WEBVTT 1 00:00:00.080 --> 00:00:15.900 On y'all. Welcome back to Blockspace Live, and an explosive day it's turning out to be, Charlie. Leopold Aschenbrenner's Situational Awareness Fund was blown up and sold its entire public equities portfolio to Citadel. 2 00:00:16.620 --> 00:00:26.280 That's right, the wunderkind sank so that we could swim. The market is ripping- [laughs]... following this news, and we will be covering that as our lead story. 3 00:00:26.420 --> 00:00:34.020 Following that, we have Zach Shapiro of the Bitcoin Policy Institute on to talk about AI in the intersection of law. 4 00:00:34.760 --> 00:00:45.400 We will also be touching on Microsoft and Meta's most recent quarterly earnings and why Meta tanked following its earnings and Microsoft soared. 5 00:00:46.240 --> 00:00:51.500 The details are in the CapEx guidance and the earnings for their computation arms. 6 00:00:52.060 --> 00:01:05.620 We will also be covering how FERC, the Federal Energy Regulatory Commission, has given approval for TerraWolf to move forward with its acquisition of Chesapeake Data with its Morgantown acquisition. 7 00:01:06.090 --> 00:01:21.960 And then we will cap off the stream today with a note on Morgan Stanley estimating that roughly 65 gigawatts of load within ERCOT's Badge Zero could be identified as base load, and why that benefits some of the principal stocks in our coverage. 8 00:01:23.320 --> 00:01:33.900 Colin, remind you to switch your microphone input. And while you do that, Blockspace goes live every single weekday at 1:00 PM Eastern. 9 00:01:34.180 --> 00:01:47.340 We are compute's daily show, featuring quick hits on AI, data centers, emerging technology, and markets. Today is a fun day to cover all of those topics. 10 00:01:47.420 --> 00:01:57.120 If you like what you hear on the live stream, it turns into a podcast shortly after we wrap up. And if you like the podcast and you wanna get more Blockspace content, I have good news for you. 11 00:01:57.640 --> 00:02:07.220 We have a website with way more stuff, a lot of written content, basically the expanded, extended universe of things that we talk about on this live stream. 12 00:02:07.260 --> 00:02:21.960 This live stream is just the tip of the iceberg of Blockspace content. You can find that at our website blockspace.media. This show is brought to you by CleanSpark, NASDAQ listed ticker CLSK. 13 00:02:22.480 --> 00:02:32.950 More on CleanSpark later on in the show. Holy smokes, Colin, the market [laughs] is on fire. Are we, are we so back? We are so back. 14 00:02:32.950 --> 00:02:44.920 [laughs] I, I think, to, before we hop into Situational Awareness, Charlie, I do think that you should get the- Yeah... DI metrics charts up. Yeah. Because a lot of the names we cover today are up, like, 20% to 30%. 15 00:02:45.320 --> 00:02:54.320 Yeah. Keel up, like, 30, Iron up 30. Basically every sector within the AI play is ripping right now. It's, it's nuts. 16 00:02:54.720 --> 00:03:05.200 Uh, you cannot swing a dead cat without hitting a major AI data center, like, stock indice that is, that's not up 10, that's not up double digits. I mean, look at this. 17 00:03:05.280 --> 00:03:10.160 So this is DI metrics, it has, uh, stuff segmented by sector. 18 00:03:10.170 --> 00:03:25.360 Memory, the biggest recovery after the Korean stock market and Korean degens fully routed, as Ben Paladian yesterday was talking about, uh, up, the whole sector's up 20%. I mean, we're talking just names. 19 00:03:25.420 --> 00:03:35.880 Micron, SanDisk, absolutely wild. Neo clouds, other sector we kinda cover. Neo clouds and power shells up 19% and 17% today. 20 00:03:36.000 --> 00:03:50.430 Um, I mean, Cipher up 28%, Keel 29%, CleanSpark 20 per- It's just like, what is going... I know what's going on. We sacrificed Leopold to the stock market gods. 21 00:03:51.360 --> 00:04:02.020 Um, yeah, so power shells, again, the whole sector up just insanely. These are, uh, these numbers are not on the day, they're on the month, I believe. Still, it, it, um, 22 00:04:03.100 --> 00:04:12.840 this, these are daily moves, um, not, like, monthly. So if we were to, like, zoom out [laughs] a bit here- [laughs] Still looking at month... okay. On the month [laughs] get a look at the month. 23 00:04:13.400 --> 00:04:29.100 The month power shells are still down 28%. On the month, hyperscalers are down 10%. On the month, storage, uh, photonics down [laughs] 35%. On the month, memory appears to be down 45%, still. 24 00:04:29.670 --> 00:04:44.469 But that's, that doesn't matter. Today we dine in the halls of Valhalla because [laughs] the, things are ripping. Um, there's a lot of hope right now. And it, and we talked about this, Colin, because, um, like, yes, 25 00:04:45.380 --> 00:04:53.040 the, there was a lot of leverage in the system. Yes, the, there, some, at some time, some later date, something runs into a brick wall. 26 00:04:53.100 --> 00:05:07.659 But, you know, as Aragon on the, you know, at the doors of, uh, at the doors of, the Gates of Mordor says, "Today is not that day." Um- Today is not that day. In fact, today is a perfect day for this meme. 27 00:05:08.000 --> 00:05:18.960 [laughs] It really does feel like this. Yeah. [laughs] Things are, things are ripping, oh my God, or it's, you know, the, we're up 1% on the month of the quarter. 28 00:05:19.480 --> 00:05:33.250 But, um, yeah, it does just so, like, show how reactionary everything is and how, um, knee-jerky the market is. Uh, why is it knee-jerky? Well, because everybody was levered to the absolute tits. 29 00:05:33.290 --> 00:05:48.480 'Cause 25-year-old was levered to the teats. Yeah. So. Child prodigy, and Columbia graduate at 19 and valedictorian, I just reali- I just learned that today. I, no, um, bro, I mean, okay. Do we wanna explain who he is? 30 00:05:48.880 --> 00:05:57.200 This is Leopold Aschen- I, I think we'll just- Okay... brief, brief bio. Okay. He used to work at OpenAI as a researcher following his graduation from Columbia. 31 00:05:57.440 --> 00:06:05.040 He ended up getting let go from the company according to some sources, and then he started Situational Awareness, his hedge fund, two years ago. 32 00:06:05.540 --> 00:06:19.852 That has been the darling- Of implausible hedge fund stories within the TradFi ecosystem, 'cause they were up over 400% on the year before this rout happened. Yeah. 33 00:06:19.872 --> 00:06:28.102 Started with a few million in his pocket, grew it to billions. At one point, I believe you'll s- you'll find the number, something like 40 billion. 34 00:06:28.132 --> 00:06:38.772 Another little detail, he is engaged to Avital Balwit, who just so happens to be the chief of staff for Anthropic CEO Dario Amodei. 35 00:06:39.402 --> 00:06:46.892 And if you'd forgotten about the storied traumatizing history of crypto in the last down cycle, 36 00:06:48.432 --> 00:07:01.832 Leopold was the, uh, investor who led the philanthropy side of now routed crypto firm FTX. Wait, really? I didn't know that. Yes. 37 00:07:01.892 --> 00:07:15.092 His job prior to OpenAI, Colin, was being the philanthropy arm, leading the philanthropy arm of FTX. So with that, Colin- Charlie, don't make me put on my, don't make me put on my tinfoil hat. Yeah. 38 00:07:15.352 --> 00:07:23.892 The, the, the coincidences just continue to build, but we'll, we'll go ahead and get to the meat of this story. So this is coming from CNBC's coverage. 39 00:07:23.932 --> 00:07:33.472 You can basically type in Situational Awareness, and you probably have 100 different articles to choose from. But the TLDR of what went down was this. 40 00:07:34.092 --> 00:07:46.772 The hedge fund unwound its entire position for public equities at Situational Awareness, selling to Citadel, as the Wall Street Journal reported this morning. That came out right before we hopped on the horn here. 41 00:07:46.892 --> 00:07:57.002 As losses on AI stocks and st- and software shorts strained the roughly 20 billion to 24 billion that it had under management at the time of the liquidation. 42 00:07:57.712 --> 00:08:11.472 CNBC reports, per your previous comment, Charlie, that the fund peaked at the start of July at 45 billion, right before this current drawdown. So the single buyer, again, was Citadel. 43 00:08:12.272 --> 00:08:21.352 It, uh, absorbed the entire portfolio of Situational Awareness' public equities, and this comes after the hedge fund was margin called. 44 00:08:21.412 --> 00:08:30.281 According to CNBC, Bank of America, Goldman Sachs, and JPMorgan Chase were helping the fund satisfy margin requirements or cut positions in an orderly way. 45 00:08:30.292 --> 00:08:36.672 And more broadly, Goldman Sachs and JPMorgan have issued margin calls to hedge funds holding highly concentrated AI positions. 46 00:08:37.132 --> 00:08:47.132 So one thing I think that is important to contextualize within this entire event, Situational Awareness is going to take a bunch of blows today for the fact that they had to liquidate these positions. 47 00:08:47.152 --> 00:09:01.352 But the fact of the matter is, this is a part of a larger story where these levered hedge funds betting on the AI trade, if they were on the wrong side of that leverage, have been getting margin called, pro- are, are probably in the process or were getting liquidated during this entire rout, right? 48 00:09:01.452 --> 00:09:16.142 And so it's too early to tell, but we could have just seen a massive flush out of leverage that would be, in all rights, if it were thorough enough, a bullish signal for potentially the rest of the quarter. 49 00:09:25.742 --> 00:09:39.672 We- May- It's increasingly seeming like we just had a pretty savage leverage flush out as indicated by this story. So a few key notes from this before I toss it to you, Charlie, for some spicy takes. 50 00:09:41.172 --> 00:09:56.512 They have a position in Anthropic, and this was not touched according to CNBC reporting. The Situational Awareness team, their representatives specifically told CNBC that they did not sell the Anthropic stake. 51 00:09:57.172 --> 00:10:10.862 But CNBC did say that it wasn't clear if that's actually true, but it's, that's at least in contention here. So we only know so far that their public positions or positions in public companies have been liquidated. 52 00:10:11.502 --> 00:10:24.132 Now, CNBC names the largest holdings as of the most recent accounting that they did with sources on the inside as Nebius Group, Sandisk, Micron, and CoreWeave. The Sandisk and Mi- and these are outright positions. 53 00:10:24.172 --> 00:10:35.472 These aren't derivative positions like puts or calls. And that would make sense in terms of getting liquidated at this point because Sandisk and Micron have been getting hammered this month. 54 00:10:35.842 --> 00:10:46.611 And so if you look at that and you think that they had large positions there, that could have put pressure on the entire book. In fact, all four of these, as CNBC notes, are down more than 35% this month. 55 00:10:47.332 --> 00:10:55.092 As of the end of Q1, the most recent filing that we have from them is around May 19th, their 13F with the SEC. 56 00:10:55.612 --> 00:11:07.972 Their largest holdings or largest, uh, new positions were VanEck Semiconductor Trust, which also is having a really rough month at 2 billion, Nvidia at 1.6 billion. 57 00:11:09.152 --> 00:11:25.512 Um, this is a new position, but it's combined with Nvidia puts that were a top holding, and Oracle at 1.07 billion. So at the end of the day, Leopold and Situational Awareness out 58 00:11:26.912 --> 00:11:35.372 getting a pit stop currently, getting their wheels changed and their oil checked. They're trying to raise money though. They're- Probably trying to get their head totally cleared. 59 00:11:35.432 --> 00:11:44.272 I mean, their brains have been scooped out. They've, they be- yeah, I mean, [laughs] if they're in the shop, they're, they're, you know, they're, they're bent over the tire pile right now. 60 00:11:44.332 --> 00:11:56.772 Like it's, it's, uh, Ken Griffin, uh, owns their portfolio now, and this is a common Ken Griffin play. Um, s- out of nowhere, somebody's really levered, somebody's over their skis. 61 00:11:57.272 --> 00:12:11.322 Who is the, you know, pro- where does the buck stop? Who is the ultimate, like, predator in the American markets? It, it's Ken Griffin and, and Citadel. And so, uh, again, putting on my conspiracy hat here, um- 62 00:12:13.028 --> 00:12:27.248 You know, uh, the entire market basically started copy trading Leopold Aschenbrenner over the past two quarters. I myself started doing a little bit too. Um, and, uh, b- because it feels like the boy's just been winning. 63 00:12:27.268 --> 00:12:27.328 Yeah. 64 00:12:27.338 --> 00:12:40.068 And he's married, or, you know, either married to or about to marry to, supposedly, chief of staff of Anthropic, which, um, like has been a little bit of an open secret speculated that he's just got inside information. 65 00:12:40.108 --> 00:12:46.608 I mean, he was let go from OpenAI for sharing insider information. I mean, 66 00:12:47.668 --> 00:13:00.618 what, what is this young impetuous buck gonna not stop doing, especially if he's got an inside route onto, you know, the most speculated, insular, mysterious company in pal- at the tip of everyone's tongues? 67 00:13:01.208 --> 00:13:13.468 He's gonna keep doing it. So, um, again, this is speculation. This is me positing, you know, intu- you know, intuiting what other people have also basically said. 68 00:13:13.488 --> 00:13:21.768 Not surprised though that Ken Griffin is there to pick up the pieces. Whether or not this is opportunistic, intentional, or coincidental, we don't know. 69 00:13:22.328 --> 00:13:47.148 It wouldn't be surprise, it would not surprise me if Leopold had kind of become the flag bearer for a lot of people who are getting levered, and, and this is just maybe situational awareness is just the, like the prominent one of the people, uh, of, you know, the prominent one that, that went, uh, belly up and just there's more other bodies floating to the surface because of the copy trade. 70 00:13:47.228 --> 00:13:52.738 So- Yeah, I wouldn't be surprised-... maybe that was, yeah... if we see other bodies floating to the surface, but it's almost like, you know, if you see an- 71 00:13:53.858 --> 00:14:01.488 a, a sperm whale float to the surface, you're probably not gonna notice all the other minor fish that are also surfacing as well. Yeah. 72 00:14:01.528 --> 00:14:15.208 My question going forward though, Charlie, is they are seeking to raise new capital to restart the fund. What, will people trust Leopold's situational awareness, dare I say again? 73 00:14:15.988 --> 00:14:26.568 You know, it's the, it's the classic fool me once, shame on you. Fool me twice, shame on me. I would imagine that there's probably a good chance that they do end up raising capital again. I bet you they, yeah. 74 00:14:27.268 --> 00:14:38.488 But- He's, yeah... but I mean, well, what, what's your, what's your take? Do, do you think that w- Do, do you think that his star has risen as far as it will? It, it, it was at a zenith, now it's at a nadir. 75 00:14:38.568 --> 00:14:44.828 Do you think that they might come back, but they just won't feature as prominently as they have in the past? 76 00:14:46.148 --> 00:14:57.108 Um, I think the mystique and mystery will be gone 'cause everyone realizes, oh, he was just levered during the right time to be levered and had all the situational attention, if you will. 77 00:14:57.708 --> 00:15:09.928 But, uh, I bet you he gets another shot at the, I ga- I bet you he gets another shot on goal. I mean, the thing is he was right during the right time with conviction. So it really, it, the irony is it was really, 78 00:15:11.588 --> 00:15:20.108 you know, obviously this was driven by hubris and risk, but that is downstream of being able to be sol- solve for that with, uh, position sizing. 79 00:15:20.588 --> 00:15:29.427 So if you can just kinda [laughs] turn the dial down a bit, if you think that this trade, that AI trade has room to run, Leopold probably has good insider information. 80 00:15:29.468 --> 00:15:37.948 I mean, do people stop texting him, you know, when they're, you know, ahead of their deal announcements? 'Cause, like, he prob- he's still got everyone in the industry's number. 81 00:15:38.008 --> 00:15:45.598 And his, his, his essay, Situational Awareness, basically was right and it has been right, basically saying- Mm... 82 00:15:45.648 --> 00:15:55.148 that there's a couple hundred people in Silicon Valley a few years ago who, like, have total vision to the entire playing field, and it really is a big group chat that you're not in. 83 00:15:55.748 --> 00:16:10.828 And, um, that only s- serves to be more right every day that this is a small amount of people with a lot of insight and agency who are basically driving the entire market. That's not to be like s- you know, call a cabal. 84 00:16:10.888 --> 00:16:23.688 I'm just saying the fact that what now drives the entire stock market was basically kind of a niche thing two and a half, three years ago, and, uh, now they're all in charge of multi-billion dollar, approaching trillion dollar companies. 85 00:16:23.748 --> 00:16:38.728 So, like, just the fact that a few engineers now kind of derive the, are, you know, drive the machine of the American and global economy, you know, he's gonna get another shot on goal is my prediction. Um... 86 00:16:38.968 --> 00:16:51.048 M- my biggest- Okay... question last before we sign off on this one, Charlie, is, like, what is left out of what was sold? Like- Well, we don't really know... after s- No, I don't- We- That wasn't in the reporting. 87 00:16:51.568 --> 00:17:10.278 So after satisfying margin requirements and clearing anything else out they needed to, how much of a haircut did they take on the positions in addition to what, uh, they already took in terms of the nominal value, considering where the stock market has, uh, considering how these positions have drawn down? 88 00:17:10.368 --> 00:17:22.368 And, like, what's left over? Do they have... I mean, 'cause you would assume that there's gotta be something. They were up, like, 430% on the year- With sides... at a certain point. Yeah, and so what's left over? 89 00:17:22.528 --> 00:17:35.608 You have to imagine there are a few billion dollars left over there. I guess that just gets returned to the, um, the LPs in the fund. I mean, I assume so, or- I mean, we'll see... 90 00:17:35.868 --> 00:17:47.488 I don't, I don't know what, what comes next. Like, do you go to the LPs and you're like, "Hey, like, you can take your money back, or you can let us try it again," you know? But, uh, definitely the 91 00:17:48.428 --> 00:18:02.308 s- I mean, from where I'm standing, Charlie, this is one of the craziest stories of the year so far, specifically because after they liquidated this position, everything, including the holdings that they had, ripped. 92 00:18:02.928 --> 00:18:13.948 So Ken Griffin and Citadel looking like geniuses right now for buying that book at a haircut. Add that to the stack of other things that they've done over the years which make them look like geniuses. 93 00:18:13.988 --> 00:18:40.224 You know, people on the internet may remember, and we were talking about this before we went live, Ken Griffin is the boogeyman, um, the ultimate arbiter of WallStreetBets, having basically been the entity that, uh, WallStreetBets failed to liquidate, who then- Uh, called all the hens home, all the cows home and said, "Ah, your little online, uh, populist movement, it actually belongs to me." 94 00:18:40.584 --> 00:18:48.824 Now Ken Griffin is building the largest palatial estate in Am- American history, either on Long Island or somewhere in Florida. I don't know which, probably both. 95 00:18:49.284 --> 00:19:02.034 But Ken Griffin is like now, you know, he is the final boss, [laughs] uh, of people that we have names of. So, uh- And one quick aside- That's- Remember when GameStop announced a Bitcoin treasury strategy? Oh, man. 96 00:19:02.084 --> 00:19:11.584 What happened to that? I feel like, like Kermit the Frog- [laughs]... at the, you know, rainy window. [laughs] Like saying, "I former- "I miss, I miss those days." Yeah. 97 00:19:11.624 --> 00:19:24.864 [laughs] All right, I think we can table this one and, and move on to earnings. Yeah. Um, before we do that, a word from our sponsor, CleanSpark. [gentle music] We are CleanSpark, 98 00:19:25.784 --> 00:19:38.264 America's Bitcoin miner, a publicly traded company with the largest operating hash rate, powered entirely by self-operated infrastructure across four states. 99 00:19:40.024 --> 00:19:56.384 This is our proof of work, and we are setting the standard for what's next. Learn more about the intersection of energy and Bitcoin at cleanspark.com. All righty, Charlie. 100 00:19:56.504 --> 00:20:05.364 I don't have a one, one article to share for this 'cause it's just a bunch of numbers that are recapped from the earnings reports, but a tale of two stocks. 101 00:20:05.924 --> 00:20:18.864 A tale of two hyperscalers, Charlie, because yesterday Meta and Microsoft both released their quarterly earnings, and they both had the exact inverse reaction in the market as a result. 102 00:20:18.924 --> 00:20:30.184 Meta dumped 7 to 8% in after-market hours. Meanwhile, Microsoft was up 9% in after-market hours. And in fact, that, 103 00:20:31.264 --> 00:20:44.764 uh, rally has extended into today as all of these stocks are bevied by Leopold Sacrifice. Microsoft up almost 17% today, which is crazy. 104 00:20:44.804 --> 00:20:59.284 Now this might be cold comfort for Microsoft holders because it's still down like 11% on the year, and it's down 6% year to date. In fact, it's been one of the laggards among the Mag Seven in terms of the AI trade. 105 00:20:59.884 --> 00:21:16.004 But yesterday's quarterly earnings might rewrite that script. Meanwhile, Meta down 12.3% on the day, and down 23% on the year, 19.3% year to date. 106 00:21:16.444 --> 00:21:30.944 So what's going on with these stocks? So more or less what came out of the earnings call yesterday was a warning on CapEx guidance and CapEx spend without clear operating income and profits from that. 107 00:21:31.484 --> 00:21:33.844 So the headline numbers are as follows. 108 00:21:34.304 --> 00:21:46.624 Microsoft f- uh, fisc- uh, their Q4 fiscal year, they do this awful thing like, uh, you know, other companies that we cover where for whatever reason their fiscal year ends halfway through the year. 109 00:21:47.324 --> 00:22:01.384 They were at $90 billion last quarter in revenue, up 18%. Meta beating them on the percentage front here, $60.8 billion, up 28% quarter over quarter for their Q2 2026. 110 00:22:02.064 --> 00:22:13.014 Earnings per share adjusted for, uh, for Microsoft was $4.74 per share. Meta $6.18 per share. 111 00:22:14.384 --> 00:22:25.074 Quarterly CapEx guidance wa- or quarterly CapEx for the quarter was 48, 41 billion for Microsoft, up 69% and 31.1 billion for Meta. CapEx guidance was cut 112 00:22:26.404 --> 00:22:34.724 from Microsoft from 190 billion to 175 billion, but this isn't actual spend, Charlie. This is an accounting gimmick that I'll get into here in a second. 113 00:22:35.364 --> 00:22:43.244 Whereas Meta raised the lower end of its CapEx guidance for 2026 from 125 billion to 130 billion. 114 00:22:43.784 --> 00:22:56.444 Now the key bit here though is that Microsoft Azure Cloud, their match, their re- Microsoft Azure Cloud revenue was up 43... Or the, uh, it grew 43% over the quarter. 115 00:22:57.584 --> 00:23:08.474 And that was really I think what was leading to the rally following earnings. There's, th- people, they're seeing clear growth in the cloud segment, and that's what investors wanna see right now. 116 00:23:08.504 --> 00:23:18.704 Their commercial RPO backlog also rose to 678 billion, up 8% sequentially and up 84% year over year, which is absolutely huge. 117 00:23:19.544 --> 00:23:35.144 And full year revenue was 338, 331 billion, up 18%, with operating income up 21% to over 155 billion. Now, let's contrast this with Meta. So their CapEx was 31 billion. 118 00:23:35.154 --> 00:23:43.414 Their full year guidance for CapEx was 130 to 145 billion. They beat on revenue, but they missed on earnings, and they got punished accordingly. 119 00:23:44.364 --> 00:23:58.304 Now it seems to me, Charlie, like the market is really truly drawing a line here by saying, "You can spend on CapEx, but we need to see that the key business segments that are going to benefit from that CapEx are actually growing." 120 00:23:59.244 --> 00:24:12.844 And it's a little bit misleading some of the reporting about Microsoft cutting CapEx guidance, because what it actually did was they extended the estimated useful life of their data center buildings from 15 to 25 years, 121 00:24:13.944 --> 00:24:19.064 which reclassifies some of the future leases from CapEx to operating expenses. 122 00:24:19.124 --> 00:24:32.944 So there's like balance sheet shuffling going on here to where they can move some of those expenses from the CapEx bucket to the operating expense bucket. And the reason for that is that, um- 123 00:24:34.340 --> 00:24:49.440 When you extend the useful life of the data centers, it changes the ratio of lease term to asset life for these assets, and it reclassifies some leases that were previously qualified as financial leases, which falls under CapEx, into operating leases, which is OpEx. 124 00:24:50.200 --> 00:25:00.880 So that's why the Microsoft CFO, I believe her name's Amy Hood, said on the call yesterday, "We're revising down the accounting guidance for CapEx, but the spend is going to remain the same." 125 00:25:01.400 --> 00:25:05.660 But the market doesn't seem to care about that because Azure had very strong growth. 126 00:25:05.900 --> 00:25:20.800 Well, and, and again, not the finance wizardry does elude me a little bit, but, uh, maybe the market sees through that and sees that structurally right now, Microsoft sells compute and Meta consumes compute. 127 00:25:20.810 --> 00:25:30.640 They are both... They're both... They both are, you know, annihilating their free cash flow [chuckles] and spending, you know, CapEx towards these endeavors, but Meta currently does not sell much compute. 128 00:25:31.100 --> 00:25:42.320 However, you know, they, they, they have said we're, we're pivoting to the, uh, the, the neocloud model or we're gonna sell compute, which you would think at some point would be the right move. 129 00:25:42.460 --> 00:25:56.640 Question is, how quickly can they roll that out? If my, you know, if my number, if, if my understanding is correct, Meta has the Hyperion site down in Louisiana, um, ETA early 2028. 130 00:25:58.360 --> 00:26:08.520 And then they just announced their other massive, massive site down in El Paso, as we covered yesterday or the day before. Again, a Hyperion model round two. Um, 131 00:26:09.640 --> 00:26:16.600 you know, while Microsoft may not be expanding quite at this rate. We'll see. 132 00:26:16.660 --> 00:26:24.390 At some point you, you know, I do wonder was Leopold Aschenbrenner also short Microsoft and now they're [laughs] going up again 'cause- Yeah... bro, I mean, they're, like- Yeah... 133 00:26:24.430 --> 00:26:37.080 the M in the Mag 7, they're up like 12, 14%. It's, it's wild. Yeah. So. And, and just to underscore that point really quickly with a few more stats about Azure. Azure's revenue is, grew 43% year over year. 134 00:26:37.500 --> 00:26:53.100 For the full fiscal year 2026, Azure revenue surpassed 100 billion for the first time. So clear growth on that cloud segment, and that to me seems to be the key driver of the current rally in Microsoft. 135 00:26:53.960 --> 00:27:06.480 Investors no longer wondering if they're going to be left behind in terms of the principles driving this AI rally, so. All right. We have Zach Shapiro in the wings. 136 00:27:06.500 --> 00:27:11.460 We're gonna bring him on up here very soon after a word from our sponsor, Luxor. 137 00:27:12.250 --> 00:27:24.080 [static] This episode of Block Space Live is brought to you by Luxor's Commander, Bitcoin miner management software for enterprise operations. 138 00:27:24.620 --> 00:27:30.230 Luxor's Commander gives you real-time fleet monitoring, bulk remote commands across your fleet, and Intelligent Miner. 139 00:27:30.710 --> 00:27:37.849 That's an automated profitability engine that runs every five minutes and tests your fleet's power settings against live energy and hash rate markets. 140 00:27:38.160 --> 00:27:52.620 In fact, ERCOT backtests show 10% improved profitability with intelligent mining versus binary mining. Commander Pro is $100 per megawatt or a 25 basis point pool fee adder, but you can also try it for free for 60 days. 141 00:27:52.680 --> 00:28:10.100 So if you'd like to learn more, go to luxor.tech/commander to get started. All right. Let's bring Zach on up. Adding him to the stage now. Zach Shapiro, welcome to the show. Thanks for having me. Good to see you. Um, so, 142 00:28:11.020 --> 00:28:22.570 uh, had, gonna have you on up here. You're kind of like everybody's AI lawyer whisperer here. [laughs] And, and I wanna talk and, you know, about your essay, uh, two clocks and associated things. 143 00:28:22.660 --> 00:28:33.740 But before that, the kind of, the story that I am still seeing, like, you know, on Main Street, my friends who are not in, really into AI, they're all angry about the Anthropic, uh, settlement. 144 00:28:33.900 --> 00:28:45.860 I don't know if you, if you, how familiar you are with this. Anthropic pirating, stealing all these books. Um, I know you're more dialed into law and AI. What's your take on this? 145 00:28:45.920 --> 00:28:52.099 What, you know, what's the landscape here? Yeah, I think this is a, um, narrower story than it sounds like at first. 146 00:28:52.400 --> 00:29:07.440 Uh, there is a big open legal question about to what extent can, uh, Frontier labs train on stuff on the internet, uh, and then use it under fair use protections, which is an exception from copyright law, uh, in order to train its models. 147 00:29:07.520 --> 00:29:11.650 And there are, you know, there's a difference between different ways of doing this. 148 00:29:11.680 --> 00:29:22.140 So, um, right now there's a lawsuit, a big lawsuit between The New York Times and OpenAI about the big question of like, was it okay that, uh, OpenAI trained its ChatGPT models on The New York Times? 149 00:29:22.160 --> 00:29:29.200 I think everyone agrees if ChatGPT were to just serve up, uh, New York Times articles on request, that would be illegal. That would be copyright infringement. 150 00:29:29.240 --> 00:29:38.239 The open question is, if you ask ChatGPT a fact that it learned from reading a New York Times article, uh, whether that is permissible or that's copyright violation. 151 00:29:38.340 --> 00:29:42.260 Um, my view, I, this is an unsettled area of law, my view is that it should be okay to do this. 152 00:29:42.560 --> 00:29:54.360 I think that LM's training on information from the internet is really more like, uh, you know, learning how to play piano from a piano teacher than it is playing a recording of someone else playing piano, and that should be the distinction, and sh- you should be able to train models. 153 00:29:54.760 --> 00:30:06.000 Um, that's an open question. Uh, what the Anthropic settlement is about is not quite that. It is, um, can you use pirated material to train the model, right? Can you steal and then train on it? 154 00:30:06.100 --> 00:30:10.300 And, uh, the answer to that, unsurprisingly, is no, and that's, that's the settlement. 155 00:30:10.380 --> 00:30:19.400 Um, there is sort of a parallel case where it's like, okay, if Anthropic bought a single copy of a bunch of books and trained Claude on it, um, is that okay and can it use the information from the book? 156 00:30:20.040 --> 00:30:27.640 Import- interesting, important opening question. If you steal the book in the first place, um, that's a slightly different scenario, even though sort of functionally it's similar. 157 00:30:27.680 --> 00:30:36.260 So I wouldn't read too much into this settlement. I don't think this is one of the sort of big open questions. Good. I'll clip that. I'll send that to my friends. I think that's a pretty concise way of doing it. 158 00:30:36.980 --> 00:30:48.262 Um, so let's get into kind of the meat of, uh, this conversation. You've published a bunch of essays, all fantastic. Uh, earlier this year, I believe it was like the Claude-powered... law firm. 159 00:30:48.312 --> 00:30:51.672 But again, that was [laughs] like eight months ago, so things have changed. 160 00:30:52.292 --> 00:31:03.552 Your latest, uh, essay, The Two Clocks, describes the, um, the cadence, different cadences of the, uh, the technology and the institutions, the law firms. 161 00:31:04.092 --> 00:31:09.612 Can you give me the short TLDR elevator pitch of this, and then we can ask some follow-up questions? Yeah. 162 00:31:10.112 --> 00:31:47.812 Uh, when I look out at the sort of economy and the macro situation, uh, the sort of most glaring fact to me, which I think is the biggest sort of high-conviction opportunity I see in the market, is that on the one hand you have these, uh, giant tech companies that are, um, even now, after the recent correction, trading at pretty, uh, incredible multiples and rich valuations, um, based on the promise that they have discovered this sort of amazing technology that is perhaps swiftly leading to AGI, and that is going to transform the economy over the next few years in a major way and justify huge amounts of CapEx on token spend by all sorts of businesses, right? 163 00:31:47.872 --> 00:31:52.052 Not just tech companies, but really, uh, sort of all white-collar type of businesses. 164 00:31:52.532 --> 00:32:02.032 Um, and on the other hand you have, uh, the industry which is starting to adopt this stuff and, and starting to spend money on tokens, and sometimes, you know, even token maxing and spending a lot of money on tokens. 165 00:32:02.312 --> 00:32:12.972 But you're not really seeing the increased revenue or increased effinci- efficiencies or increased productivity, uh, you know, outside perhaps of big tech, which is seeing more of this leverage. 166 00:32:13.392 --> 00:32:24.752 Um, and you know, this is a problem, right? If the efficiencies don't actually materialize, then, then w- you know, the current valuations may start to look like a bubble, at least for now. 167 00:32:25.192 --> 00:32:34.112 And so the tech companies, uh, sort of drastically need to see efficiencies in the regular economy in order to justify, uh, their projections. 168 00:32:34.172 --> 00:32:46.762 The, uh, businesses in the real economy are having to face, um, AI native startups that are doing things more efficiently than incumbent businesses can, and, uh, questions from their customers and clients about where is the AI efficiency. 169 00:32:47.562 --> 00:32:57.311 And so both parties really need this gap to close. And my thesis is that, um, each of the parties is seeing something real, um, but missing something else. The tech companies are right about the technology. 170 00:32:57.352 --> 00:32:58.572 The technology is really amazing. 171 00:32:58.912 --> 00:33:09.212 It can do all sorts of stuff, and if AI models don't get any smarter than they are now, they are smart enough to deeply transform white-collar industries like law, accounting, finance, and stuff like that. 172 00:33:09.612 --> 00:33:16.912 Um, what they're wrong about is that this technology will automatically just work and diffuse in the economy without, uh, a lot of effort. 173 00:33:16.972 --> 00:33:26.792 Um, now, the businesses, right, the law firms, the accounting firms, the investment firms, um, they are right that the technology is not magic and it is not fully working for them yet. 174 00:33:26.812 --> 00:33:37.612 But their conclusion that they tend to draw from that, which is the technology is over-hyped, uh, and, you know, is, like, really fancy auto-complete, and it's, you know, it's a neat parlor trick but it's not ready for prime time, that's wrong, right? 175 00:33:37.732 --> 00:33:44.292 Uh, the problem is it's a skill issue. You actually need to figure out how to use this technology well and how to reorient your business around that. 176 00:33:44.302 --> 00:33:48.252 And so I see this gap, you know, these two different clocks that move at different speeds. 177 00:33:48.312 --> 00:33:59.552 There's the speed of the technology, which is genuinely impressive and unprecedented and fast, and then there's the speed at which institutions move, which is slower and sort of the biggest, you know, necessary thing that needs to happen in the economy. 178 00:33:59.652 --> 00:34:05.352 And, uh, the area of the economy I'm most bullish in is converging those clocks, uh, towards each other. 179 00:34:06.392 --> 00:34:21.532 Do you think people's expectations are just out of sync because of that, in the sense that the tech industry is probably overestimating how quickly we will get to AGI, and the private sector is saying it's gonna take way too long, and they're just a mismatch? 180 00:34:22.112 --> 00:34:29.152 Because I often see what you said about the... People say, "Oh, it's just a fancy auto-complete." And if you wanna use an analog, it's like, well, you know, 181 00:34:30.092 --> 00:34:38.512 um, you know, dial-up internet was really freaking slow, and now we have satellites in space that can beam you Wi-Fi anywhere on Earth. Yeah. 182 00:34:38.542 --> 00:34:48.812 And it, it seems like there's one sector of the economy that thinks it's just never gonna get there. Yeah. I, actually, uh, that might be true. Um, but I actually don't think this is a question about AGI. 183 00:34:49.372 --> 00:34:55.322 Uh, we don't need to get to AGI for the AI to be deeply transformative. It's, it's good enough already, right? 184 00:34:55.392 --> 00:35:06.372 Like, I run a three-person law firm, um, that is AI native, and the things that I do and the efficiencies I get and the leverage I get, uh, are just way, way better than big law firms. 185 00:35:06.412 --> 00:35:12.252 And I have a little bit of authority to say this 'cause now I'm also spending a lot of my time inside big law firms, helping them figure out how to diffuse this stuff. 186 00:35:12.272 --> 00:35:22.152 And it's just the, the capabilities are different when you don't have institutional constraints. Um, what needs to happen is not, like, to figure out exactly when AGI's gonna happen and then that'll solve it. 187 00:35:22.252 --> 00:35:30.672 Uh, what we need to figure out is how to re-architect enterprise in order to incorporate the technology that already exists today, which is good enough. 188 00:35:30.732 --> 00:35:41.771 And, and in my piece I use this historical analogy of, um, when industry started to electrify 100 years ago. Uh, you know, electricity had been around for certain applications. Obviously, it's impressive. 189 00:35:42.562 --> 00:35:52.392 Um, but, uh, early sort of Industrial Revolution factories ran on the steam engine. Uh, what happened was... You know, so factories were oriented around a central drive shaft. 190 00:35:52.452 --> 00:35:55.902 In the basement of the factory, you'd have this big steam engine that sort of chugged along. 191 00:35:55.972 --> 00:36:06.292 Uh, you'd have the drive shaft, and then you would attach all of the machines in the factory to that central drive shaft along the assembly line, uh, in order to power all of the different machines that needed to work in the factory. 192 00:36:06.322 --> 00:36:19.812 And when, uh, the economy started to electrify, what they did, which is totally understandable, uh, is to, uh, take that steam engine motor in the basement, replace it with an electric motor, uh, and then use that to power the drive shaft. 193 00:36:19.832 --> 00:36:28.472 And, uh, you know, maybe you saved a little bit of money not having to have folks s- shovel coal in the basement. But, you know, the economists at the time were like, "Oh, my God, electricity's gonna change everything." 194 00:36:28.812 --> 00:36:34.932 And when you replaced the steam engine with the electric motor for the drive shaft, it, it didn't change everything, it just made things a little bit more efficient. 195 00:36:35.332 --> 00:36:46.222 Um, and the efficiency came 30 years after that, when they realized, oh, that's not what electricity is for. It's not to drive the drive shaft. What we can do with electric motors is put them in every individual machine. 196 00:36:46.232 --> 00:36:52.632 And so they actually tore up the factory floor, and they just remade the assembly line. It doesn't need to be along a central drive shaft anymore. You can... 197 00:36:52.772 --> 00:37:01.172 All these machines that have their own electric motors, they can be electrified, and now this is, like, a fundamentally different thing that is much more efficient and much more powerful. 198 00:37:01.252 --> 00:37:07.432 That is what needs to happen in industry, right? Like, the AI is here in the way that, in the way that the electricity was here. 199 00:37:07.812 --> 00:37:27.555 It's that the sort of incumbent enterprises are designed around, um, You know, assumptions about the scarcity of intelligence and, you know, what tasks require humans and what levels of review are needed that is just not compatible with the, like, truly agentic AI future. 200 00:37:27.896 --> 00:37:31.096 And so it, it's really much more like a reorganization that needs to happen. 201 00:37:31.136 --> 00:37:42.636 And so you need much more investment, and some of it will come from the labs, and some of it will come from industry, and some of it will come from, you know, consulting or change management businesses, um, but to actually sort of re-architect the way that firms work writ large. 202 00:37:43.976 --> 00:37:54.196 You talk about re-architecting, and you say that you are now, um, you know, working inside other large law firms. And in your piece you talk about, like, the forward deployed absorption role. 203 00:37:54.956 --> 00:38:05.916 Calling back, I think, Palantir is where I, you know, see the forward deployed engineer or, or some-something like that. Um, is that part of what you think rearchi- architecting needs to look like? 204 00:38:05.996 --> 00:38:16.056 Is that a short-term bridge, like, period where we do this as we approach AGI? And then, like, you know, the drive shaft, great example, but we're talking about intelligence now. 205 00:38:16.116 --> 00:38:29.176 This is a very much less tangible, amorphous idea, really hard to quantify. What does the architect- re-architecting look like, Zach? Yeah. Uh, great question. So I think this is roughly a 10-year project. 206 00:38:29.276 --> 00:38:34.936 Um, I think beyond that, it's really hard to know what this all looks like. I don't think we're gonna get to AGI in the next couple of years. 207 00:38:35.236 --> 00:38:46.876 If we do, then this all might be out the window and we enter a singularity, and who knows? Um, but assuming that that doesn't happen, um, w- the forward deployment, uh, what that really is, is you need to have... 208 00:38:47.116 --> 00:38:54.856 It might be one person, it might be teams of people, who understand both the actual substantive task. So this can't be McKinsey. I don't think this will be Palantir. 209 00:38:54.896 --> 00:39:01.456 You need actual subject matter experts, um, you know, in law, accounting, [audio glitches] what call, uh, industries. 210 00:39:01.896 --> 00:39:25.086 Um, and then you need that person or someone else who closely works with them to understand how to talk to AI, how to translate the substantive task into essentially prompting and then custom skills and then, you know, automated workflows, which might be some amount of generative AI and some amount of sort of more deterministic software sort of built around it, right? 211 00:39:25.136 --> 00:39:36.495 Not every task is, uh, meant for AI. Um, I have a, uh, basically a, a software tool I built to help automate, uh, venture deals, uh, for startups when they raise money. 212 00:39:36.836 --> 00:39:40.396 And, uh, part of it is just a plugin to Claude CoWork. 213 00:39:40.796 --> 00:39:50.476 Um, but part of it is also a separate deterministic software engine that I vibe coded that basically does all the math of, like, share counts and price per share and, you know, uh, creating the cap table. 214 00:39:50.756 --> 00:39:55.116 That's not something you wanna hand to an LLM because, you know, an LLM might hallucinate. 215 00:39:55.136 --> 00:40:00.736 It'll give you a different response every time, uh, you put it, this, the same input in, and that's no good when you're trying to do math. 216 00:40:00.816 --> 00:40:10.056 Um, but if I'm trying to add, like, legal language into the deal documents to capture a specific intent, um, you need an LLM for that, right? 217 00:40:10.076 --> 00:40:13.836 That's a qualitative request that you're not going to have, like, a logic tree do. 218 00:40:13.876 --> 00:40:20.916 And so some of this is gonna figure out what is AI, what is not AI, how do these things work together, what, if any, is the UI behind all of this. 219 00:40:21.916 --> 00:40:44.936 Um, but the forward deployment is really about translating the intent of the subject matter experts into sort of agentic automated workflows over time at each level of the stack, whether that it's at the most basic level, a prompt or a skill that is, that is built by looking at a ton of different prompts or, you know, some fancier software tool that, that, like, is a hard shell around sort of the skill workflows. 220 00:40:45.356 --> 00:40:52.816 Uh, and then the end state of this, um, [lips smack] done a lot of thinking about that also. Uh, and, and here I'll be more specific to law. That's, that's the area I know best. 221 00:40:53.136 --> 00:40:56.396 Uh, I think that law is going to bifurcate into two different markets. 222 00:40:56.716 --> 00:41:04.696 I think there is sort of, like, commodity legal work, which is y- you know, like, regular low-value contracts for big business where you do bulk review. 223 00:41:05.056 --> 00:41:08.976 I think that is just gonna fully, uh, automate and commoditize to the price of the token. 224 00:41:09.396 --> 00:41:17.376 Um, there will be the minimal amount of human review that regulations allow for, and, you know, that is what it is, and people will make money, uh, fully automating that. 225 00:41:17.486 --> 00:41:26.156 Just like they've made money, um, sort of shunting a lot of that work to LPO offshore sort of low-cost review in India. Um, it'll just be machines instead of people in India. 226 00:41:26.496 --> 00:41:33.515 Um, at the high end though, I think what is... And that this is where I'm more interested, and the reason why I am sort of embarking, uh, to, to work with big law firms on this. 227 00:41:34.406 --> 00:41:36.716 [lips smack] Uh, I think, you know, if you're hiring, if you're a, 228 00:41:37.656 --> 00:42:06.176 you know, wealthy person, a big company, investment firm, and you're, you're hiring the very best law firms in the world to help you, whether w- you know, it's like a bet-the-company litigation matter that's really important to you, it's a important sort of white collar criminal defense case where you might go to prison if you lose, uh, whether it is sort of like headline grabbing M&A where a huge amount of money is at stake, um, you are hiring the law firm to get access to the very best lawyer at the top based on that lawyer's judgment, their ability to make good decisions in the face of uncertainty. 229 00:42:06.306 --> 00:42:14.056 Uh, and I, I don't see LLMs as being really good at that. LLMs are great at execution when you specify a task well enough in a prompt. 230 00:42:14.476 --> 00:42:29.296 Um, they're very good at giving you information and talking through stuff with you and helping with analysis, but, uh, I wouldn't want an LLM to make decisions for me, and I just don't think that, like, because they are prompt-based, because they're generative, I just don't think that's what they're great at, so there will be a human. 231 00:42:30.256 --> 00:42:37.536 [lips smack] And sort of the promise and the challenge of that for, like, big law firms specifically, uh, is the economic model for big law firms right now. 232 00:42:37.896 --> 00:42:45.716 Um, you know, if you look at the sort of partners at the top of big law firms, sometimes they're making, you know, $10 to $25 million a year, um, per partner. 233 00:42:46.116 --> 00:42:55.186 Uh, and y- yes, they bill out at very expensive hourly rates, you know, two to $3,000, sometimes a little bit more than that. Uh, even [audio glitches] rates, there are not hours in the day 234 00:42:56.976 --> 00:43:01.896 to earn their yearly draw just by working even at very expensive hourly rates. 235 00:43:02.116 --> 00:43:09.876 The way they actually make money is billing out armies of junior lawyers at, like, $1,000 an hour on, you know, doing grunt work on big tasks, right? 236 00:43:09.916 --> 00:43:14.856 In a litigation matter, there's doc review, just a, a bunch of documents you need to read to see if they're relevant evidence. 237 00:43:15.256 --> 00:43:23.996 Uh, in a, in a, you know, large transaction, there's due diligence where you gotta look at all of the company's contracts and stuff to see if there's something in here that's gonna blow up the deal. 238 00:43:24.416 --> 00:43:35.856 And, um, that is called the leverage model, right? It is, it is having many non-equity partners, right, junior associates, mid-level associates, do a bunch of grunt work that all feeds into the profits for the partners. 239 00:43:36.892 --> 00:43:52.692 Um, now you'll notice this creates sort of a perverse disconnect between what the client is trying to pay for, which is the judgment of the very senior partners, and what their bill is mostly made up of, which is many hours of grunt work from junior to mid-level lawyers who, you know, recently graduated from law school. 240 00:43:53.112 --> 00:44:01.242 Um, that is the economic engine for Big Law. Now, the, the bad news for them is that, uh, that's gonna go away. There's no way that five years from now, right? 241 00:44:01.272 --> 00:44:03.822 Again, even if AI doesn't get any smarter than it is now, right? 242 00:44:03.832 --> 00:44:16.252 If it just stays this good, no smart general counsel of a public company is going to want to pay millions and millions of dollars for recent Harvard Law grads who don't know anything about practicing law to, like, catch typos, right? 243 00:44:16.332 --> 00:44:27.872 Like, AI is just gonna be good at that, right? So the, the Big Law business model is going to break. But the good news is the clients don't give a shit about, like, you know, grunt work by junior associates anyway. 244 00:44:27.912 --> 00:44:29.572 What they're trying to buy is the judgment from partners. 245 00:44:29.852 --> 00:44:41.252 And so the big law firms need to figure out a way to move from this cost-plus model that they, you know, have been on for a long time, and doesn't make a lot of sense, to a value pricing model, which is what the client is trying to buy anyway. 246 00:44:41.692 --> 00:44:57.032 Um, and so step one is, like, figure out, and this is what I'm helping with, the internal systems that allow you to efficiently, like, deliver and execute on the judgment of the senior partners, uh, without having to rely on the sort of pyramid-shaped leverage model with the billable hour in order to make money. 247 00:44:59.572 --> 00:45:07.932 So I could go a lot of ways with this, but I'm actually gonna ask you to, you know, look into the crystal ball here. And [clears throat] 248 00:45:08.332 --> 00:45:17.062 'cause this has big implications for just the market of, I would say, knowledge work. And [clears throat] you know, it's, it, it... 249 00:45:17.092 --> 00:45:32.932 I, I get the thesis, the high-end knowledge work judgment, more valuable than ever and value directly accruing to that. But, um, I have a hard time imagining that, like, the net market value of knowledge work increases. 250 00:45:32.972 --> 00:45:45.932 In fact, it almost seems like it would decrease, like the overall market cap, because a lot of the, what we would call m- uh, knowledge work otherwise, the legwork, the billable hours, um, that, that collapsed to token price. 251 00:45:46.532 --> 00:45:59.392 Um, what do you... You know, I'm, I'm spitballing a scenario here. Where am I wrong? What is your view? W- where does, like, the future of knowledge work go, you know, with legal as the, as the, as the case study here? 252 00:45:59.692 --> 00:46:11.252 Yeah. I think what you are wrong about is Jevons Paradox. Um, I, like, there is so much underconsumption. I mean, there, there are a lot of answers I could give as to why Jevons Paradox makes you wrong about this. 253 00:46:11.312 --> 00:46:19.212 Um, uh, the first is that there is a lot of underconsumption of law right now. Um, you know, uh, like you guys are spoiled, uh, because you have me as a lawyer. 254 00:46:19.492 --> 00:46:29.332 Uh, but, but for most people, hiring a lawyer, uh, it's like a painful process. It's, like, really difficult to know how much it's gonna cost. The billable a- Like, nobody likes the billable hour. It's a baroque process. 255 00:46:29.592 --> 00:46:38.892 A lot of lawyers sort of give, like, hard-to-understand advice. It's, like, so expensive. Um, and so lots of people just don't do it. And so there's a huge amount, right? 256 00:46:38.932 --> 00:46:43.592 Think about, you know, the price of lawyers going down and the pricing making more sense moving away from billable hour. 257 00:46:43.832 --> 00:46:49.752 I think there are a lot of people that are, should be consuming legal services that aren't right now, uh, and they will in the future. 258 00:46:49.872 --> 00:47:00.252 And, like, that's another source of, of demand for law that w- a huge one that'll be unlocked. Um, also just the leverage that these tools give, uh, lawyers. 259 00:47:00.312 --> 00:47:10.322 It's, it's easier now than ever to start a small or solo practice. You just don't need the same infrastructure you did before. Uh, you can... Like, legal research is easier and cheaper than it's ever been. 260 00:47:10.752 --> 00:47:20.332 Uh, writing emails, reviewing contracts is just much easier. And so, uh, like, setting up a law practice will be much easier. Um, uh, think about litigation. Uh, it used to be 261 00:47:21.352 --> 00:47:28.672 that in order to send a scary demand letter, right, something that, like, looked credible, making a threat, you needed to hire a litigator who would vet your case. 262 00:47:28.772 --> 00:47:33.002 You'd have to pay them a bunch of money, and they would need to agree that, like, they could put their name on something. 263 00:47:33.332 --> 00:47:43.452 Now anybody with ChatGPT can spin up a credible-looking demand letter, and a really long one, and some of the claims might be real and some might not, and so you need an actual lawyer, uh, to look at this, right? 264 00:47:43.732 --> 00:47:54.902 There'll be all sorts of new legal questions that will arise because of AI that are novel and not something that LMs, uh, will be super well situated to deal with, right? 265 00:47:54.931 --> 00:48:06.412 To, to choose an area that, like, you know, you all will know about, right? Think about the intersection of agentic commerce and digital assets. Um, right? Think about, you know, like what OpenClaw does today. 266 00:48:06.832 --> 00:48:17.262 Uh, let's imagine three or four generations smarter than OpenClaw is now. What are people gonna do with those agents? They're gonna say, "Go make me money." And the agent will, you know, think, you know, really hard. 267 00:48:17.292 --> 00:48:27.092 How am I gonna make money? Okay, there's this DeFi protocol that's inefficient. Um, you know, like, uh, I, I can find a way to arb it. And so, you know, and so agent will come back to say, "All right, I have a plan. 268 00:48:27.392 --> 00:48:36.952 You know, give me, you know, a million SATs or give me, you know, 10,000 USDC or whatever." And you'll say, "I don't really have a 10,000 USDC. Here's 1,000 USDC." Agent will say, "Fine, let me see what I can do." 269 00:48:37.501 --> 00:48:43.272 Agent will say, "All right, I can't arb this protocol with 1,000 USDC, but this is a really good idea and I could convince others. 270 00:48:43.312 --> 00:48:54.172 So let me go post, uh, on, you know, Reddit or Twitter, and I'm gonna do a token offering. And my token offering is going to give people upside in this arbitrage thing, and so I can, I can raise money." 271 00:48:54.992 --> 00:49:05.992 And so it issues a token, and then it, it takes in a couple million USDC. And then, all right, like, this is gonna be really complicated to, um, arb this protocol, so let me write this sort of trading bot, uh, code. 272 00:49:06.132 --> 00:49:07.052 All right, like I figured that out. 273 00:49:07.092 --> 00:49:16.112 Right now I arb the protocol and, you know, uh, and then it makes all this money, and then it needs to pay out all of the token holders, and then it gives you money and, and, you know, it's, it's fulfilled its reward function. 274 00:49:16.132 --> 00:49:25.152 It's made you money. In the course of the hypothetical I just made up on the spot here, right? Like this, you've not foreseen anything that that agent would do. 275 00:49:25.172 --> 00:49:34.612 You just said, you, you put in a prompt, which is, "Go make me money." And then the agent did an unregistered securities offering, wrote valuable IP in terms of the, the trading bot. 276 00:49:34.652 --> 00:49:45.032 Maybe the arb that it did is a sandwich bot attack, which is securities fraud. It also made you money, which has tax gains and losses. And again, just as a reminder, you just said, "Go make me money." 277 00:49:45.142 --> 00:49:54.872 None of this was arguably foreseeable. So figuring out even what this means under the eight legal regimes I just mentioned is, like, just a plethora of problems. So we're not gonna run out of law stuff to do, right? 278 00:49:54.972 --> 00:50:04.868 It's gonna be all-time highs for law questions. Um, uh, like I, I think there will be roles for humans. The thing that is going away, and right, you know, a lot of people email me and say, "I'm, I'm-"... 279 00:50:04.948 --> 00:50:06.748 think about going to law school, like is now a good time? 280 00:50:07.038 --> 00:50:20.628 And my answer is, uh, if you actually want to be a lawyer, if you have a sense of like why you wanna do this, now might be a terrific time to go to law school because it's never been eas- Like for high agency people, people who just go out and do things, this is a terrific time to be a lawyer, like genuinely. 281 00:50:20.987 --> 00:50:33.428 Um, if you're going to law school because you don't know what else to do and you think it is a guaranteed path to sort of an upper middle-class life, that's a really bad reason, 'cause that is probably going away in the same way that it has sort of gone away for entry-level software engineers. 282 00:50:33.488 --> 00:50:47.848 That used to be just a great ticket to a great salary right out of school, and now not so much. The same thing's gonna happen with law. But that doesn't mean that the, there won't be value for humans. I'll wrap this up 283 00:50:49.088 --> 00:51:02.588 with a curveball, which is, okay, got you to talk about law. What's something interesting you're doing with AI that's like not related to, to law or legal? I mean, um, you spend all day staring into the, 284 00:51:03.588 --> 00:51:10.928 in, into the terminal or the desktop application, depending on your preference for the harness. I don't know. 285 00:51:11.068 --> 00:51:18.648 What's something interesting you've done that's outside of the, you know, the legal beat and the, uh, you know, I'm just curious. 286 00:51:18.788 --> 00:51:27.348 What's, you know, give me a, a side project or a interesting thing that you've seen that's caught your interest. Uh, I'll give you a boring answer and then an interesting answer that, that I probably shouldn't get into. 287 00:51:27.368 --> 00:51:30.908 But, um, the boring answer is I use it as a, as a really good travel guide. 288 00:51:30.948 --> 00:51:39.168 I've been traveling a lot recently for my consulting work, and it's just amazing to have this thing in your back pocket to suggest restaurants, to do a walking tour of a city. 289 00:51:39.488 --> 00:51:44.388 It's just infinitely knowledgeable in exactly the domain you want it to be knowledgeable, right? 290 00:51:44.448 --> 00:51:48.468 The, the things that would be interesting to me from a tour guide are, are different than things that would be interesting to you. 291 00:51:48.828 --> 00:51:53.748 And so having that level of just like instant, on-demand information personalization is really cool. Um, 292 00:51:54.788 --> 00:52:06.428 I've also used it, so I find, you know, call it two to four times a year, I find, uh, doing like a psychedelic reset to be very useful for just like, you know, personally, professionally, my mental model of the world. 293 00:52:06.528 --> 00:52:12.688 I, I find that's very helpful. Um, but, you know, as people who've done this before might know, um, it's not without risk. 294 00:52:13.088 --> 00:52:18.108 Uh, and, uh, it's something that you wanna like take seriously and, and spend a lot of time preparing for. 295 00:52:18.148 --> 00:52:26.448 And so I created a Claude Cowork session, which is like a psychedelic therapist where, you know, I talk about sort of my ideas and my experiences and, and all the preparation I'm doing. 296 00:52:26.778 --> 00:52:38.317 And I'll say to it like, "I want you to do like PhD-level research on what, you know, Freud, and Jung, and the Buddha, and Ludwig Wittgenstein, and all of the sort of like knowledge traditions would say about this type of thing. 297 00:52:39.068 --> 00:52:50.347 Look at all of the things that you think would be analogous to this, and then come back to me with like the two or three things, you know, from philosophy and religion and human history that you think are the best lens with which I should look at this experience I had." 298 00:52:50.988 --> 00:53:02.648 And it's fucking amazing at that. Like, it's really, um, it's not something that you could have asked of a human. Uh, and like I found that to be a sort of surprisingly very powerful use case. Curveball. 299 00:53:02.728 --> 00:53:10.018 I love that greed. Yeah, that's crazy. We both got long hair. I think you can probably- [laughs] Yeah. You're, you're in good- I will say, you know, s- someone who isn't me- You're in good company, Zach... 300 00:53:10.148 --> 00:53:22.278 may have tried to use a keyboard when under the influence of some unknown thing. It, yeah, good luck typing, um, [laughs] under the influence. No, this is a, this is a before and after thing. Oh, okay. Okay. Yeah. 301 00:53:22.588 --> 00:53:32.348 I, I would think, I would think during, i- if you could even manage that, uh, would be a recipe for a bad time. Yeah. Do you take notes? Like, are you lu- are you, are you lucid enough to have notes make sense? 302 00:53:32.398 --> 00:53:32.848 Yeah, I usually, I usually take notes. Okay. 303 00:53:32.908 --> 00:53:43.708 But, but like my view on this is like the purpose of that experience is to experience it, and then like hopefully it's not going to be subtle, the things you're supposed to realize, right? 304 00:53:43.788 --> 00:53:52.768 I'm, I'm a big believer in like, you know, I don't use drugs recreationally, but like, you know, I- I'm b- big dose, blindfold on, like reset. 305 00:53:53.058 --> 00:54:05.348 And, um, yeah, there's usually, there's like a lot of notes and thought about it afterwards, and that, that is where the AI is very helpful. Well, if you're in Bitcoin, you've also got, uh, similar adjacent interests. 306 00:54:05.448 --> 00:54:12.648 Many of Bitcoin's early stories come out of similar tales. Zach, thank you so much for your time and your insights. 307 00:54:12.808 --> 00:54:22.308 Really appreciate you having me on the show and, uh, would love to have you back whenever we need to come back to the story of legal and knowledge work. My pleasure. Cheers. Thanks, Zach. 308 00:54:24.448 --> 00:54:38.148 We need a neologism for that, like psycho clanker punk or something like that. [laughs] Yeah. Um, uh, trip prompting or... Yeah, I'm trying to like- Ooh, I, I like trip prompting. Yeah. Uh, yeah. Um, okay. 309 00:54:38.248 --> 00:54:47.148 We'll, we'll have to go back to the drawing board. We'll have to- The clank hole. Yeah, the clank hole. [laughs] The clank hole. [laughs] That's, that, that could, that, that could be misconstrued. Yeah. 310 00:54:47.208 --> 00:54:59.428 Anyway, um, really fascinating kicker question there from- Yeah... or answer rather from Zach. Yeah. Um, we're gonna keep on going. We're gonna go back to the markets. We're gonna go back to NeoClouds. 311 00:54:59.508 --> 00:55:11.008 If you thought we were done talking about data centers, you're wrong. We're gonna go back to data centers, um, talk about some furk, some wolf, other four-letter words that you're allowed to pronounce on polite show. 312 00:55:11.068 --> 00:55:28.688 But before that, we go to our sponsor, Ligos. [whooshing sound] Oh yeah, baby. Bitcoin is up with the rest of the market, but that doesn't mean that you're not at risk 313 00:55:29.868 --> 00:55:36.548 if you're using a Bitcoin-backed lender that doesn't respect your right to self-custody and is playing funny games with your coins. 314 00:55:37.468 --> 00:55:48.208 The good news is, with our sponsor, Ligos Finance, our preferred Bitcoin-backed lender here at Blockspace, you don't have to worry about that. You don't have to worry about becoming the next Celsius or BlockFi victim. 315 00:55:48.218 --> 00:55:56.237 With Ligos Finance, you can use Bitcoin native smart contracts for their Bitcoin-backed lending to make sure you always hold your own keys. 316 00:55:56.248 --> 00:56:04.248 That's right, Ligos uses Bitcoin native smart contracts so that you self-custody your coins when you're using their Bitcoin-backed lending service. 317 00:56:04.688 --> 00:56:19.218 With Ligos, there's no funny business, no rehypothecation, no bridging, and no wrapping. Just rates as low as 8.5% to 10% APR with pure self-custodied Bitcoin-backed lending. Go to ligos.finance to learn more. 318 00:56:21.296 --> 00:56:32.596 All right, Charlie, we're gonna hop on over to one of our favorites here in terms of our universe, not preferential, but a company that we cover a lot, and that is TerraWolf Fun company to cover. 319 00:56:33.356 --> 00:56:46.716 Fun company to cover, and a big news item for them today. The Federal Energy Regulatory Commission has authorized TerraWolf subsidiary Chesapeake Data to purchase the Maryland's Morgantown generating station. 320 00:56:47.376 --> 00:56:59.536 This hit today, and it is a big milestone for TerraWolf in terms of getting this site online and running. That being said, there are still other hurdles for it to clear. 321 00:56:59.636 --> 00:57:13.396 Before I get into the details of what went down here, this is only, uh, green-lighting the acquisition of this data center and its energy-producing assets, or this site and its energy-producing assets, excuse me. 322 00:57:13.876 --> 00:57:25.406 The order only clears the ownership transfer. It does not permit data center construction, new or repowered generation behind-the-meter power sales, and any changes to interconnection or capacity rights. 323 00:57:25.796 --> 00:57:37.896 So it's just one step of many, but it is a very significant hurdle for them to clear in order to get this site online. Now the-- Now a background on this site for those of you who may or may not remember. 324 00:57:38.896 --> 00:57:42.096 TerraWolf announced this acquisition in February. 325 00:57:42.136 --> 00:57:52.796 The initial phase is eyed at five hundred megawatts on the two hundred and fifty acre site, scaling to as much as one gigawatt of load, and it's paired with additional generation and battery storage. 326 00:57:53.536 --> 00:58:06.576 So the idea here is there are coal plants on this site. TerraWolf plans to convert them to nat gas and gas fire generation, and they will not be resu- restarting the retired coal units. 327 00:58:07.296 --> 00:58:20.796 And they hope to have batteries on site to bank the excess energy that they produce on site and have optionality to sell it back to the grid and support wholesale prices on the grid accordingly. 328 00:58:21.186 --> 00:58:36.516 We have a really good interview from Nazir Khan back in February or March of this year, where he talks about how they are structuring this to make sure that they can be both consuming this power on site and also piping it to the grid when needed. 329 00:58:37.016 --> 00:58:52.536 If y'all are interested, definitely recommend go checking that out. So what happened here, Charlie, is FERC overruled some local regulators in terms of their opposition to this site. 330 00:58:53.066 --> 00:59:02.656 The first one was a PJM capacity fight, and this is really the bulk of the story as to why this was kind of up in the air until FERC gave the go-ahead. 331 00:59:03.336 --> 00:59:15.176 PJM's inter- independent market monitor, Maryland's Office of People's Counsel, opposed the approval, arguing that redirecting Morgantown generation to the data center would tighten supply and raise wholesale prices. 332 00:59:15.756 --> 00:59:24.116 FERC rejected this condi- rejected, um, this condition trying to bar TerraWolf from pulling the units from the capacity market. 333 00:59:25.136 --> 00:59:37.236 And the reasoning for this was that the commission found TerraWolf owes-- owns no other PJM generating assets, so it doesn't have an incentive to withhold the Morgantown capacity to benefit its portfolio. 334 00:59:37.796 --> 00:59:47.716 But if TerraWolf did own another PJM asset or tries to purchase another PJM asset, that could be an issue. The reason here being, if all else being equal, 335 00:59:48.816 --> 01:00:07.976 TerraWolf pulls this capacity from the grid for its own use, but it also owns another generating site, then there are misaligned incentives because if TerraWolf is pulling capacity, all else being equal, wholesale power prices could go up, and then they could benefit from that from their other site by selling energy to the grid. 336 01:00:08.056 --> 01:00:20.296 So it creates these perverse incentives that the FERC regulatory-- that, that FERC says it wouldn't be okay with. But they don't own anything else in the PJM market in terms of power-generating assets. 337 01:00:20.336 --> 01:00:32.756 In fact, this is the only site that TerraWolf is eyeing that has on-site generating capacity. Their flagship site, Lake Mariner, does not, nor does Lake Cayuga, nor does the Hawesville, Kentucky site. 338 01:00:33.356 --> 01:00:43.236 So pr- FERC said that's not a problem. They don't own anything else, so they won't be able to materially benefit from any spike in energy prices 'cause they don't have another site. 339 01:00:43.936 --> 01:00:55.136 The other thing that FERC shot down is this challenge to the Google warrants that Google holds as a result of their-- its Fluid Stack deal in Lake Mariner. 340 01:00:55.556 --> 01:01:13.176 So Google has a stake in TerraWolf in the form of warrants, and Public Citizens, the NAACP, and Port Tobacco River, um, Conservancy tried to dismiss this application because of the warrants, saying, "This isn't just about TerraWolf. 341 01:01:13.236 --> 01:01:21.216 If Google has a beneficial ownership of TerraWolf, then they also need to be considered in this, and they have a huge data center and energy footprint." 342 01:01:21.816 --> 01:01:31.636 So those warrants, they said, created something of a conflict in terms of the approval because you have to bring Google into this if they have beneficial ownership. 343 01:01:32.476 --> 01:01:42.276 FERC accepted TerraWolf's position that the unexercised warrants, which they are un-unexercised, confer no current ownership or voting rights or control, so there's really no problem here. 344 01:01:42.836 --> 01:01:46.516 So those were the two big pushbacks on approval for this site. 345 01:01:46.996 --> 01:02:00.096 Going forward, TerraWolf has approval to finalize the acquisition, so we will be keeping an eye out over the coming weeks and months to see if the acquisition finalizes. Thank you for that, Colin. 346 01:02:00.356 --> 01:02:11.996 What a doozy [laughs] of, like, a detailed, uh, cascade of different names and, uh, jurisdictions. Appreciate your, uh, focus on that. Yeah, there's a lot there. 347 01:02:12.156 --> 01:02:26.796 And Paul Prager, if I can find it on the fly here, had a really good response to a Twitter reply this week where, I don't know if it was a TerraWolf shareholder or if it was just a talking head, but, uh... 348 01:02:26.856 --> 01:02:42.346 Yeah, here we go. I'm gonna just throw this up here Um, let me get, here we go. Perry Lynn, uh, was commenting on a reply to one of his tweets, says, "Do you think Wolf has found a bottom now? If no, what is?" 349 01:02:42.446 --> 01:02:54.386 I hope so, but probably not. There's potentially a rally. There's potential- there's a potential rally to a couple of dollars, but I think we can retest lows and maybe head to 16. Who is John Galt? Classic. Mm. 350 01:02:54.706 --> 01:03:03.186 "Strategic mistake for TerraWolf to think they could do biz in liberal states. Execution risk is real. They need to pivot to free states and not overly regulated liberal states." 351 01:03:03.986 --> 01:03:12.746 And Paul Prager responded, "Actually, our sites, if anything, are more valuable than ever, given they are fully contracted, already approved for operations with committed energy and existing energy infrastructure. 352 01:03:12.806 --> 01:03:15.666 We are fortunate to have been first movers and early." 353 01:03:15.706 --> 01:03:25.366 He's mostly talking about the Lake Mariner site here, for sure, because there was the data center moratorium in New York, and people were freaking out about what this means for TerraWolf. 354 01:03:25.966 --> 01:03:33.486 They said that the Lake Cayuga site's energization timeline is so long as such that they don't think that the moratoriums will be an issue. 355 01:03:33.546 --> 01:03:39.966 They think they'll be lifted by the time that they seek energization and permitting at that site, or rather energization, excuse me. 356 01:03:40.666 --> 01:03:49.426 But for Lake Mariner, that is not affected because they already have full regulatory clearance to run that site and to-- and it's already operational. 357 01:03:49.986 --> 01:04:03.126 But I do think the reason I wanted to bring this up is because when you look at Pennsylvania too, not a, you know, no ma- data center moratorium like New York, not as stringent, but there was significant pushback 358 01:04:04.446 --> 01:04:11.266 on this site specifically. And excuse me, I, I need to revise that. Um, not Pennsylvania, Maryland, excuse me. 359 01:04:11.306 --> 01:04:21.146 But so still a blue state and for this site specifically, you do have to ask yourselves like how-- are, are these headaches worth it? 360 01:04:21.926 --> 01:04:26.806 I mean, if they can get the deal done, obviously they are, but there's a long uphill battle with these things. 361 01:04:26.846 --> 01:04:40.195 The fact that they had to basically quell two dissenting viewpoints from the localities and local regulators, uh, kind of shows that there are certain states that just fundamentally are going to fight you more on these builds. 362 01:04:40.286 --> 01:04:55.426 But TerraWolf securing a win here with FERC dismissing those concerns. [sighs] Yep, they move fast, so speed to power also means speed to permits. So, uh, uh, yeah, love to see it. 363 01:04:56.206 --> 01:05:05.936 This wraps up our live- Oh, no, Charlie. Got one more story. Oh, oh, oh. Whoa, whoa. We got- Got it. Doesn't wrap it up We got one- I'll tag you back in. Pull me out... 364 01:05:06.046 --> 01:05:12.866 one more to just briefly note on, um, this interesting- Oh, yeah, this is going down-- We're leaving New York, we're going down to ERCOT. 365 01:05:13.266 --> 01:05:21.326 Yeah, just, uh, going down to the Lone Star State where they want you to build and rip energy-generating assets all day long. 366 01:05:21.346 --> 01:05:34.706 This is a note from Morgan Stanley where they see Cipher Galaxy upside from ERCOT's sixty-five gigawatt batch zero. Uh, this-- The headline muddies this a little bit. 367 01:05:34.746 --> 01:05:49.206 What's specifically happening here is that Morgan Stanley estimates ERCOT classify about sixty-five gigawatts of large load requests as base loads, which potentially clears advance- advanced, uh, data center projects to move forward toward grid connection. 368 01:05:49.906 --> 01:06:05.266 The TLDR here is that if the base load, if these loads are considered base load, that covers operating projects, uh, sufficiently, um, base load covers operating projects and sufficiently advanced developments supported by qualifying prior studies. 369 01:06:06.026 --> 01:06:21.376 Ma- meaning that they could circumvent some of the more onerous, [clicks tongue] uh, some of the more onerous paperwork for the batch zero, and they might be able to duck having to be included in that process entirely. 370 01:06:22.176 --> 01:06:26.376 Morgan Stanley says, [clicks tongue] um, it-- the, um, 371 01:06:27.546 --> 01:06:37.426 I'm quoting here directly from the Blockspace article, "Morgan Stanley's estimate is thirty gigawatts above the bank's previous expectations and could benefit site operators with mature interconnection positions." 372 01:06:37.466 --> 01:06:50.946 They specifically identified Cipher Digital and Galaxy Digital as having preferred exposure to the process. Morgan Stanley expects about two gigawatts of Cipher-submitted requests to receive base load treatment. 373 01:06:51.006 --> 01:07:00.846 And again, if they receive that base load treatment, they aren't going to be as scrutinized or have to go through as much rigmarole as ones that aren't considered base load in the batch zero process. 374 01:07:01.726 --> 01:07:18.226 So potential tailwinds there for Cipher and for Galaxy. Okay. So go build in Texas. Maybe don't build in New York. We'll see. On that note, we're wrapping up. 375 01:07:18.586 --> 01:07:30.046 This is Blockspace Live, going live every single weekday at one p.m. Eastern. We are Compute's daily show featuring quick hits on AI, data centers, emerging technology, and markets. 376 01:07:31.046 --> 01:07:44.426 If you like the show, you'll love the rest of our content. Find that at our website, blockspace.media. That's blockspace.media. This show is brought to you by CleanSpark, Nasdaq listed ticker CLSK. 377 01:07:45.706 --> 01:08:00.166 Thank you for listening. I'm Charlie. I'm Colin. We'll see you tomorrow. [outro music]