Alman stealing an H100. Here's Sam Alman. Also, this is when he
39:05 was um storming the capital on January 6th. Here he is at when he was working
39:10 at Google. Yeah, lots of but it's really good and they are basically taking a ton
39:15 of risk and solving some problems with IP. As we all know, the IP outputs is
39:19 where people think you're going to have to be really thoughtful or get a bunch
39:24 of lawsuits. On this app, you can opt in and make your persona like Sam did
39:29 available to everybody to use. So that whole concept of notable persons
39:34 allowing their image to be used, you opt into that and that's pretty clever. So
39:38 you can let your and you can make it so your friends can, you know, basically
39:42 make videos of you but nobody else can. It's it's a thoughtful way of doing it.
39:46 However, very controversially, this thing had everybody's IP in it and you
39:51 have to opt out if you don't want your IP used. That's going to get him another
39:53 whole collection of lawsuits to go with the New York Times and Z Davis ones. And
39:57 there have obviously been a bunch of settlements now, uh, Anthropic settling
40:03 their book thing for 1.5 billion. So, anybody play with these tools yet? And
40:06 what do you think, folks? And what's the point of these? Do we think this is like
40:10 a Tik Tok competitor >> tomorrow? Do you think it's just back
40:14 door to training data? What do you think? >> The closest thing is a Tik Tok
40:17 competitor, but I I use it. I thought it was okay. But again, the thing that I
40:22 have that I keep in mind whenever I try these apps for the first time is
40:27 >> today is the worst it'll ever be. >> Sure. >> It it only gets better from here. And so
40:33 if you look at the starting point, it won't take but a year where this thing I
40:38 think or maybe two years where this thing is legitimately excellent. It has
40:42 to get the scripting right. It has to get the prompting right. It has to be a
40:46 little bit easier for you to use. There was a bunch of prompts that I used that
40:51 were rejected by so or by the IP, right? >> Well, it just said use me, but I
40:55 couldn't validate that I was me. And so, you have to take a picture of yourself.
40:58 It's a it's a little clunky the app right now, but you're right. It's going
41:00 to get better in each version. The one by >> Zuckerberg is called Vibes. I you know,
41:07 I was looking at these sacks and I don't know that this is intended to be like
41:11 the next great social media app as much as it's a data play to get folks to
41:16 train data. when you see them, what are any thoughts on them other than
41:20 interesting? Yeah, >> I haven't played with it yet, so Oh,
41:23 >> sorry for me to say. >> Freeberg, you got any thoughts on it?
41:26 Just >> uh No, I I don't have like thoughts. I I think, you know, we're kind of early
41:32 innings. I do think there's like new categories of media that none of us are
41:36 really considering today. Like traditional media, as I've mentioned in
41:39 the past, is like centrally produced and then broadly consumed. And I think that
41:43 there's models of media that are going to emerge that are going to create new
41:46 business categories or new business models and and also new media categories
41:50 that are all about kind of distributed production and not necessarily like
41:55 central production, distributed consumption. So that that kind of
41:58 changes things quite a bit and I think maybe this is going to start to open
42:02 that door a bit. One of the things I because I thought about this and I I
42:04 mentioned this in the past where I'm like everyone's going to make their own
42:07 movie, their own video game, their own music, but there is this notion of like
42:13 shared cultural context. Like everyone wants to talk about, you know, how did
42:18 the 49ers do this weekend or did you guys see that show adolescence? Did you
42:23 guys like we want to have a conversation about some shared stories that's the the
42:28 basis of kind of societal interaction and mimetics. So I think like there are
42:32 elements of this being the beginning of the enabling tools, but I don't think
42:34 we've actually seen what's going to happen, which is how do you take one
42:39 story and then create a distributed way of consuming that story where everyone
42:43 experiences and consumes it differently. So I do think like this notion it's like
42:47 hey everyone's making fun of Sam or does some like maybe there's some cultural
42:50 context about Sam Alman that we all share and then we're all like engaging
42:53 with Sam Alman in different ways you know. So, so I think like there's we're
42:56 very early and we don't yet know kind of how it's all going to play out, but I
43:01 think that's really critical to >> bring it is something is lost because we
43:05 used to all talk about the latest Tarantino movie or the latest, you know,
43:10 Sopranos episode. We don't do it anymore. >> And I I do share stuff. We do talk about
43:15 tweets and stuff and you know there's other forms of groups but it's it's not
43:21 like it used to be where 30 40 million people would see Raiders of the Lost Arc
43:24 and it would be the discussion of the summer or whatever it is. And so I I
43:27 literally bought 20 tickets to the new Paul Thomas Anderson one battle after
43:32 another just so I could have a conversation with 20 friends about the
43:37 new PTA. And so people really are longing for this shared experience.
43:41 >> Paul Thomas Anderson he did the master there. just there will be one of the
43:47 greatest ever he is top five director of all time but I know you don't care about
43:51 culture um but is he like is he like Michael Bay >> it would no opposite of that actually
43:57 Michael Bay makes things that go boom Paul Thomas Anderson's that make makes
44:00 things that make you go >> Michael Bay super cool fun to hang out
44:03 with fun to party with >> right okay well way to bring it back to
44:07 you um okay hold on you dropped a name here is >> I don't know Paul Tom Sandra but it was
44:13 a heck of a film as Sax. Sax is actually very cultured when it comes to cinema. Did you see it
44:17 yet, Sax? >> I have not seen it yet. No, >> it's it's of the moment
44:25 >> and it's heard it was anti- conservative. So, it doesn't have some
44:29 leftwing take. >> No, it kind of mocks the left and the
44:33 right. It's kind of mocking both extremes. You'd love it. >> I think you very much appreciate it.
44:36 >> All right, I'll check it out. >> Yeah, I would check it out. Uh,
44:39 >> hey, I have an idea. Why don't we find a topic that's interesting to talk about?
44:42 >> Yeah. Okay, great. Yeah. Well, that's a well, if you contributed to the docket
44:45 or showed up on time, maybe we could do that. Um, so unbelievable. Just so you
44:49 know, the inner workings right now, there's a little resentment in the group
44:54 because one of us decides to change the time of the pod for four weeks in a row
44:57 and then show up half an hour late. I won't say which person that is, Sax. Uh,
45:02 Sax, but here's an interesting topic from Red Meat for you. Deepseek, the
45:04 Open source AI: DeepSeek's new model, pressure on US AI industry
45:06 Chinese LLM, just dropped their latest model 3.2 EXP. It's faster, it's cheaper, and it
45:14 has a new feature called DSA, Deepseek sparse attention, which makes
45:18 it faster to do uh training and inference at larger tasks. The key
45:23 takeaway is it can reduce API cost by up to 50%. The new model charges 28 cents
45:28 per million inputs, 42 cents per million outputs. Claude, which is a leading
45:32 model from Anthropic that a lot of developers use, a lot of startups use,
45:38 is like $3.15, so 10 times 35 times more expensive. Obviously, people are cutting
45:42 their prices pretty quick. But, uh, Sax, this is your wheelhouse as our ZAR of
45:47 crypto and AI for the United States of America. What are your thoughts here on
45:53 the continued execution of the Chinese government with Deep Seek?
45:56 >> Well, I want you to hear Freeberg's thoughts on this because he was paying
45:59 attention to this, weren't you? Yeah, I mean I think there's a total
46:03 rearchitecture underway and we're at the earlier stages of cost per token in
46:07 terms of dollar and energy. My understanding is there's actually a lot
46:10 of work going on with US labs right now in a similar kind of track that's going
46:14 to result in similar results. Maybe they're a little bit ahead of the curve,
46:18 but we should really pay attention to the curve. I think you know what do the
46:23 models say in terms of energy demand in terms of cost per token if these
46:28 architectural changes really do drive down 10x 100x a 1000x 10,000x um over
46:33 the coming months >> and this is open source so just so
46:38 everybody understands it's available on AWS it's available on GCP at least 3.1
46:42 is I don't know if 3.2 too is available there now, but I'm hearing from a lot of
46:45 startups, I don't know if you're hearing this in the field, Chimoff, that they're
46:48 testing it and playing with it in some cases using it because it's uh so much
46:52 cheaper. Are you seeing that? >> We are a top 20 consumer of Bedrock. So,
46:59 let me tell you what it looks like on the ground. We redirected a ton of our
47:04 workloads to Kimmy K2 on Grock because it was really way more
47:09 performant and frankly just a ton cheaper than OpenAI and Anthropic. The
47:15 problem is that when we use our coding tools, they route through Anthropic,
47:19 which is fine because Enthropic is excellent, but it's really expensive.
47:24 The difficulty that you have is that when you have all this leaprogging, it's
47:28 not easy to all of a sudden just like, you know, decide to pass all of these
47:35 prompts to different LLMs because they need to be fine-tuned and engineered to
47:38 kind of work in one system. And so, like the things that we do to perfect codegen
47:44 or to perfect back propagation on Kimmy or on Enthropic, you can't just hot swap it to deep
47:50 speed. All of a sudden, it comes out and it's that much cheaper. It takes some
47:54 weeks. It takes some months. So, it's a it's a complicated dance and
48:01 we're always struggling as a consumer. What do we do? Do we just make the
48:05 change and go through the pain? Do we wait on the assumption that these other
48:10 models will catch up? >> So, >> it's people are making tools now
48:17 >> that and by the way, I can't just make it easier to switch between them.
48:20 >> No. And like you know this weekend a different company with a huge model came
48:25 to us and gave us the preview of their nextg model. Okay and it's incredible
48:31 but then when I sit on Monday morning with my team and I'm like okay what do
48:36 we do? We don't know what to do. Do we cut it? Do we move over and say great
48:41 we'll refactor all these workloads to run on on this new model? It's a it's a
48:45 really hard problem and it's getting worse the more complicated tasks that we
48:49 undertake. Okay. And just for people who don't know, Kimmy is made by Moonshot
48:53 AI. That's another Chinese startup in the space. Sack, your thoughts.
48:57 >> Well, I think this is actually a really interesting topic. This topic of open
49:01 source. I'm a big fan of open source software because it's a it's a check on
49:05 the power of big tech in a way. What we've seen in the past in the history of
49:11 technology is that these major categories end up getting dominated by
49:14 one or two big tech companies and they have all the power and control. And open
49:19 source provides an alternate path, right? Because the community of open
49:23 source developers just puts things out there and then you can take it and run
49:26 it on your own hardware and you're not dependent, right? It's a path to sort of
49:31 software freedom, if you will. So, so far so good. I think the thing that is
49:35 now tricky about this is that all the leading open-source models are from
49:41 China these days. China has made a really big push on open source.
49:44 Obviously, DeepSeek is an open source Chinese model. That was the first big
49:50 one. Kimmy is one. Quen from Alibaba. And so I think that if you want the US
49:54 to win the AI race, then we're all kind of two minds about this. On the one
49:58 hand, it's good that there are open- source alternatives to the closed source
50:03 proprietary models. On the other hand, they're all coming from China. Now,
50:08 there were some American efforts that have been important. So, Meta most
50:12 notably has invested billions of billions of dollars in llama. But the
50:16 release of Llama 4, I think, was considered disappointing by a lot of
50:19 people. And now there's statements by Meta that they might be backing away
50:22 from open source and just going proprietary. OpenAI released an open
50:27 source model, but it's nowhere near their frontier. And there are some startups that are
50:34 trying. So there's one called Reflection that looks promising is developing an
50:38 open- source American model. But so far, this is maybe the one area in AI where
50:44 the US is behind China. as this sort of open source models. I'd say every other
50:48 part of the stack, closed models, chip design, chip manufacturing,
50:51 semiconductor manufacturing equipment, every other part of the stack, even data
50:55 centers, I would say we're we're ahead, but this one area of open source is a
50:58 little bit concerning. >> Interestingly, Saxs, the two things of
51:05 note is OpenAI. The open was originally that they were supposed to do open
51:10 source. So, that's kind of hilarious. But the second is that Apple, which is
51:13 the furthest furthest behind of everybody, they have a really
51:16 interesting open source model. So when you're behind like Apple is or the
51:20 Chinese were, you're open. You're you do open source and when you're ahead like
51:24 OpenAI became with ChatgBT, you close it down. But >> that uh
51:29 >> can I tell you open Elm Open ELM? Yeah. Efficient language models from Apple.
51:32 Keep an eye on that one. >> Can I tell you what's going to make this
51:35 open source closed source battle even worse? Because effectively what this is
51:40 is the US versus China. the US is closed and China is open at least at the scaled
51:45 models that work. >> But that doesn't have to be the case,
51:47 right? Because we could release open models too. >> No, no, you're right. I'm just saying
51:51 today if you look at the conditions on the field, the closed source, highly
51:55 performant models are American. The open- source highly performant models
51:59 are Chinese. And you would say, okay, well, what is the next downstream thing?
52:03 It's what Freeberg mentioned, which is the energy and the cost of generating
52:07 these output tokens. And I talked to somebody yesterday who runs a huge
52:11 energy business and I have to tell you it's not in a good place. Meaning you saw I think this
52:20 week where the residents of Indianapolis were able to reject or get their city to
52:26 reject a billion dollar data center that Google was going to build near
52:31 Indianapolis largely because of concerns of price inflation around electricity.
52:37 And what this energy CEO told me is, look, the next five years are baked. And
52:44 if we don't find some compelling solves, and I'll tell you what the two ideas
52:47 were, but if we don't find some compelling solves, electricity rates
52:52 will double in the next 5 years. Now if you think about how then consumers will
52:57 view the use of AI and then if you think about companies like us and others trying to use the
53:02 cheapest version so that we are minimally impacting the downstream cost
53:06 of these things because it will become an energy problem. This is a very
53:12 complicated thing. Now his idea and it's a huge PR crisis because if you want to
53:16 take big tech which is already viewed negatively and make their perception
53:20 even worse. If you start to finger point to them and say these guys are the
53:24 reason my electricity costs have doubled in the last 5 years that is no bueno for
53:29 them and they need to find an offramp asap. >> It's a bad look
53:35 doubling this could take your jobs right. >> Yeah it's terrible. Whether you believe
53:38 that's true or not, that is the perception. >> There are two offramps. There are two
53:43 offramps that he suggested which I think are worth considering.
53:47 >> Offramp number one is what's called a cross subsidy which is essentially to
53:53 say that they pay a rate card which they can absorb with all their free cash flow
53:58 materially higher than what other rate payers would pay in that geographic
54:03 area. So the homeowner his or her electricity costs stay flat to down. The
54:10 data center costs are higher and it's the Metas, the Googles, the Apples, the
54:14 Amazons who have hundreds of billions of free cash. They absorb it. That's that
54:19 was idea number one. And idea number two is to start to set up some mechanism so
54:24 that they can install things like batteries at every single home in and
54:29 around these data centers to allow those homes to have a better chance of
54:33 actually um absorbing some of this inflation without having to pay it.
54:36 >> That's a really good idea and this is playing out sachs in Virginia in a major
54:39 way because that's where data center alley is and 40% of the energy in
54:45 Virginia now is going to data centers. This is becoming acute. So what what are
54:48 your thoughts here, Zar? >> Well, Chris Wright spoke to this pretty
54:51 well at the all-in summit in terms of what we have to do. I mean, there's no
54:54 question that AI is going to create a huge need for power over the next 5 or
55:00 10 years. I think on a 5 to 10 year time frame, the answer is probably nuclear or
55:04 at least that's a big part of it. But nuclear takes at least 5 years. Within
55:08 the next 5 years is probably gas, you know, natural gas. But the issue there
55:13 is there's a huge backlog for gas turbines. basically the engines that
55:19 burn the gas to create power and there's like a two to three year backlog for
55:22 those to spin those up. So the question is what do you do in the next few years
55:25 and I think Chris Wright talked to this and I've heard this from other energy
55:28 executives which is we just need to squeeze more out of the grid. If we were
55:35 to shed just 40 hours a year of peak to say backup generators, diesels, things
55:38 like that, you could get an extra 80 gawatt out of the grid. This is what one
55:43 energy executive told me. The reason is because they build the grid and they
55:48 have regulations on it based on the peak, right? Which is basically the
55:50 coldest day in winter or the hottest day in summer. And the same way that you you
55:55 know you build your church for Easter Sunday, the rest of the year it runs at
56:00 50%. Same thing with the grid. And so if they could just reduce the the peak 40
56:04 hours, if they could shed that load to backup to generators, to diesel, things
56:08 like that, then they could run the grid to squeeze an extra 80 gawatt out of it.
56:13 And I think that's the bridge over the next few years that we need to then get
56:17 a lot more gas and then eventually some nuclear as well. But unless you want to
56:22 keep talking about electricity, I think there's some other things to talk about
56:26 on the open source cuz I think it's a pretty interesting topic actually. And
56:31 if can we just go back there? Yeah. Yeah. Yeah. I was I was just trying to
56:35 paint the case that my economic model for going to open source is better
56:40 because I can't pay $3 an output token >> and then also pay for all this
56:43 >> actually I want to I want to ask you when you're running like Kimmy or
56:47 something like that. So I think it would be good just to explain to the audience
56:49 how this works because I think there's a lot of confusion about what it means to
56:54 be an opensource model. A lot of people think that when a Chinese company
56:58 publishes one of these models it's still somehow theirs. >> No. But the reality is once it's
57:03 published, it's no longer theirs. It belongs to anyone who wants to take that
57:07 code and you're not running that on a Chinese cloud or something like that.
57:10 The data is not going back to China. >> You're taking that model and you're
57:15 running it on your own infrastructure. >> Can you just explain this?
57:19 >> Yeah. So when I first started 8090, my only solution was uh Bedrock, which is a
57:24 service that Amazon provides that allows you to essentially get inference as a
57:28 service. Right? So as we are building our product and we need inference and we
57:34 need inference tokens, bedrock basically handles everything. So it's it's what
57:40 AWS is but for this vertical of AI, right? So they have the servers. These
57:45 are in American data centers. They're managed by Americans and what they do is
57:50 they take a handful of models and they make sure that they can support usage of
57:56 those models. That was how we started. But as with everything, we have to
58:01 manage our costs and our operating profile. And so we're always looking
58:06 for, are there other models and other places other than Amazon that can
58:10 service our needs? Because in fairness, Amazon is very expensive. So, a
58:15 different company that I helped get off the ground, Grock with the Q, they have
58:21 a cloud and what they've been doing is they've been working with initially
58:26 Llama, then they work with OpenAI to bring their open source model, but they
58:30 also brought online a couple of these Chinese models. And what they do exactly
58:34 as you said, Sax, is they take the source code, they basically implement
58:37 that, fork it, they fork it, they fork it, and and now it's implemented
58:42 domestically. on American soil by Americans inside of an American data center. So there's
58:50 China gave us kind of the the way the road map if you will the architectural
58:54 plans but we as in you know the American company in this case Grock built the
58:58 house and then launched it and so now we as 8090 basically made a cost decision
59:02 to move to this open source model because it was just materially cheaper
59:06 >> right and what Grock with a queue will give you at you're the application
59:09 company 8090 >> they're like Amazon for us they're >> they'll give you an API
59:11 >> exactly >> so the same way if you want to use a
59:15 closed model like open AI or you know chat GBT They'll give you an API. You
59:19 submit prompts. They give you answers. Basically, tokens in, tokens out.
59:23 >> What what Grock does is they will take this open source model, run it on its
59:26 own infrastructure, and then give you the API so that you can then get tokens
59:30 in, tokens out through their API. >> Well, for me, as a consumer, it reduces
59:36 us to a pure economic decision. Where is it cheaper? And you know, it's not
59:40 dissimilar to the last generation of the internet. You'd run on AWS, but then
59:44 you'd bid it against GCP. You'd bring in Azure. you'd say who is cheaper because
59:48 ultimately you're running a database, you know, you're running, I don't know,
59:52 pick pick your service, Snowflake, >> right? >> It didn't really matter where it was.
59:55 You were just really trying to find the cheapest vendor. >> Right. Now, here here's here's what's
59:58 compelling about it. So, first of all, like you said, it's cheaper to just run
60:01 it on your own infrastructure if you know what you're doing. Also,
60:05 enterprises like it because it's more customizable and there's going to be a
60:08 lot of fine-tuning of these open source models for specific applications
60:13 >> 100%. And enterprises frequently want to run these models on prem in their own
60:17 data centers because they want to keep their own data on their own
60:20 infrastructure. But now the challenge is you've got these models that they're no
60:24 longer Chinese. They've been forked. It's an American company but they
60:28 originated in China. >> That's right. >> And they could be running on some
60:31 critical infrastructure and that you that that does raise issues. I mean, do
60:36 what is Grock doing, I guess, to like test whether these models are safe,
60:41 whether they could be backdoored. I mean, how do they think about that?
60:44 >> They they have an entire pipeline of stuff that they do, the details of which
60:47 I I don't exactly know because I've not asked exactly what they run through, but
60:51 >> yeah, they're big rub in this. >> They go through an incredibly rigorous
60:54 >> they basically do like safety testing to make sure. Absolutely. So, I mean,
60:56 because a lot of people think that if you run a Chinese model, the data must
60:59 be going back to China, but it's not if it's being run on your own
61:01 infrastructure. I think the issue is more theoretical that like could a
61:06 Chinese model somehow be backdoored with an exploit or vulnerability somehow.
61:09 >> Well, if you take a compiled version, sure, but if you just take the open
61:13 source and you do it yourself, no. >> Right. Well, that's the thing. So, I
61:16 mean, and if someone did discover a vulnerability, it would get widely
61:20 shared in the community very very quickly. I think you can I think I think at this
61:25 point you can expect that every single major company that is in security that
61:30 is in a cloud vendor and also every single major model maker is trying to
61:35 prove and invalidate how the other models are inferior or bad in some way
61:40 and so that's where the competitive cycle I think is really valuable because
61:43 you do have the best and the brightest computer scientists like you know
61:48 yesterday a certain person he's Italian that's I a leading security guy at one of these
61:55 model makers just talking to him. He's in charge of this security stuff.
61:58 They're hammering everything to try to figure out whether there's a there's a
62:02 vulnerability because it slows these other folks down. So that made me feel
62:06 quite positive that we haven't seen anything yet on any of these models
62:10 which is to say that generally everybody is actually been a pretty good actor so
62:13 far. >> The other piece to this puzzle sachs is there's a lot of crypto distributed
62:19 projects. The one I've been working on is Bit Tensor and Tao. I think you've
62:23 also done a deep dive on this Chimath and I'm a partner in a you know an
62:27 emerging crypto fund called Still Core Cap and we're buying Tao and we're looking at
62:35 Bit Tensor and all of these subnets that are being made to do distributed
62:38 computing and this is a big push for Apple as well. A lot of these M4 Mac
62:43 minis you've seen out there. Their plan is to put all of this uh all these LLM
62:48 sacks on people's personal computers and then distribute them and have this like
62:53 SETI at home and an incentive layer. And I think that's going to be a big part of
62:56 this. People are not going to want their AI jobs to go to the cloud necessarily.
63:01 They might want to do it locally and I think that's where the phones and all
63:04 this silicon is going with um you know Apple's big focus on it. It's going to
63:07 be >> Yeah. Well, brave new world. >> Yeah. You bring up an interesting point.
63:13 You know, in the early years of this AI revolution, I'm talking about like 2023,
63:16 2024. I mean, this started in the last three years. There was this analogy that
63:22 AI was like nuclear weapons. I mean, you hear the the doomer crowd, the safety
63:26 advocates saying this that like AI was this really threatening technology.
63:31 And they would even say things like GPUs are like plutonium, you know, things
63:35 like that. And I think that model of the world is just wrong, right? Because what
63:39 we're seeing is um and Justin actually had a pretty good line about this. He
63:43 says nobody needs nuclear weapons. Everyone needs AI. And it's true like
63:47 every consumer, every business is going to want to run AI. A lot of them are
63:50 going to want to run it on their own infrastructure. Consumers are going to
63:53 want to run it on their own phone. You're going to have an AI that's highly
63:57 personalized to you. And so everyone's going to have AI. It's not like a
64:00 nuclear weapon where we want to stop all proliferation. AI is a consu first and foremost a
64:06 consumer product that is going to proliferate and so the question is
64:11 bearing that in mind how do you then create an appropriate response for the
64:16 national security risk but this idea that we're just going to stop AI and
64:20 only have two or three companies who who have it which I think was the view a few
64:25 years ago among policy makers ridiculous even now yeah they they were thinking in
64:29 very centralized terms and I think what we're seeing now is regardless of what
64:34 certain policy makers might want. It it's already highly decentralized,
64:38 right? You've got five major American disclosed source companies. You've got
64:43 eight major Chinese models and then you've got everything that's happening
64:47 with startups. So, this is going to be highly decentralized and
64:51 >> and verticalized, right? All the hugging face models, there specific ones on
64:54 images, specific ones for video. Like, it's it's going to be super fragmented.
64:58 >> The vast majority of this activity is benign. I mean, that's the thing. These
65:01 are business solutions. These are consumer products. These are viral
65:06 videos. Most of the stuff does not rise to the level of a nuclear weapon or
65:10 something like that. >> This is a good chance for us maybe to
1:05:11 State AI regulation frenzy: States' rights vs Federal control, overregulation
65:14 talk about AI regulation. There is uh a lot of and and maybe we'll get to
65:19 Wikipedia as well, but there's a lot of states that are starting to look into
65:25 regulating AI. California SB53, the Transparency in Frontier Artificial
65:32 Intelligence Act, is working through the system. It's going to serve as a
65:35 template possibly for other states. It was introduced in January as an
65:39 alternative to the more sweeping bill, the SB 1047. This would require AI
65:46 developers to conduct extensive safety tests before rolling out the models. It
65:50 got a lot of push back from tech obviously and Newsome ultimately vetoed
65:56 it. But this new law focuses only on the most advanced large frontier models that
66:00 we just talked about. And it requires companies to release a framework for
66:04 knowing how they're approaching safety issues, including standards and best
66:07 practices, whatever that means, and however safety is defined.
66:12 These are models, I guess, in this definition, that have half a billion in
66:15 annual revenue. I don't know how they pick that out, but it requires these
66:19 companies to release transparency reports before deploying. So they're
66:22 going to be like the app store, I guess, if this gets through to approve frontier
66:27 models with updates. Oh, that sounds great. You got to go to the government
66:31 to release a new model. Your thoughts, David Sachs, >> of AI,
66:37 >> I think it's very concerning. There's a a regulatory frenzy happening at the
66:41 states right now. Just to be very clear about what happened in California, there
66:46 was a original bill SB was it 1047 that was incredibly obtrusive that Newsome
66:51 vetoed that, but now they've passed a new one which is called SB53.
66:58 And like you said, it's not as burdensome and intrusive as the previous
67:05 version. It focuses on making frontier AI models report safety risks.
67:11 They're supposed to report if they have >> Can I stop you there for a sec? What is
67:14 the safety risk they're going to be required to report? That's it's such a
67:19 nebulous term. What safety? What? That it's going to jump out of the computer
67:21 and murder me? Safety that it's going to give me the wrong answer.
67:25 >> They're supposed to they're supposed to report on potential catastrophic harms
67:31 related to cyber attacks, bio threats, model autonomy, which is the Terminator
67:37 scenario. And they're supposed to >> okay >> let the government know if there's a
67:42 safety incident. I mean look all these things are quite nebulous.
67:45 >> It's almost like a nuclear power plant having to report if there was an
67:50 incident. Are any of these in your mind thoughtful? >> Let me just let me just let me just
67:55 interrupt for a second. I think it's the equivalent >> of saying I need any factory to report
68:02 to me on the risk of something of a nuclear explosion. Even though the
68:05 factory might not be working with nuclear material, you see it like it it
68:08 uses a >> trying to get out here. I'm confused. >> I mean, it it effectively uses
68:13 terminology that makes everyone nod their head and say, "Oh, yeah, that
68:17 makes sense. That's a good idea." When the reality is that the legislators have
68:22 actually no concept of what they're talking about. They have no concept of
68:26 how these models are built. They have no concept of how they're deployed. and
68:30 they're using language that they think is inevitably going to result in giving
68:34 them ultimately tools and control over a private market system. And that's
68:37 fundamentally what I think a lot of this comes down to. Think about this issue
68:42 that's going on with free speech in California. this hate speech bill SB771
68:46 that's sitting on the governor's desk to be signed right now where effectively
68:50 the state of California's administrators have the ultimate say of what is deemed
68:55 hate speech and not which if you think about it if they had this bill in
68:59 Alabama during the civil rights era there would have never been the ability
69:02 to have the protest and realize the equal rights that arose from the civil
69:05 rights movement because the government would have said those are inappropriate
69:09 hate speech things that you guys are saying and we're now putting those same
69:12 tools in the hands of the legislators They're going to do the same thing with
69:16 AI. They're giving ownorously powerful tools to the legislators to let them
69:20 decide what is and isn't appropriate for private market actors when they actually
69:24 have no sense and no sensibility about what they're talking about.
69:25 >> So, >> yeah, I'd actually I think that's a really important point. Just let me give
69:31 you some stats on this this regulatory frenzy that that's happening. So, all 50
69:38 states have introduced AI bills in 2025. There's been over a thousand bills in
69:43 state legislatures. 118 AI laws have already been passed across the 50
69:48 states. The red state proposals for AI in general have a lighter touch than the
69:51 blue states. But everyone just seems to be motivated by the imperative to do
69:55 something on AI, even though no one's really sure what that something should
69:57 be. Exactly. >> And there's no real agreement on like what all these AI regulations are
70:01 supposed to do. So, they're just making things up >> or what the risks are.
70:05 >> That's what I'm trying to get at. So, let me ask you a specific question.
70:07 >> Yeah. Well, I was going to finish the point about California. So, so look,
70:12 California, they've kind of gotten to this point where now it's about
70:16 reporting on all these safety risks. And if this is all it was, then it would
70:21 just be basically a bunch of red tape and it wouldn't be so bad. The problem
70:25 is that you've got to multiply this by 50 states. So, you've got 50 different
70:29 states, each with their own reporting regime, which is going to be a trap for
70:32 startups. They've all got to figure this out about what they're supposed to
70:35 report on, what the deadlines are, who to report to. I mean, this is like very
70:39 European style regulations. Actually, maybe even worse than the EU because the
70:43 EU tried to basically harmonize to get to one authority. We're going to have
70:47 50. They're going to have one. But the other problem is that this is just the
70:50 camel's nose under the tent. So even in California, Scott Weiner, who's the
70:55 legislator who did SB 1047, now he did this. He's got a block of legislators
71:00 and they have 17 more AI regulation bills that they want to pass. So this is
71:03 just the beginning. And if you want to see where this is going, okay, look at
71:07 Colorado. We should talk about this Colorado bill because this has already
71:11 been passed into law and it's called SB24-205, Consumer Protections for Artificial
71:18 Intelligence. It was passed all the way in May of 2024. So, it was one of the
71:23 first to pass. Even though they didn't really know what they were trying to
71:26 regulate, no one's quite sure how to implement it. But what the law does is
71:31 it bans something they call algorithmic discrimination. Okay? And algorithmic
71:37 discrimination is defined as unlawful differential treatment or disperate
71:42 impact based on protected characteristics. So things like age,
71:47 race, sex, disability. If any of those factors drive an AI decision and it
71:54 results in a disperate impact, then both the developer of the AI model and the
71:58 deployer, which means basically the business that's using it, can be in
72:01 violation of this law and they can be prosecuted by the Colorado attorney
72:05 general. Let me give you a practical application here. So, let's say that you
72:10 got someone like a mortgage loan officer who's reviewing applications, okay? And
72:16 let's say they don't even discuss race. There's it's not on the form, okay?
72:19 They're just using race neutral criteria like a credit rating or financial
72:23 holdings, something like that. If the result of their decision nevertheless
72:28 had a disperate impact on a particular protected group, its decisions could be
72:33 found to be discriminatory. And moreover, the developer of that model
72:38 could be liable even though their model just gave an answer that under the
72:43 circumstances was truthful. The only way that I see for model developers to
72:47 comply with this law is to build in a new DEI layer into the models to
72:53 basically somehow prevent models from giving outputs that might
72:58 have a disperate impact on protected groups. So, we're back to woke AI again.
73:02 And I think that's the whole point. Yeah, >> that's the whole point of this Colorado
73:04 law. >> But let's get Shimoth in on this discussion. Shimath,
73:08 >> I think that this is really, really dumb. What's happening? If you have 50
73:13 sets of rules, what you will have are some conservative versions of AI. You'll
73:19 have some progressive leaning versions of laws. These 50 series of laws will
73:24 essentially just render this industry impotent and incapable of maximizing
73:29 itself and and actually doing what's necessary to drive productivity and GDP
73:33 on behalf of the country. There is no conceivable way, as Freeberg
73:38 said, that anybody in Sacramento or Little Rock or, you know, name your
73:43 state capital will have the intellectual wherewithal to get to an answer as good
73:49 as the federal government will and as Sax will just to be totally honest with
73:53 everybody. So what should happen here is that there needs to be a complete
73:57 moratorum and the federal government should be given the time to figure out what the
74:03 framework should be so that there is a one size one set of rules. Now if that
74:08 doesn't happen and this is allowed to stand there is a perfect example of
74:13 where this has happened before and that is in the car market because in the car
74:18 market what happened was there is a complete set of rules in California
74:24 for emissions that is entirely different than the rest of the country and you can
74:29 look and see what it did now that's just two sets of rules >> and what let me let me let me finish
74:35 >> okay >> and so what these two sets of rules going from one set of rules to two. What
74:41 did it do? It drove most of these companies to go towards barely break
74:46 even or massively money losing. It has been something that the entire industry
74:51 has been fighting back on for now 10 plus years now. Can you imagine instead
74:57 of two sets of rules you have 50? I think you know what the economic
75:01 consequences will be. you'll render this entire category incapable of being able
75:06 to generate any positive economic output. So I guess the steel man if we were to
75:14 make one is transportation, education, abortion, taxes, alcohol,
75:19 cannabis. I think I mentioned those are all state >> cannabis is a poison and uh it is the
75:25 the worst thing in the world, >> right? But for you, >> okay, that's your opinion. Great. But
75:31 should states have some general are trash? >> Oh, okay. We know your position on that.
75:35 I'm talking about the different which is what should states zombies.
75:40 >> Perfect. What are I don't disagree with with that statement.
75:45 >> The question I'm asking is we let states just to steal man this for the audience
75:50 decide how they want to execute against things like taxes, alcohol, education,
75:56 abortion, transportation. Should David Freeberg states have some rights here?
76:00 This is the I'm just stealing here. I'm not saying this is my opinion. But if
76:04 this is the most transitional technology of our lifetime, >> shouldn't states have a say or what's
76:09 the argument for states having a say? >> It's the United States. It's a federated
76:14 republic. I am 100% in favor. I think what we're pointing out is the idiocy of
76:18 these decisions for number one. Number two, so so the internet created a
76:24 virtual network system for media, communications, content, productivity.
76:32 So, you know, we're talking about something that stretches across the
76:35 federal landscape. What needs to happen is there needs to be federal
76:39 preeemption. So, the federal government, Congress, needs to pass a law
76:43 >> that says, "Here are the standards that we are going to set or here's the the
76:47 rules that we think are relevant for AI. here are the things that states can and
76:51 can't do if we want this country to succeed on the uh opportunities and
76:56 advantages that will arise from AI. The second thing I'll say is that much of
77:00 the the law that's being drafted by these state legislators are regulatory
77:07 oversight laws, not laws that define a new civil or criminal penalty because of
77:12 something you did that caused harm. They are specifically written in such a way
77:16 that they say we need to have oversight. we need to have review. We need to have
77:21 control over your systems because we get to review them. They don't say, for
77:26 example, if your AI kills someone, you are going to jail. That is what they
77:31 should say. And in fact, one could argue that much of the civil and criminal
77:35 statutes that already exist in the states cover much of the harm that is
77:40 already being talked about as the potential safety risk associated with
77:43 AI. You don't actually need more because at the end of the day if the AI system,
77:47 the producer of the tool, the user of the tool causes harm to someone or
77:52 something or some business, there is already statute to protect against that
77:56 harm. The statute that's being drafted is all about oversight. It is about
78:01 giving the government the regulatory control, the ability to go in and
78:06 interrogate and investigate and create approval systems on whether or not what
78:10 you're creating as a private market business or citizen is appropriate to be
78:14 used. And it is one of many points of overreach that this federated republic
78:19 has been able to withhold itself against historically. And after 250 years, the
78:25 day may be up. This makes >> so saxs in the case of a large language
78:31 model being constructed in a non-thoughtful way so that it could be
78:36 used to do cyber attacks and you know dox people or I don't know be used for
78:42 impersonation there should the law should be able to I'm trying to think of
78:46 a scenario here when they give these security things that would be concerning
78:50 and the law should >> I don't know if openai allowed their to
78:54 go hack credit cards That's already illegal, right? >> It's already illegal to to conduct a
78:59 cyber attack. And if you manage to take an AI model and use as a tool to perform
79:03 a cyber attack, that's still going to be illegal. Same thing in Colorado. Okay,
79:08 they've got this bill that they want to outlaw algorithmic discrimination, but
79:12 discrimination is already a violation of the law. So, what they're doing there is
79:17 they're not just going after the business that's performing discrimination. That's already illegal.
79:23 What they want to do is get into the tool itself, right? And they want to
79:27 make the developer liable. If their model creates an output that supposedly
79:33 ends up creating a disperate impact in a decision >> and imagine if we did this with the
79:36 internet. Imagine if we went back to the start of the internet and we said, "Hey,
79:39 if someone uses the internet to do something bad, therefore the government
79:41 needs to approve everything that's done on the internet." I mean, you can talk
79:45 about mobile communications. You can say, "Okay, Verizon's responsible if
79:48 people use it in a terrorist attack. Verizon's not responsible if people use
79:52 it to coordinate a bank robbery. That's so obvious. So, yeah, this does seem
79:55 like it's overreach. >> Zach, what is the situation on Capitol
79:59 Hill in having a conversation about creating federal preeemption, passing a
80:03 a bill that says the federal government's going to set standards
80:07 around AI utilization that states cannot kind of intervene on and creating a
80:11 mechanism that allows this market to to develop and allows things to prosper.
80:14 Well, here here's the situation is in the big beautiful bill, there was a a
80:19 federal moratorum on state AI regulation, and I think it was
80:23 well-intentioned and well motivated by the fact that we do see this huge
80:27 knee-jerk reaction to state legislators wanting to do something without knowing
80:30 what it is they want to do. However, there was not enough Republican support.
80:34 There wasn't enough Republican or Democrat support for it. And I think
80:37 that part of the reason why Republicans in particular have been opposed is just
80:42 because there's so much anger at the big tech companies right now for all the
80:47 censorship that happened during especially co but even before and you
80:51 still see it. You saw with this Wikipedia news where they're banning all
80:55 conservative publications from being sources. There's just a lot of anger
80:59 towards the big tech companies and tech bros and and basically there's a lot of
81:02 Republicans who don't want to get on board with anything that is perceived as
81:08 helping tech. Now the reality is who does that ultimately benefit? I mean
81:12 ultimately it benefits the blue states who are in the lead on this type of
81:17 regulation. It's Gavin Newsome who just signed this new bill. It's, you know,
81:21 again, it's Jared Polus in Colorado who ultimately signed this Colorado law. And
81:26 if and if there is no federal standard, what you're going to see is that the
81:30 blue states will drive this ban on quote unquote algorithmic discrimination,
81:36 which will lead to DEI being promoted in models, which is what the Biden
81:40 administration wanted. You will see the return of woke AI at the state level.
81:43 It's not something any Republican should want. I mean, I understand the the
81:47 justifiable anger at these tech companies because their behavior in the past has been
81:53 really bad towards conservatives. I mean, they did engage in a lot of
81:57 censorship, shadow banning, demonetization, debanking, all that kind
82:01 of stuff. So, I get it. But we have to look at what the results are going to
82:05 be. And the single federal standard is the best way to make sure that we do not
82:10 have woke AI, that we do not have insanely burdensome regulations that
82:14 allow China to basically get ahead of us in this AI race. And it's to ensure that
82:20 we actually have truthful, unbiased AI instead of highly ideological AI.
82:24 >> Do you think you can get it done? >> Let me go to poly market. The US enacts
82:29 AI safety bill in 2025. Not getting done this year. >> Well, here here's the good news. It
82:33 doesn't really matter what what I think. The important thing is what President
82:37 Trump thinks. And in his July 23rd speech on AI, he was really clear that
82:41 there needs to be a single national standard for AI. He said it was
82:45 impractical. It doesn't make sense to have 50 different regulatory regimes and
82:51 that that could cost us the AI race. And he would like there to be a single
82:54 federal standard just like he promoted for vehicle emissions because again, we
82:58 didn't have a federal standard there. And then it was California taking the
83:02 lead and then the blue states set the standards. President Trump didn't think
83:06 that made sense for California to be setting the rules for the whole country.
83:09 So the feds preempted that. And I think we should do the same thing on AI.
83:12 That's what the president basically said in his speech. So I think the
83:15 administration ultimately will support this. And I think I think more
83:19 Republicans will come on board as they realize what the blue states are doing
83:25 here is not helpful for conservatives. is not helpful for having an unbiased
83:29 information environment. >> I'm torn on this one. I, you know, I
83:33 moved to the great state of Texas to get rid of, you know, to have certain
83:36 freedoms that we have here that we don't have in other states. And I I kind of
83:40 like the idea of states having certain rights, but I don't like the way these
83:43 laws are being written. So, I remain torn and the devil's going to be in the
83:47 details on this one. Chimath had to bounce. >> Well, do you do you like the Colorado
83:51 law? Would you like to have >> No, of course not. So that it's how
83:54 these laws are executed that, you know, are my concern, you know, and I had this
83:59 concern with gun rights in California, like you should have the right to own a
84:01 gun and then they're just like, well, you can't have a gun. Okay, well, you
84:06 know, and then the states have to go back and forth in these lawsuits to see
84:12 can New York City, San Francisco ban guns and one of the reason crime is out
84:15 of control in some of these places because homeowners can't have guns and
84:20 the stand your ground laws, etc., etc. And one of the nice things about this
84:23 country is you can pick a state where hey I want to live in a state where
84:26 abortion's legal. I I don't want to live in a state where abortion is legal. I
84:30 want to live in a state without taxes, state taxes, ones with taxes. You get to
84:33 choose. It's one of the powerful things and we get to debate these things in
84:37 real time. So I do have a concern of centralized government and overreaching
84:41 federal governments, especially with the way executive power is being deployed
84:46 these days from Obama to Biden and to Trump. This too much executive power in
84:49 my mind. So I have concerns on both sides of it, but you know it's this is a
84:53 devil's in the details of the execution and I trust you to come up with
84:57 something good as our civil servant. So come up with something good, Sax.
85:00 >> Well, we will. But you know, just just to go back to to one of your points on
85:03 states rights, look, there's a commerce clause in the Constitution. And the
85:06 reason that exists is to create a seamless national market economy. One of
85:11 the reasons why the US has such a strong economy, why it's the number one economy
85:14 in the world, is because we have a single national economy, which is the
85:19 largest market for products. Imagine if we had 50 separate markets, each with
85:23 their own rules and regulations. And then doing business in the US would be
85:26 like Europe. Remember, one of the reasons why the US dominated the
85:31 internet in the '9s is because if you launched a startup in America and you
85:35 won the American market, you were basically right there in terms of
85:38 winning the global market. Whereas, if you were in a European country and you
85:41 wanted your local country, whether it was the UK or Netherlands or France or
85:45 something, you would just want a small part of Europe and then you would have
85:48 to go figure out all the rules and regulations to get into just the other
85:52 30 European countries, never mind the rest of the world. So, it's that
85:56 seamless national market that's given our companies the scale they need to
86:01 then dominate across the world. And if you restrict that by making every state
86:06 have different laws for every product, we're going to lose that massive
86:08 advantage that we have. >> Here's the thing. You know, I look at
86:12 the car standards with which Chimath brought up Friedberg and uh you know,
86:16 Trump, I guess, doesn't want to have California having their own car
86:21 standards. That got rid of 70% of the pollution in California. I was in favor
86:25 of that. I wanted to see higher standards, not lower standards, because
86:29 I don't want to pollute. And the the the smog over California was just especially
86:33 Los Angeles was insufferable at times. Those standards which led the nation
86:37 which led the world did they add extra cost? Of course, but it made California
86:40 a great place to live because it's car culture there and people were dying and
86:44 taking years off their lives from the smog. So that's an example of it I think
86:48 working really well. And I am for cannabis regulation and for it being
86:52 legal. And California led the country in that whereas other states want to ban
86:57 cannabis and they don't want to have higher standards for pollution. I like
87:01 the fact that California led in those two ways. Now it's all in the execution
87:04 of course. And so >> the problem is that because California
87:07 is such a big market, those vehicle emission standards that may or may not
87:11 have been right for California apply to every other state because the car
87:14 companies can't manufacture different models for different states. Nor should
87:17 they have to. >> They did though. >> Practically they did produce different
87:22 models for different states. But yeah, it definitely >> you want you want to have different AI
87:26 models for every state. >> You want to have you want to have a DEI
87:30 model for Colorado. You want to have >> if the in the case of cars, I do like
87:35 the fact that they California did push the car companies to make cleaner cars.
87:39 Now, in the case of AI, that's why I was asking you which safety concerns you
87:43 have, cuz I'm trying to find a safety concern that we can all say is a legit
87:47 concern for AI, and we can't come up with one. So, that's the interesting
87:51 part about this is like they're obviously overreaching laws right now
87:54 because we can't come up with something where AI is going to jump out of the
87:57 computer and do something in the real world that regular laws don't account
88:02 for. I we can't come up with an example here and we're deep in this industry.
88:05 Can you come up with a single example of AI doing something bad in the world that we
88:10 should be concerned about that isn't covered by existing laws? I can't.
88:13 Somebody in the audience figures that out, please email me. Another amazing
88:18 episode of the All-In podcast. Great to see you, Chimath, who had to jump. David
88:23 Freeberg, and of course, my bestie, my bestie, David Zach, our Zar getting it
88:28 done in DC for the country. Well done, and we'll see you all next time on the
88:32 All Podcast. Bye-bye. will let your winners ride. >> and it says we open sourced it to the
88:43 fans and they've just gone crazy with it. >> Love you queen of quinoa.
88:52 [Music] >> Besties are Oh man, my habitasher will meet.
89:04 >> We should all just get a room and just have like one big huge orgy cuz they're
89:07 all just useless. It's like this like sexual tension that they just need to
89:16 >> Your feet. We need to get our [Music]