US Data Center Share vs. the World, How Stock Values Track GPU Rental Prices, the BIS is Bearish AI

Summary

---

Blockspace: AI & Bitcoin

Episode: US Data Center Share vs. the World, How Stock Values Track GPU Rental Prices, the BIS is Bearish AI

Date: June 30, 2026
Hosts: Charlie Spears & Colin Harper
Guests: Rob Hamilton (Anchor Watch), Kush Bavaria (Oren)

---

Episode Overview

In this episode, Charlie and Colin dive into three research-driven topics at the intersection of AI, bitcoin, and global infrastructure:

1. An in-depth breakdown of the International Data Center Authority’s massive report on the AI data center Capex boom and global energy headroom.
2. A conversation with Rob Hamilton (Anchor Watch) on the current state of bitcoin markets, developer funding, and the growing impact of AI on developer productivity and startup workflows.
3. GPU rental prices, stock behavior, and the hardware market with Kush Bavaria (Oren), followed by a review of the Bank of International Settlements (BIS) latest report, which frames the AI boom as a classic bubble with systemic financial risks.

The tone is data-driven but irreverent, with plenty of inside jokes and a healthy skepticism toward both hype and fear-mongering in the AI and bitcoin sectors.

---

1. The Global AI Capex & Data Center Boom

[02:24–11:57]

Key Points & Insights

• USA’s Dominance in Data Centers:

  • The US leads with 29.2 GW of data center capacity—43% of the world total (67.7 GW).
  • US centers now consume 6% of national electricity, up from 1.7% globally in 2024.
  • China follows with 8.5 GW but this is just 0.8% of its national consumption, signaling much more headroom.
  • Germany: 5.5 GW = 9.5% (!) of national usage; a byproduct of decommissioned nuclear power.

• Growth & Headroom:

  • The world’s data center power usage has grown 36% in two years.
  • China has ~59 GW of electricity headroom—almost double current US data center consumption.
  • US only has 1.2 GW of headroom, signaling an imminent energy bottleneck unless investment ramps up.
  • Smaller countries like Singapore are already at 20% of national grid usage for data centers.

• Community Pushback Threshold:

  • Community pushback against data centers emerges when usage crosses 5% of national electricity.
  • Noted in Germany and the US, but little in China/Japan.

• Environmental Data:

  • Data centers in the US use 1 gallon of water for every 10,000 gallons for human use.
  • No direct link has been found (yet) between expanded data center power use and consumer price increases—but most new centers aren’t operational yet.

Notable Quote

"China, with almost 60 gigawatts of headroom, should be a pin dropping in a quiet room moment...we need to start building out more capacity if we want to compete on a long enough time frame." – Charlie [07:50]

Timestamps

• Discussion of report and main stats: [02:24–10:59]
• Water use and price discussion: [10:59–11:57]

---

2. Market Update & Bitcoin Developer Funding

Guest: Rob Hamilton
[13:16–30:48]

Key Points & Insights

• Bitcoin Market Sentiment:

  • Ongoing drop to ~$58K, with Michael Saylor’s entities (Strategy, Stretch) making moves.
  • Saylor’s preferreds and common shares are behaving differently; preferreds (STRETCH, SATA) show higher yields and volatility.
  • Huge short interest rates on SATA (68% annually) signal unsustainably expensive bearish bets.

• Bitcoin Developer Funding:

  • Three largest funders: Chaincode Labs, Brink, Spiral.
  • Grants often paid in bitcoin; security and stability vary (e.g., Spiral is under Block—more stable).
  • Funding has not dried up despite the price downturn; matter of ecosystem stewardship.

• AI’s Impact on Developer Productivity:

  • AI tools (e.g., Cursor, Claude code) have transformed prototyping and iteration.
  • Rob: mobile app prototyping that previously required dedicated headcount now happens overnight via AI.
  • Discussion on security with AI-assisted dev, especially in high-stakes crypto.

• Open Source, Tokens & The Future:

  • Joke about AI + tokens: “Tokens are the future, just not the way we thought.”

• AI Regulatory Issues:

  • New US restrictions may hobble domestic AI innovation while boosting open-source Chinese models.
  • Attempts to “control math” (AI model exports) echo failed cryptography export bans in the 1990s.

Notable Quotes

"The bitcoin miners can absorb that energy, and in many ways this feels like a second bite at the apple to build a new Internet." – Rob Hamilton [13:05]

"With Bitcoin, the code is the money and the money is the code." – Rob Hamilton [24:18]

"The US government's job is not to protect OpenAI and Anthropic’s margins...the capital and growth explosion could happen from AI regardless of the model’s country of origin." – Rob Hamilton [29:33]

Timestamps

• Saylor/Strategy/Stretch yield discussion: [13:27–18:46]
• Bitcoin developer funding and ecosystem: [19:44–23:05]
• Rob on AI-accelerated development: [23:12–27:03]
• AI regulation and global dynamics: [28:11–30:48]

---

3. GPU Rental Markets, Stock Valuations, and Open Source AI

Guest: Kush Bavaria (Oren/Orn)
[32:00–45:25]

Key Points & Insights

• GPU Rental Prices and Stocks:

  • Nvidia’s stock now highly correlated with H100 GPU rental prices—up or down, they track one another.
  • Odd dynamic: B200s, H200s have falling utilization, but rental rates keep rising. Longer-term contracts drive up “on demand” rates.

• Depreciation of Chips:

  • Widely held belief in rapid chip depreciation is called out as myth—real-world data shows old chips (even 6–7 years) still command strong usage and prices.
  • Chips can even appreciate as demand spikes and new workloads (esp. inference) emerge.

"You can even think about chips as an appreciating asset. At least for the last six years the values have gone up." – Kush Bavaria [35:17]

• Open Source & Chinese Models:

  • Recent US export/model bans create a vacuum—Chinese open-source models (GLM 5.3 rumored) are rapidly catching up.
  • Global compute markets are seeing increasing preference for open-source as US restrictions bite.

• Stock vs. Rental Price Discrepancies:

  • CoreWeave’s GPU rental rates rose 59% in six months, but stock only gained 9%—market seems to be pricing debt loads and balance sheets, not just rental rates.
  • More factors (debt, management, etc.) drive stock prices beyond GPU hourly rates.

• Supply & Demand, The GPU Crunch:

  • The “compute crunch” is not near resolution—most of the world still isn’t using AI models regularly.
  • Expect demand to only grow as AI is further commoditized across the globe.

• ASICs & Custom AI Chips:

  • OpenAI’s “jalapeno” ASIC chip is more about replacing Cerebras than threatening Nvidia.
  • CUDA ecosystem and data center familiarity keep Nvidia dominant; ASIC talk is “media hype”.

"If you ask a startup, 'Hey, do you want a Cerebras chip or an Edge chip?' No one's going to say, 'Yeah, I want it today.' Everyone’s: 'Give me Nvidia or don’t give me a chip at all.'" – Kush Bavaria [44:13]

Timestamps

• Nvidia/GPU price correlation: [32:55–33:55]
• Chip depreciation reality-check: [35:20–36:08]
• Open-source model trends: [37:00–38:31]
• Stock vs. rental price dynamics: [38:31–41:08]
• Supply-demand analysis, future outlook: [41:08–42:57]
• ASIC and chips commentary: [43:29–44:40]

---

4. The BIS "Bearish AI" Report: Bubble, Boom/Bust, and Systemic Risk

[45:32–63:28]

Key Points & Insights

• Capex & Debt:

  • Hyperscaler Capex ($160B in 2026) now at 0.4 of revenue, a jump from 0.1 in 2024.
  • Relentless borrowing boom to build out infrastructure.

• AI Boom Dynamics:

  • Compared to 19th-century “Canal Mania”, railroads, the Roaring 20s, and dot-com. AI ramp-up is historically unprecedented in speed and scale.
  • These historic booms ended with busts—report positions AI as at risk for classic bubble collapse.

"For these kind of being normalized of multiples of boom bust...for the 1920s which preceded the Great Depression—to be the smallest boom-bust cycle on this chart is pretty dramatic." – Colin [49:34]

• Valuation & Retail Exposure:

  • Mega-cap AI stocks sport higher multiples than dot-com era; broader retail (esp. lowest and highest American wealth quintiles) exposure signals risk to household net worth if boom goes bust.
  • "This is barbell theory: the lowest percentiles and the highest percentiles have seen their net worth invested in the stock market go up since the dot com bubble." – Charlie [57:19]

• Credit & Contagion:

  • Signs of rising credit risk, particularly as multiple Business Development Companies (BDCs) make loans to the same AI firms (“circular financing,” increased contagion risk).
  • If one large AI firm fails, many lenders are exposed—could spark a cascade.

• Macro Backdrop:

  • Geopolitical risk (Iran, energy shocks, inflation) overlays the bubble risk—pressure on energy costs will ripple into AI and bitcoin mining, given their heavy electricity reliance.

"Regardless of your view on credit risk or the AI bubble...we do have this almost unavoidable wall of inflation slowly marching towards Bethlehem." – Colin [65:19]

Timestamps

• Capex and bubble context: [45:32–52:15]
• Valuation and retail risk: [52:15–58:23]
• Credit market contagion: [58:23–63:28]
• Macroeconomic/geopolitical backdrop: [63:28–66:51]

---

Memorable Moments & Quotes

• On US and China’s future in AI:

  “If they're [the US] being serious about AI as a national security risk...China, with almost 60 gigawatts of headroom, should be...a pin dropping in a quiet room moment.” – Charlie [07:50]

• AI-accelerated dev work:

  “What I really love using it for is iterating new prototypes and features...now I just queue it overnight...and I have a new app in test flight by morning.” – Rob Hamilton [24:22]

• Stock market wealth barbell:

  “The 6040 portfolio is dead.” – Colin [58:20]
  “50% stocks, 50% bitcoin.” – Charlie [58:23]

• On ASIC AI chips:

  "If you ask a startup, 'Hey, do you want a Cerebras chip or an Edge chip?' No one's going to say, 'Yeah, I want it today.' Everyone’s: 'Give me Nvidia or don’t give me a chip at all.'" – Kush [44:13]

---

Final Thoughts

• The US remains a massive leader in AI infrastructure but faces energy headroom constraints and rising societal resistance.
• Bitcoin markets are undergoing volatility with developer funding stable—AI is transforming productivity even more than market cycles.
• GPU markets are hot—a “compute crunch” is likely to last years; older chips retain value, and open-source AI models are accelerating, in part due to US regulatory missteps.
• The Bank of International Settlements is flashing systemic risk signals: unprecedented capex, rising debt, and the potential for highly concentrated, contagious defaults.
• All of this unfolds against a background of stubborn global inflation and geopolitical instability.

For more deep dives, subscribe to the Blockspace newsletter at newsletter.blockspacemedia.com.

---

Transcript

[Charlie]

Foreign. What's going on, y'?

[Rob Hamilton]

All?

[Charlie]

Welcome back to Blockspace Live, brought to you by CleanSpark Charlie it's the report edition. Today we've got two very juicy reports. One from the International Data Center Authority that we will lead with that takes a look at the current AI capex boom across multiple jurisdictions, not just the United States. And we will bookend the live stream with a look at everyone. Chill out the bank of International Settlements in their annual report, take a look at the AI boom and run some numbers with regards to valuations and debt to give a temperature check on where we are in the current AI bubble. The globalists are out there trying to make everyone scared of getting in on the last leg of returns. Joking, joking. Anyway, those are our two segments today. For news, for interviews. We have Rob Hamilton of Anchor Watch on to talk a little bit of everything Bitcoin as bitcoin hovers around 58k and as the stretch crisis is not resolved. And then we have Kush Bavaria from Ornon to talk about GPU rental prices and their findings recently that stocks are starting to track these indices as investors wise up on ways to actually valuate these companies in real time.

[Colin]

Yeah, I hope you like reports. We are diving deep into the research. Not just two long form research papers, but additional research and insights from Oren today with Kush. Blockspace goes live weekdays at 1pm Eastern featuring quick hits on AI data centers, Bitcoin mining, emerging tech and sometimes bitcoin. If you like what you hear on the live stream, it turns into a podcast shortly thereafter. Wherever podcasts are found and also a newsletter in your inbox every single day. Go to newsletter.blockspacemedia.com to sign up for that. So Colin, Bitcoin's crashing. We'll talk about that later. We'll probably riff on that with Rob. But I think what we're going to do today because it's a little quiet on the news front, that means we get to roll up our sleeves and crack open a textbook or in this case long form PDFs, get dirty with some data.

[Charlie]

Yeah, what we're doing today. So our first report here comes from the International Data Center Authority and they published a pretty meaty piece that looks holistically at the data center industry. Everything from design to funding to concentration in certain markets. They even address public outcry against resource use which we'll get into here shortly.

[Colin]

Shortly.

[Charlie]

To start off, I think this gives you a good idea of where the world is in terms of the AI Capex boom. I found this to be interesting. They look at major markets and show how many gigawatts are currently active for all data centers in these countries and how much of that equates to the percentage of electricity consumption within these countries. No surprise. The US tops the list.

[Colin]

I mean, you say it's no surprise. Like, I knew it was big, it was the biggest, but I didn't realize that. USA topped with 29.2 gigawatts versus the second place, China, with 8.5 gigawatts. That's what's most surprising to me.

[Charlie]

Yeah, it's four. So the US is 43% of the global data center market. As Charlie just said, the US commands 29.2 gigawatts out of a 67.7 gigawatt estimate for the entire world. The entire world's estimate is a growth rate of 36% over the past two years. And now they estimate that data centers consume 2% of the world's electricity, up from 1.7% in 2024. China, as you said, Charlie, is 8.5 gigawatts. That's only 0.8% of their electricity. This is a really important point for an index that I will pull up later in terms of future growth potential for data centers. The United States at 29.2 gigawatts. That represents 6% of all of the nation's electricity. Now, that seems high, but if we look at Germany, it could be a lot higher. Germany is 5.5 gigawatts, and that's 9.5% of the nation's electricity, which is pretty astounding.

[Colin]

Yeah. Germany, in a degrowth country trying to be powered by much more variable, unreliable renewables. Not surprising that data centers are taking up the highest percentage, which is very ironic because, you know, I would say, like the eu, not really known for their data center, booming economy. And yet here we are, the flagship country of the EU being their grid dominated by the most by data centers.

[Charlie]

I mean, does any. Is there any better stand in, though, for the DE growth initiatives in the eu? Because the country with the largest footprint is the one that's been decommissioning nuclear.

[Colin]

Yeah.

[Charlie]

So, you know, Germany, I'm pretty sure did away with like all of their nuclear. They might still have some, but that was under Angela Merkel's reign and it was. It continued after she left. But at 9.5%, that is a crazy load for a smaller footprint compared to China, the second biggest. And the US then UK is at 2 gigawatts, which is 5.8%. And Japan is 1.7 gigawatts at 1.5%. The interesting part about this for me, Charlie, is if you actually look at the, if you look at the usage of data centers for overall output for these countries, it puts into perspective how friendly or unfriendly they are in terms of trying to bring new capacity online. This chart Here on page 15 sums this up nicely. You can see estimated dentistry electricity usage by country versus national percent of electricity grids. The US is there's just incredible amounts of terawatt hours, 220 in 2026. And it's just a small slice of the pie. China, it's even smaller. If you look at the 80 terawatts per hour in 2026 estimated again that 0.8% total utilization of their grid. Then you get to some smaller companies out here like Singapore, where they're using 20% of the national grid and they're barely using 15 terawatt hours so far this year. But the scary thing here, Charlie, for me to look at is if you look at the data center headroom chart, this is a chart that looks at potential.

[Colin]

Let's zoom in a bit here. Really what's up? Yeah, zoom in a bit here on this.

[Charlie]

So what this shows is a country's ability to add new capacity online without having to build a gigawatt to service that capacity. China's headroom is at 58.9 gigawatts. So they theoretically have enough freed up electricity to deploy more than double what the US is already consuming for data centers, followed by India at 12.7 and Russia at 6.7. Japan coming out with more headroom than I would have assumed given their current footprint and the electricity use that is going towards it as a percentage of national electricity use coming in at 5.5 gigawatts and the US is down here at 1.2 gigawatts. Now this really, I think, speaks to the bottleneck narrative with energy specifically. And if you look at what China has been doing on the national front in terms of building electricity assets, they're deploying everything. They're building a lot of coal, they're building nuclear, they're building NAT gas. Yes. They're also building solar farms and wind farms. But if you look at those as a percentage of actual generating assets, all of the carbon based power plants and the nuclear power plants outstrip them incredibly. But I think that this also sums up why AI is a national security risk and a priority for the government. If they're being serious about it. Because China, with almost 60 gigawatts of headroom, should be a kind of pin dropping in a quiet room moment in the sense that we need to start building out more capacity in line with that if we want to be able to compete on a long enough time frame, assuming the demand for AI keeps up. But those were just a few things I wanted to highlight in the report. It's very Meaty, it's about 85 pages. So if y' all are interested in more stuff in it, I, I recommend you dive in.

[Colin]

Yeah, there's a few more insights on that and we'll probably because it's taking longer than I anticipated, we'll probably skip a few of them. I'll point out that on page 44 it does a little treatment on the highest concentration of data centers in the world in Loudoun County, Virginia, the world's densest data center hub. I think maybe like 20 to 30% of Virginia's or that that county's revenue comes from taxes, property taxes of data centers. And in that area, very pro data center area, everyone is now being weighed, now being delayed up until 2032 for new powered land deals. Which validates the thesis of the as this report explicitly mentions, the crypto to AI Bitcoin miner powered land deal. Another interesting insight on this is that what this report also found with regards to community pushback at a certain threshold of data center penetration in the overall energy grid mix, it finds that around once data centers Start to consume about 5% of the nation's power, then community pushback begins. And it identifies specifically Germany and the U.S. notably, you see in Japan, China and Japan and China, there's as far as we know, little to no data center pushback. Again, China's a difficult beast. But you know, if we want to consider what it looks like for local, regional and community pushback against these data centers, whether it be power or criticisms of water, you can kind of project that forwards and say what does it look like for these growth economies for Japan, going from 1.5% of overall energy mix to 5% energy mix, what does that threshold hit? And then you can perhaps project or judge accordingly when the people will start showing up with pitchforks at your local city council meeting.

[Charlie]

And on that note of pitchforks, one other thing that's interesting. Their findings said factually speaking, data centers throughout the US today consume one gallon of water for every 10,000 gallons consumed for overall human use. They also found no direct connection between new data center electricity consumption and immediate consumer price increases. There's been A decent amount of research on that.

[Colin]

Actually there was the Nick Carter piece on a few weeks ago. Great piece.

[Kush Bavaria]

Yeah.

[Charlie]

And I think that was a synthesis of outstanding research, kind of like a meta analysis. And that is something to watch because it stands to reason that right now when we haven't seen the full weight of the capex cycle play out, because we still only have if we look back to 2024 when this started kicking off. A lot of the data centers that are being bankrolled right now are not even close to being at full capacity. Many of them haven't been built. So I'd be curious if that stands in a few years.

[Colin]

All right, we have Rob Hamilton in the wings. We're going to bring him on up here to talk about. Yes, you got it. Bitcoin, but also some AI before. Rob, a word from our sponsor, CleanSpark.

[Charlie]

We are CleanSpark, America's Bitcoin miner. A publicly traded company with the largest operating hash rate powered entirely by self operated infrastructure across four states. This is our proof of work and we are setting the standard for what's next. Learn more about the intersection of energy and [e-mail] if Bitcoin's actually the best money and it's the thing that people should accumulate and it's the best risk

[Kush Bavaria]

adjusted asset, I'd lose zero sleep about

[Charlie]

whether or not that's gonna happen. I just ask the question of when

[Kush Bavaria]

is literally matrix math that you're running

[Colin]

on large pieces of data.

[Kush Bavaria]

The bitcoin miners can absorb that energy

[Colin]

and in many ways this feels like

[Rob Hamilton]

a second bite at the apple to build a new Internet.

[Colin]

All right, we got Rob Hamilton. We're going to bring him on up here and rip about some bitcoin. So welco, welcome back to the show, Rob.

[Rob Hamilton]

Hey guys, thanks for having me.

[Colin]

Good to see you. So bitcoin's crashing. Everybody's talking about Sailor. People are mad at Sailor. What's your temperature check? What's your read on the vibes, Rob?

[Rob Hamilton]

You know, I think it's not much has changed since I was on the two weeks ago. I know Sailor yesterday came out with that press release kind of shoring up the bat, the balance sheet and the capital structure signaling that they're being more thoughtful as it come as it relates to things like the amount of cash they hold on their balance sheet and kind of defending those preferred dividend interests and giving themselves a permission structure to sell bitcoin if needed to cover everything. I mean, Sailor's been the large marginal Buyer out in the market for, you know, at least the past, you know, four or five months, especially once STRC really started taking off. My assessment right now is he came out that big release yesterday, there was a huge rip like 10, but it was still 15 under par. If you look at the live price right now, STRC is trading at $83 off from the hundred and SATA is trading at 91. And between those two, like that's a 14 and a half percent yield right now. And there's less than a point difference between what SATA is paying, an effective yield, and what STRC is paying. And it kind of just tells you this is currently what the market's pricing the risk at. It just is what it is. And if Sailor is able to Increase the rate, 50 basis points a month we're talking about, and you know, probably a four month slog, four or five month slog to get to that rate naturally through the STRC dividend mechanic. But until then, the market's saying that this is the fair price of lending Saylor out that money in perpetuity.

[Charlie]

And what's interesting to me about this, if we look at the preferred versus the common stock, Stretch has held up from that rally much better than strategy actually has. I'm gonna get the chart back up here really quickly. Stretch is at 83. It's down 3% today. But as you said, Rob pumped a nice like 10%, 11% following the news. Strategy is kind of reverted. I mean, it's still above where it was before a sale.

[Colin]

One strategy equals stretch, you know, that's the.

[Charlie]

I mean basically at this point, but it's kind of fallen and it's a little bit above where it closed last week, just a couple bucks above. But to me, it almost seems like this signals that investors see that capital plan is more bullish, at least for Stretch, than strategy, as bullish as it can be. And I guess partly because what Saylor basically announced with that, with opening Strategy up for 1.25 billion worth of bitcoin sales and saying that they would buy back, I believe Stretch as well to try to retain the peg. It seems to that's naturally bearish for the common stock. Just because you're selling bitcoins, the whole raison Detroit. Right. Yeah.

[Rob Hamilton]

And the entire mechanic of Saylor's recent ability to aggregate capital has been just going to the ATM and issuing more common. Right. And it's not something he's able to do as easily with the preferreds like one, the preferred's sitting higher up in the capital stack. But again, like the whole notion of this was that in the prospectus and how it was kind of marketed and delivered in all the disclosures was they would do those at the market offerings only when you got to the hundred dollars. So in theory they could issue more shares right now, but then they're kind of locking in that percentage of the 15 points, 14 and a half percent. It would kind of shake things up. So I think it's an interesting waiting game now because Saylor has so much cash sitting there hanging out, he could just go take off for a year, right? You go take off for a year and then also pay the converts and like have cash to do that, right? Like he, he's sitting on such a, a dollar war chest at the moment that he's able to kind of wait the market out, which actually was an interesting piece for the other perpetual preferred that gets a lot of attention. SATA, Jeff Walton from Stride came out with this really interesting stat that I saw earlier today that the current short interest rate is like a little over a day. But what's more interesting is that the, the rate to borrow SEDA if you want to go short, right? So mechanically, if I want to go short a stock, someone has to lend me the stock, I sell the stock and then I owe them back one share to return it. And the actual like what you have there is like the actual short interest. It's almost like two days of the volume of what happens for SEDA is currently locked up in short positions and they're paying 68% at the moment. So if you had a share of SATA, you could lend it to someone and have a 68% annual rate. Plus SATA doing that daily dividend. They also owe that 13% annualized dividend. So like that's going to have to. I don't see that being a long term sustainable short position. Like that's incredibly high. That has to probably get resolved sooner or later. And SATA is trading at 91. So if you had like a huge, you could easily see a huge pop up from there just from people having to necessarily cover their shorts. Because 68 interest per year is a really expensive position to hold. You're, you're borrowing a 90 stock and you're going to pay what, $65 a year just to hold the ability to go short. It, it's not sustainable.

[Colin]

Okay. So I do find it really ironic that the biggest bitcoin bull in the world is sitting on the most this incredible treasury of dollars. But we'll we'll kind of put that over here from now. Let's talk about 58k Bitcoin first of all. Hell yeah. Meme coin meme number 58K. We're here. It's purgatory. But at my heart, you know, I want to go back, I want to talk about something where I'm much more personally interested in which is the landscape of like who builds bitcoin, who works on this, the bitcoin developer scene. A lot of people don't know this exists. This is a beat we've covered for years at Block Space. You, Rob, are pretty dialed into this. A lot of people don't know that there's bitcoin developers out there and they get funded often in bitcoin. It's cyclical with the bitcoin market. Rob, what's your sense of what's going on in the world of bitcoin developers? Are they able to pay the bills? What's going on here?

[Rob Hamilton]

Yeah, so there's multiple players in the ecosystem. Right. Maybe your two largest ones when you're thinking about it. The three largest, I would say when you're thinking about the bitcoin, the core protocol itself, you have chaincode Labs, Brink and Spiral. So Spiral is part of the Block square umbrella as an organization. So from heuristics of, they are employees of Block and because of that they maybe have a little bit more stability than someone who is working on open source grants on a month to month or year to year basis. And then chaincode and Brink are, have a lot of, you know, grassroots support as well as like industry donations of people who give long term visions. They understand that bitcoin as a intergenerational project, you need to have stewardship of people that are just working on the boring, unglamorous things. They're often not working on fancy new upgrades or any sort of flashing headline things. It's just the, the toiling away of that regular work. And then you also have organizations like OpenSats and you have like the Human Rights Foundation. The Human Rights foundation tends to focus on things for like bitcoin accessibility around the global world and use cases of being able to help people, you know, use Bitcoin to fight against oppressive governments. And then you also have open sats which also does do all of their grants paid out in bitcoin. And typically people get a grant, either they'll do a small one off for a project or they also will have things called like long term support grants where they'll actually get a two year grant. And it's paid monthly out in bitcoin, no matter what. Right. And so I think what's actually most interesting about that is less the bitcoin price. It's actually more of an AI conversation. Because when you think about this, if the marginal utility and productivity, if the marginal productivity of these developers can be increased, multiple folds, one, you would assume either you're going to have like, you're gonna have to do more work for your dollar, right. Like the work, maybe a year's worth of work in 2023 is, you know, you have to be doing like, you know, at least 2, 3x of what you're able to do now. Right. And so you could easily have the same amount of dollars go a lot further or maybe you have a contraction in the actual funding. I think I haven't seen any sort of changing and you know, grants are still being issued from all of these organizations. There hasn't been any sort of like hiring freeze or slowing things down. But it's what I kind of think of the role of an open source developer. Over time with these tools, I think it's really compelling for people to be able to quickly build, iterate and do these own projects themselves. We're using a bit of AI in the prototyping stage of things at Anchor Watch and ideas of like building a mobile app, which would have been like a whole dedicated headcount, would have been a lot of overhead. I'm able to kind of cobble together at least working prototypes that I've been playing around with right now. And I just have it running overnight. So I'll just overnight, I'll go to bed, I'll queue up a bunch of instructions of what I want to get done and then first thing in the morning comes back and I have like a new app in test flight I can start testing, which is one step outside of the like full open source developer ecosystem. But I think it's all part of the story though for like where devs are being compensated for their time, whether it's through grants or their jobs. And just trying to understand, I think it's way more of that tech productivity story than the dollars of bitcoin flying in and out of the ecosystem.

[Colin]

Imagine we started compensating devs in tokens. But yes, it's AI, it's inference tokens.

[Rob Hamilton]

All of the crypto, the crypto people were right. Tokens are the future. It just wasn't in the form that anyone expected it to be.

[Colin]

So you brought up using AI in development, you've Got us start up. I don't even know if we call. It's a real insurance company now, so. And you've got a number of developers, small cracked team.

[Rob Hamilton]

Yes.

[Colin]

Tell me about AI. Using AI in a. In a developer environment. How do you handle security? How does this change your workflow is. Is what you're doing now. Was it like with AI? Did you anticipate this two years ago when you're thinking about corporate roadmap?

[Rob Hamilton]

No, I definitely was not. I was tinkering with those tools. Back in those days, what you do, you would take ChatGPT and you would copy and then you would paste it and then you would copy the error message and paste it back in. Right. This is before the Claude code harness cursor was kind of the first one to start. I first started using that really like this harness idea of being able to reflexively iterate through the whole stack of everything. I want to say though, I'm fortunate because there's a lot of things that vibe coding is fun for. But with Bitcoin the code is the money and the money is the code. So it's not something you want to just trivially like make me a bitcoin wallet. Make no mistakes. Right. So what's really been beneficial for us is that the core logic and infrastructure of what we built at Anchor Watch was largely set in stone before the AI wave came. And we were able to do is we were able to go in there and do iterative testing. We were able to kind of like battle test things, add more documentation, kind of build on top of that, which has been great just from having a strong foundational code base. What I really love using it for is for iterating new prototypes and features. Right. Like I mentioned before, like an Anchor Watch mobile app, I can actually take some of the core logic of how we actually manage Bitcoin on our side. I could port it into a different programming language using Bitcoin Dev Kit. So we're built on top of Bitcoin Dev Kit. I should have even called this out during the funding of open source developers. We're a member of the Bitcoin Dev Kit Foundation. We give money to Bitcoin Dev Kit every year because it's a very strong, robust library. And that originally I think was part of the spiral arm of projects between that part of the open source ecosystem. It's really fun and interesting and we also have places where strictly for prototyping, getting things started, having non engineers build initial prototypes of this is what I want this website to look like. And then rather than having to start with like a blank page from the engineering side to understand what you want, or like maybe little cocktail napkins with like pictures of frames, someone who's not an engineer can just hand us like basic wireframes and like a very basic click through the website saying this is what I want to do. And then we can iteratively go back and forth on those feedbacks. We've done that for a couple of products where we'll hop on a zoom call, record it, go through all of the things that we want to change, what we like, what we want to get rid of, and then we'll just transcribe that into cloud code, send that as a product requirement document. And since it's not live yet, we have this great flexibility where I'll just kick it off to run overnight and by the morning we have a new updated website that's already deployed for us to start playtesting with this stuff. It allows us to faster iterate and think through that creative process. And also just empowering non engineers to be able to start leveraging these tools, not in a way where they're. I do grimace when I saw Brian Armstrong at Coinbase said that they're non engineers at pushing to production. Like I actually break out into hives in a cold sweat at the idea, especially of all businesses, a crypto exchange doing that. It's not like Coinbase was able to keep uptime, you know, during like market volatility anyway, let alone now all you need is like a bunch of non engineers like put in marketing people pushing code directly to a website.

[Colin]

Hey, maybe it'll actually help. I don't know, maybe.

[Charlie]

Who's to see?

[Rob Hamilton]

Well, it was really funny. You did see that base went down for like a couple hours. And the problem is when you are the block, it's one thing if Coinbase the exchange goes down. And there's actually some funny postmortems where you saw that like the entire thing ran in one Data center in US East 1 and there was the data center, like there was a leakage of a coolant system and you could mock them for that, but at the same time to run a central limit order book at the size of Coinbase, you can't make these a bunch of microservices spread all over the world because you need to market make and like fractions of a second, like microseconds are actually meaningful for being able to run a reliable, good exchange at scale. All this to be said though, I think it's fascinating and one to your point, like small crack team. I think that's actually more the direction where things are going to be going, where it's easier to maintain context with fewer actors and fewer people going around and changing the code and curating the engineering talent as is, and having good engineering talent that can be accelerated by AI, rather than trying to take median performers and make them better. With AI, you're just going to get further returns on investment with that for sure,

[Colin]

I guess. Last question. Fable looks like it's coming back out. Rumors. I think they're all but officially confirmed, but the government is gatekeeping. What are your thoughts as a startup, as a US citizen, as a technologist? What are your thoughts about this new brave new world of the government now getting involved with what we can and can't do with AI?

[Rob Hamilton]

I understand the reactionary sense of, oh, we need to hold on to this, right? Like this is the US's golden egg. So much of the GDP growth is tied to AI infrastructure and spend. There's a couple of things, though. And one, this is just math. This is matrix multiplication. Just like in the 1990s where cryptography fell under export rules. And this is where Adam Mack and others were printing out T shirts with large numbers on them because they were the prime numbers that were used for the encryption for rsa. You really can't put the toothpaste back in the tube. This is just linear algebra, matrix multiplication. The premise of making math a controlled export is a really difficult thing. Information ultimately wants to be free. And I think it's going to be a tough thing to reconcile if I'm trying to pull up the poly market odds right now for when Fable 5 is supposed to be coming back. I saw it was bouncing around last week by, by July 1st was like, you know, over half. So today's June 30th. Maybe we're closing in on that now. I think it's something that the US is gonna have to figure out, though. And the interesting thing is that even if the Frontier Labs start to lose their edge, I don't think it's the US government's responsibility to protect their margins. It was actually a Nick Carter post I saw earlier today. The US government's job is not to protect OpenAI and anthropics margins because all of the data centers, all of the semiconductors, all of the infrastructure still is going to be used. Whether it's a model from China or a model from the us. You still can have this massive capital and growth explosion and productivity boom that could happen from AI. I just don't like the posturing and positioning of. I feel like Dario at Anthropic tried to make it sound like, oh, we don't like this, this is bad, I don't want to do this. I think this was ultimately his intention was to try and make this a controlled substance and pull the ladder up behind him from export controls. So I think it's going to be a growth phase and we're going to have to learn how to navigate it. And as if we start restricting the flow of these things, the open source models are going to start catching up and it's going to make it kind of an irrelevant point is my theory. So we'll see ultimately where it goes over the coming months.

[Colin]

Hey, I think the bitcoin folks really do vibe with the open source model angle. Rob, thank you so much for your time. Thanks for our token bitcoin conversation on this live stream. Appreciate you. See you again soon. Peace.

[Rob Hamilton]

Thanks for having me, Rob.

[Colin]

Okay, we've got Kush Bavaria in the wings. We'll bring him up here. We'll talk about some compute here. Shortly before that, a word from our sponsor, Luxor Foreign.

[Charlie]

This episode is brought to you by Luxor's Commander Bitcoin miner management software for enterprise operations. Luxor's Commander gives you real time fleet monitoring bulk remote commands across your fleet and intelligent miner that is an automated profitability engine that runs every five minutes and tests your power settings against live hash rate markets and energy markets. Ercot back test show 10% improved profitability with intelligent mining versus binary mining. Commander Pro is $100 per megawatt or a 25 basis point pool fee adder roughly half the cost of competition and you can try it free for 60 days. So if you're interested in learning more, go to Luxor tech forward slash commander to get started.

[Colin]

All right, we have Kush Bavaria up in the wings. We'll bring him on up. Let's see if the mics work. Kush, welcome to the show.

[Kush Bavaria]

Thank you for having me again. We're excited to see you guys.

[Colin]

Yeah, likewise. Thanks. Third time's a charm. I'm going to drop you right in. I mentioned that I read through the latest ORN data research some great blog posts on compute futures and equities. I thought this was an interesting subject to talk about. You guys identified trends in GPU prices, rental rates and corporate stock performance with some counterintuitive insights. Let's start with the biggest one. It looks like you guys identified that Nvidia and H1 hundreds are correlated. Can you tell me more about what you're seeing here?

[Kush Bavaria]

I mean, I think it's pretty simple in the sense that Nvidia sells H100 chips. And then the easiest example that we look at is if you look at Exxon price and oil prices, if WTI prices usually go up, Exxon stock usually goes up as well. And we think the similar trend is true with Nvidia. So if Nvidia stock goes up, you see H100 prices increase as well. And the vice versa is also true in the sense that if H100 prices increase, then Nvidia stock goes up. And I think that correlation is stronger than the opposite correlation, if that makes sense.

[Colin]

Yeah,

[Charlie]

go ahead,

[Kush Bavaria]

you go.

[Charlie]

One of the more counterintuitive findings that you all had was this idea on utilization and rental rates. B2 hundreds and H2 hundreds have seen utilization rates slip, but the prices to rent them are still going up. Can you unpack that for us? It seems totally counterintuitive. What dynamics are at play, do you think that are causing that?

[Kush Bavaria]

Yeah, I mean look, I think the price are going up just because a lot of the contracts right now are being deployed on a longer and longer timescale. The rates we track are on demand prices, which means that utilization rates are going down. People are usually locking into longer term contracts. And I also think that that's primarily true. Prices have gone up since like May and April this past year. And it's something that we've sort of like tracked for a while now. And as you guys know, when we came back on the last episode we talked about how prices have increased and that price is still continue to increase and now it's sort of found a level that's stabilized. So I still think demand is very high. Just people are switching to more and more longer term contracts to sort of lock in the prices. I think the market's starting to realize that this sort of compute crunch that's happening right now is going to be last for a lot longer than people realize.

[Colin]

And this gets into an insight you guys brought up which is a 1/ hundreds, I guess you call it a vintage chip now from 2020 they're getting more expensive. And I think you even compare this to some crypto mining dynamics too. What do you think is I asked you last time, I want to ask you again. Chip depreciation, what does it look like? Is Jim Channels wrong saying that three years is a better chip depreciation timeline?

[Charlie]

Yeah, I'll ask it more directly. Are the depreciation concerns, totally overblown.

[Kush Bavaria]

I think it's a lot of it is just like news and media and people want to be like right on Twitter and go on CNBC and say that oh the chips are going to depreciate faster because I have a different take. And that's just not true. If you look at aws, gcp, Azure, any sort of pricing in the market today, every single chip, even the Ampere series, the Hopper series from Nvidia are being utilized the same level that they were six years ago or seven years ago. Which means that the value of the chip has not gone down over time. And in fact I would argue with new inference workloads coming online, the prices have actually gone up over time. And so you can even think about chips as an appreciating asset. At least for the last six years the values have gone up.

[Charlie]

Do you think that that indicates that end users are getting a little bit smarter about their token usage? You know what I mean? I can run an older model on an older GPU and maybe save on tokens rather than ripping newer GPUs on newer models if they the task is simplistic enough.

[Kush Bavaria]

Yeah, I think so for sure. And also people are writing better kernels for the newer GPUs and I think the biggest advantage that you have now is this chip state For a longer period of time people can optimize their code to run better inference on these chips. If you look at the GLM 5.2 outputs you see on I think it's artificial analysis. There's literally every single person has a new they try to outdo each other on throughput and so I think that's going to start to turn true as well as more people optimize their code to run better on even older chips and newer chips just because of how the sort of supply and demand is today.

[Colin]

Yeah, I was looking at total tokens based upon model on tracker recently and it looked like the Chinese models, the more open models really started being utilized like crazy over the past couple months. Especially post especially as like these frontier models started to get really expensive and around like the inference cost like freak out. I think that the Uber CEO said they spent tons of money on this. You know it seems like the frontier and the right behind frontier timeline is collapsing and I'm curious your thoughts on that and how it relates to the global compute markets.

[Kush Bavaria]

Yeah, I think look the US has made a mistake in sort of banning some of these frontier models. I think Fable 5 and then now GPT 5.6 as well. And that allows Chinese models and foreign companies to sort of come into this place and compete against these U.S. companies. I think we'll start to see that performance similar to Fable 5 or GBD 5.6 will come out of companies out of China and those models will be open source before most people realize. I think there's a rumor going around that GLM 5.3 is coming out soon that has capabilities similar to Fable 5. Whether whether that's true or not, I can't comment on that. But in general I think more and more people are going to switch to open source as you start to see government restrictions around closed source US models, which I think is just a mistake by the government.

[Colin]

So you guys have another piece published a couple weeks ago on Core Weave. I thought this was pretty interesting. Said Core Weaves GPU fleet which is eight one hundreds, two hundreds and Blackwell two hundreds saw rental rates climb 59% over six months. But the stock only gained 9% in the same window saying the market's pricing Core we've off its balance sheet and debt load, not rental rates. So is the market maybe lagging on this news or is it pricing something in that maybe the index doesn't capture again a little bit outside the like the orange wheelhouse. But what are your thoughts?

[Kush Bavaria]

Yeah, look, I think what's happening is that Core we obviously run very different business, right. So GPU rental rates are just part of the actual story of the whole chip. And it's similar to why we actually built a market around GPU hours. If you want to hedge against AI today, then your options are you could buy a put on Core Weave, you could buy a put on Nvidia. But that gives you exposure to many different things throughout the business. Whether that's your debt to equity ratio throughout the business, your cash flows, the CEO, how good is the company's management leadership, et cetera. And so I think core we have many different factors that influence and I think GPU hours are just one factor of that. And you see the price has gone up for Core we stock because of GPU hours. It's same for Net Nebs and all sorts of other NEO cloud players as well. But there's different factors in the stock price. It's not just pure based off of GPU hour pricing.

[Charlie]

Yeah. And that's on demand GPU pricing correct for what it's tracking. So part of me wonders, you know. Yeah, the rates went up 59%, 9% move still pretty big. And you know investors are likely factoring in how much of their revenue actually comes from the on demand portion rather than contracted revenue with, you know, other tech companies. But that also. Do you think that they might also be saying, you know, assuming this is surprising to the underside and not as much as you would expect to see move, Maybe they're saying they don't expect the rental prices to stick.

[Kush Bavaria]

I think the rental prices will continue to stick. And that's partly my take what it is, as prices continue to increase, people have more and more companies and more and more NeoClouds have updated their pricing schedules. I think the most Recent one was DigitalOcean that updated their pricing schedule. But many of these companies will continue to update their pricing schedule. And I also honestly believe that this sort of pricing crunch isn't like temporary. It's going to last for a lot longer than people realize just because of how insatiable demand is right now for AI chips.

[Charlie]

Do you have any idea from Yalls research? So this is going to be kind of a curveball and you can just duck it if you all haven't done this. But when supply would outstrip demand, is there at a point in which that you all have modeled out where you expect the supply to catch up with what we're seeing on the demand side?

[Kush Bavaria]

I mean look, in my opinion this is very much like a GMod's paradox that I don't think that it will ever sort of catch up to what the demand is going to be for AI chips. I think at some point, maybe down the line in 20 to 30 years from now, when the markets are find a better level and there's less demand coming from newer models or chips, et cetera, that will be true. But I also look at the metrics of how many people use these frontier AI models and I think the metric's very low. It's like 80% of the world has still yet to use any sort of like AI, whether that be chatgpt or even like anthropic. And only around like 2 to 3% of the world are regular users or power users of it. And so you can imagine around five years from now when the whole world is essentially a power user of this technology, what demand will look like and what new use cases will we will find for this technology as the price continues to go down per token and intelligence continues to increase. I think the metric we like to look at is intelligence per dollar and that number is just continuing to increase over time. I don't know. I think a good example again is also just looking at cell Phone usage. Right. I think most people in the world now have access to a cell phone and that's only true for the last five years, I'd say. So six years before that if you went to many of these third world countries, like no one had a cell phone. And now that technology has become commodified and so I think the similar thing will happen to AI.

[Colin]

Good point. Yeah. It's only very recently that most people in the world have Internet access. So last question, curveball question. OpenAI announced the jalapeno chip. It's an ASIC though different than GPUs. Nvidia's had the CUDA standard for forever. It's kind of defined this market. What are your thoughts on OpenAI's ASIC chip? And maybe just ASIC, the trend of ASICS in the near to medium middle future.

[Kush Bavaria]

Look, I think a lot of it is hyped up more than people think. And I think at least for OpenAI's case, I would be very bearish on Cerebris. More than anything, what OpenAI just did is replace the Cerebras chip. They were using Cerebris for a lot of their inference on their newer chips. And so I would honestly be more bearish on Cerebris than I would on Nvidia. OpenAI probably has some of the best relationships with Nvidia and most of their engineers code on CUDA and many other technologies that Nvidia provides. And I think it's still true that around 90% of the market in chips is still just Nvidia and that's due to many different use cases. But I think a lot of this ASIC stuff is just these are how media lifecycles work. Right? One person says it, another person says it and it keeps sort of like getting over and over again and then over time you'll see when you just actually look at the data in the market, who is using Asics? And the answer is like nobody. And then if you ask a startup, hey, do you want a Cerebras chip or an Edge chip? No one's going to say, yeah, I want it today. Everyone's give me Nvidia or don't give me a chip at all.

[Charlie]

And kush, is that just largely because of CUDA and the software mode that Nvidia has? Why wouldn't you move to towards ASICS at this point?

[Kush Bavaria]

I think it's that also. And also the financing mode that Nvidia has, they're faster to plug in chip into a data center. The Data centers know actually how to operate Nvidia chips. The Neo cloud family that Nvidia has built is just absolutely incredible and they've done a great job of distributing their chips as many players as possible possible. And if you look at companies like Cerebras and AMD and it's just the neocloud ecosystem is not there, they continue to have issues with their chips. No one essentially is using them in a real world use case today.

[Colin]

Kush, thank you so much for your time, really appreciate the insights. Shout out to Oren for some very good you guys have handheld me through this rapid acceleration into the world of AI compute. Thank you Kush.

[Kush Bavaria]

Thank you so much.

[Charlie]

Have a good one Kush.

[Colin]

Highly recommend everybody check out Orin the website. Read the blogs, they're fantastic. Pretty evergreen. We're gonna keep rolling. We are gonna dive into one last research report from the bank of International Settlement. But before that a word from our sponsor Lygos.

[Charlie]

58k gang was right. We are back at $58,000 per bitcoin and as a result lending desks are reeling. Especially if they have rehypothecated your precious, precious coins. Do yourself a favor, don't become the next FTX or Celsius victim. And if you're using a Bitcoin lending service, migrate to Lygos asap. Lygos Finance is our preferred Bitcoin backed lender here at Block Space they use Bitcoin smart contracts to make sure that you are always in control of your Bitcoin and no one else. With Lygos Finance there's no rehypothecation, no wrapping, no bridging, just 8 to 5, 8.5 to 10% APR financing with Bitcoin that's always in your control. To learn more go to Lygos Finance. We will go ahead and move on to what the globalists Charlie are saying about the AI bubble. They want to scare you out of the final melt up everybody. Not financial advice but they are actually pushing some pretty interesting data with regards to what we're seeing out of the current AI Capex cycle and why it might be time to start thinking about an exit. So here is the article. This came from their annual report which was very ominously titled, at least the first section, Progress and Peril. This is a larger Macro Econ report. It dealt a lot with the impact of tariffs on the US and global economy. Also of course the war in Iran and its impact on energy prices and everything downstream of that. But they also had a very fat section on AI's long run impact and its current trajectory. Also what they're seeing right now with regards to valuations and debt loads and where we might start seeing cracks in credit markets that could indicate this thing is turning south. So they have a section on long run impact about business productivity. I didn't find these that interesting actually because the base cases seemed kind of strangely lackluster. And you know, the mega impact growth line trend for transformative AI is kind of just you, you know, it's just way the upside. But they kind of obviously come to the conclusion that unlike other transformative technologies, you know, in the industrial revolution and et cetera, AI changes how thinking is done and the ramifications of that are not fully understood. But the more interesting part of the report for me were these kind of fundamental questions about sustainability and overall investment. If we look at Here on graph

[Kush Bavaria]

11

[Charlie]

hyperscaler CAPEX surging as a share of revenue. It is now estimated as of 2026 hyperscaler capex to hit 160 billion for the 160 billion for debt issuance for the capex boom in 2026. But that the Capex as her share of revenue has hit a 0.4 just above 0.4 ratio estimated for 2026 and that is up from 0.25 roughly or sorry, yeah just about 0.25 in 2025 and, and below 0.1 for 2024. And this is obviously something we talk about every day Charlie, with the debt deals being signed also here on the,

[Colin]

here on this section in the report. Graph C Comparing boom bust cycles Historically I've seen a lot of comparisons of the AI data center, build out the US highway system and other boom busts. And here we have the AI boom compared to railway mania, the roaring 20s, the.com boom. I think everybody's familiar with all those and one thing, Canal mania. Canal mania. I was not very familiar with AI 1830s Europe. Yeah, yeah, exactly. So learning about that now frankly AI dwarfing all these in terms of speed of ramp up. I think another thing that I, that I am just observing on this chart where the chart is obviously positioned as a way to say well it's gonna go down, the higher it goes, the lower it's gonna crash at the same time for these kind of being normalized of multiples of boom bust for the 19, the roaring 20s, the 1920s which preceded the Great Depression. For that to be the smallest boom bust cycle on this chart is pretty dramatic. So I feel like there's something missing here because if you Were to compare the AI boom over any to overlap it on like S curves of adoption, I think it would look pretty normal. So this does become an existential question. Is this just a market cycle? Is this the secular adoption of new technologies? What is it? Is it perhaps all of the above and coinciding with a breaking of the financial system in some way? I think the answer is all the above.

[Charlie]

And I think the other interesting thing here too is that the dot com bubble is actually not the greatest parallel to what we're seeing right now. Yeah, because that was largely investments in software. Obviously there was a fiber optic boom at that time, but it wasn't the same level of infrastructure spend as we're seeing. The better scenario or better examples there being the canal boom and the railroad boom. With the canal boom this happened in the 30s and 1830s, right before railroads became commercially viable. And then that boom came in the 1840s, 50s and 60s. And basically that's why you're seeing the canal boom being the worst crest to trough here on this chart. Because railroads ended up quickly superseding canals as a more efficient way to actually transport goods.

[Kush Bavaria]

Right.

[Charlie]

And transport, you know, transport raw materials for industrial uses. But so if we want a lot of other folks that said this, the better analogies are for the Capex boom, oil and gas and the railroad boom in terms of over investment or in terms of just total dollars spent at that time. So but the TLDR here, you know, is that they, they kind of frame this as a winner take most dynamic where there's so much pressure to invest right now because everyone is scared about being under allocated and being left behind in this race. And that's leading towards what they find to be probably more spending than is necessary for the current environment or for what demand might look like in the future for all these data centers that folks are building. All right, the next section that I think is really worth looking at is the valuation section. A lot of really good charts here and probably a good place to start would be the 30 largest stocks across the S&P 500 topics. And the stocks Europe 600 are exceeding the highest historical performers in those indices. So the basket of companies that are trading on this AI boom right now are now seeing valuations and multiples that are higher than the outliers within these individual indices during other booms and busts. And this feeds into the fact that investors are now getting increasingly more comfortable with volatility within this cohort. One of the more interesting charts Here on graph 12, chart B, they show Sharpe ratio differences between pre and post pandemic markets. The Sharpe ratio is a measure of risk adjusted returns. It looks at the returns of a stock and then takes into account the volatility of the stock over that time period and gives you a ratio for the returns relative to volatility. And a lower Sharpe ratio shows a greater risk for the return. Higher Sharpe ratio, less risk. If you look at this post pandemic stocks are now trading with much lower Sharpe ratios than they were pre pandemic. And the bank of International Sentiments points out investors are getting increasingly more comfortable with higher levels of risk for greater or sometimes similar returns that they were seeing pre pandemic, which is really setting up a scenario for, as they call it, complacency. And this kind of reminds me of the new paradigm memes with, like John McAfee, Charlie, with crypto, where people are starting to get the idea that this is the way an AI market trades. Maybe this is the way the markets are going to continue to trade. Maybe the valuations are actually not that engorged. When we consider the value creation that AI will bring to the market. No one knows that right now. Obviously that's why these things are trading so volatile because people are really struggling to figure out what's the actual fair price for these AI companies considering the revenue that they're bringing in. The other big one here, going back to using the dot com bubble as a corollary, stock ownership among Americans is much higher now than it was in the dot com bubble. It is significantly higher, which you could read as a bullish and bearish indicator in some ways because if we got to like we're, we're approaching or@.com bubble valuations in terms of multiples for most of these companies, if you look at price to sales and, and price to earnings. So if, if you were a bull on AI, you would probably use this to say, well, we have higher to go here because actually there is a greater level of ownership, so the concentration is maybe less so. And who's actually invested right now versus in the dot com bubble. So maybe there's more room for this to grow. But they're looking at income percentiles and from the 0 to 20% income percentile that saw the greatest increase from roughly 5% of 6% of net worth up to 16 to 17% currently the 20th to 40th income percentile actually dropped their stock ownership. Funnily enough, more hollowing out of the lower middle classes here. 40 to 60% went from 10 to 15 from dotcom to the current AI bubble the 60th 80th percentile didn't move that much. It went from about 14% to 16 and then 80th to 99th went from about 11 to 16 and then the ultra rich, the 99th to 100th percentile saw the second largest gain outside of the bottom, the bottom quintile. It went from 29 or sorry 39 to about 51% of net worth in the stock market. I think this is a horseshoe theory for.

[Colin]

This is horseshoe. This is barbell theory. This is barbell.

[Charlie]

Yeah, this is barbell theory.

[Colin]

Excuse me.

[Charlie]

Thank you Charlie. For financial ownership, the lowest percentiles in the US and the highest percentiles have seen their, their net worth invested in the stock market go up since the dot com bubble which to me it's kind of funny. It's like the, the you know, the capitalists and the actual shakers and movers for the economy versus like the Wall street bet degens aping their entire paycheck into you know, Gamestop or what was the AMC theaters.

[Colin]

Wendy's. Wendy's is really popular right now if you want to know what's going on on the Internet. Wendy's.

[Charlie]

So but I thought this one was extremely notable. If you look at the fact that their point in pointing this out is like with the dot com bubble you have a chance to really disrupt people's or household net worths here. And savings account which the stock market has become a savings account for majority of Americans.

[Kush Bavaria]

Right.

[Colin]

The 6040 portfolio is dead.

[Charlie]

Yeah, we're going 50% stocks, 50% Bitcoin. But I think this is worth flagging though in terms of existential risk. It's not just corporate balance sheets that are going to be potentially hampered by a credit crisis within the AI boom real retail investors could lose out on their savings if this thing goes south. And that brings me to the next interesting section which looks at debt. They have a few charts here. They show that AI credit risk is rising when you look at the change in spread credit spreads versus other benchmarks. Now I think there can be some noise here and some signal because we've actually seen for certain companies like Core Weave and other high performing AI companies in neo clouds, their credit spreads have been shrinking recently largely because their bonds have been or their credit risk has been downgraded. Their credit profiles have been upgraded to triple B and sometimes in some cases a tranches. But for the sector as a whole, the bank of International Settlement shows data here to say that the actual credit Risks or credit spreads for these AI companies are rising. And also they have a good chart here on circular AI financing. So we haven't really covered circular financing that much. But the basic idea here is that like one company invests into another company, so company A invests into company B, then company B then pays for company A services. So it's kind of the idea that you're making an investment in a company that's going to use you and then those investment dollars kind of flow back to that company. And it shows here that the chip makers, it's like all circular financing for them. And then the hyperscalers, it's split about 50, 50. And probably the most salient graphs though for gauging actual credit risk or concentration within the market comes in graph 14 here. And I'll close on this Charlie, and throw it to you for closing thoughts. They show a chart that looks at corporate equity and credit risk in terms of month over month returns from January 1997 to May 2026. And it tells a very clear story that's pretty intuitive. When the S and P rips, the credit spread for bonds shrinks because risk is going down for holding those bonds because the companies are cash flowing, they're beating earnings. When the S and P tanks, the spreads blow out and they start widening. And if you want to see how, for when the, if you want to see nightmare scenarios for this for high yield bonds, which are not investment grade bonds, they're companies with higher credit risk, you can see these spreads blow out in October of 2028 or sorry, it's October 2008 during the great financial crisis and March 2020 during the flash crisis when Covid started becoming a reality for economies around the world and everyone started shutting down. Those are the highest points in terms of credit spread increases on the chart. The other chart that's really worth mentioning is exposure to software borrowers is becoming increasingly concentrated. This, in this report, they found that currently the number of divisors, development companies as lenders per issuer in the seven plus tranche. And then what's. What, what that's saying is there are seven plus credit development companies which are basically funds that are issuing credit to these, to these AI companies. The tranche of seven plus in terms of. I'm sorry, I'm talking over myself. Let me, let me start over here. Multiple credit or multiple business development companies are issuing to the same borrower. So in this chart it shows that 40% of software loans in 2025 were coming from or were taken in by borrowers that saw seven or more lenders to that company. So basically you have multiple lenders issuing to the same company, to the same counterparty, meaning that they all have the same risk. Risk is becoming more concentrated as these companies end up lending as multiple companies lend out to one borrower. And bank of International Settlements basically points out to say it's like this could lead to real contagion. If all of these business development companies are lending to the same borrower, that borrower goes under. You have contagion across multiple different firms. And these business development companies are stood up by investment banks and they're a way for retail investor, or sorry, institutional investors to fund the business development company and then they will go out and lend to these corporations that are seeking debt to build these AI data centers. So the fact that you have more than ever multiple different business development companies lending to the same borrower, bank of International Settlements flags that as a chance for existential risk in the system if one of those defaults.

[Colin]

And to backdrop all this, you know, it's been a whole, it's almost been a whole week since I've seen a headline about Iran and Hormuz. But realize that just we just spent the past two or three months going back and forth on this. Maybe there is like resolution in just the fact that perhaps the eye of Sauron, the United States is going to pay attention to something else. But the reality is all of those bombs and missiles and closing and hemming and hawing according to this report, does create inflationary pressure. So regardless of your view on credit risk or the AI bubble, the big looming thing is not really specifically whether AI is in a bubble or not is this, this global inflation across the board you see on this chart, severe energy shock increases, inflationary pressures, fertilizers, plastics, skyrocketing over the past couple months. It hits mainly as you can see here, mainly Asia, Asia, the like majority like the, the lobes like productive manufacturing, production, production capacity. So these semiconductors, this is where things are built. And so this is the backdrop for all of the doomerism described in this report. So you know, whether or not it's this AI thing which is totally going to dislocate markets, we do have this almost unavoidable wall of inflation slowly marching towards Bethlehem to borrow.

[Charlie]

Yeah, and I think how deadly that inflation will be will be directly linked with how quickly the US can actually extricate itself from this war. The peace deal seems to be kind of faltering right now.

[Colin]

Yeah, I mean hard to tell.

[Charlie]

I honestly at this point. Any peace deal comes out, I kind of just say, okay, well, it's probably not going to work this time. This is like what the umpteenth time tried to do it.

[Colin]

I know how to solve this. Just get us involved in a different war. Let's go to Cuba. It's funny because it's true. It's funny because it's true. We got to move the fleet down to the Caribbean and then do what we have to. We have to complete the trifecta of the three pariah states that Trump doesn't like the most. And that's how we extricate ourselves from Iran. We basically get distracted, like, like Andy putting down the, you know, the play with you. I don't want to play with you anymore. And it's him picking up buzz and it's Cuba.

[Charlie]

And, you know, last note on this, Trump basically came out, I believe it was in. I don't know where the press conference was held, but he basically said part of the reason why they want to stop the war is for the stock market, which. There's a very, There's a better way to say that. It's like to make sure that global economies don't fracture, which is basically what he's saying in some way. I mean, like a charitable reading is. That's what he's saying. But to just go out and say that and show your hand and show that, that's the one thing that's really making you lose sleep about this.

[Colin]

That gives your enemies and your opponents a very obvious way to attack you. Make the stocks go down. It's ridiculous. Meanwhile, mainstream media is obsessed with this, like, pond somewhere in Washington, D.C. obsessed with whether or not, like, there's some, some markings on the bottom of it, not whether or not we're, Whether or not the President is, is. Is creating havoc for international economies and lives. No, we got to care about whether they drove across a reflecting pool.

[Charlie]

Priorities, Charlie.

[Colin]

Priorities, priorities, priorities. Priority right now is to get out of this stream. Thank you so much for listening. We go live every single weekday, 1pm Eastern. Featuring quick hits on data centers, AI, Bitcoin, Bitcoin tech markets and emerging technology. We feature guests of the day, really insightful folks, and sometimes deep dives into

[Charlie]

research and investigatory pieces.

[Colin]

If you like what you hear, you'll love reading it. Newsletter.blockspace media.com this live stream is sponsored by CleanSpark. Nasdaq listed ticker CLSK. Thank you, Clean Spark. Catch you all tomorrow.