← Blockspace: AI & Bitcoin
Blockspace: AI & Bitcoin
HIVE Exec Chair Interview, BTC Mining in South America, Kimi K3, Compute Exchange’s Secondary GPU Market
01:08:56
Loading summary
Transcript120 lines
- [00:00]
A
What's going on, y'? All?
- [00:00]
B
Welcome back to Blockspace Live for the final live stream of the week on this beautiful Friday. We've got a packed guest schedule today, Charlie. We've got Executive Chairman of Hive Digital, Frank Holmes on for our first segment after our hash rate index update. Following that, we have Luxor's business development lead for Latin America, Alessandro Cerseri, on to talk about bitcoin mining in South America, across Brazil, Venezuela, Paraguay and everything in between. Then we will be touching on what else but Kimik3, the open source model that has thrown the AI sector into a panic. And this pairs really well with Xi Jinping's latest speech on AI and showing the world that, guess what, the Belt and Road Initiative is coming for the clankers as well. And after that, we have Carmen Lee, the CEO of Compute Exchange, on to talk about their latest service, specifically a used and refurbished hardware market for GPU trading.
- [01:04]
C
That's right, Block Space goes live weekdays at 1pm Eastern, featuring quick hits on AI data centers, emerging markets, emerging tech, if you're lucky. We'll talk a little bit about bitcoin. Not that much today, though. If you like what you hear, you'll love our newsletter in your inbox every single day shortly after we wrap up the stream. You can find that@newsletter.blockspacemedia.com and everything we do, written work, published work, RSS livestream, you can find on our website at Blockspace Media. There's a host of content there. Speaking of hosts, this show is brought to you by CleanSpark. Nasdaq listed ticker CLSK. More on CleanSpark later on in the show. Colin, let's kick it off with some hash rate index. Let's do this today.
- [01:55]
B
Yeah. This is bitcoin mining data brought to you by Luxor's Hash Rate Index. And a bit of a bright spot here, Charlie. Not much. One, we're taking all the rays of sunshine we can get, but bitcoin is back at roughly 64,000 spot. Hash price is at $31.62PETA hashes per day. We've got an estimated Difficulty adjustment of negative 1% coming up in about two weeks. We're halfway through. We'll see if that holds. It is curtailment season across the US and in Texas, so good chance that it does. And we're just coming off of a minus 5% difficulty adjustment on the most recent round of difficulty adjustments, so kind of more of the same. As we covered with Khan Ferrani last week, we're actually seeing a Hash price regime in which BTC's dominated hash price has been doing a lot better than expected. It's actually been recovering since November of last year. October of last year. November of last year. Given the fact that hash rate has really not had a good recovery. We're below 1 zeta hash. We're currently at about 900 exahashes a second. And so if it weren't for the fact that bitcoin price has depressed, you know, the hash price regime actually looking pretty decent now, if bitcoin's price was flying from here, no doubt we would see that BTC denominated hash price fall a little bit as more miners come online to take advantage of the increase in revenue. But overall, not too much different from what we had last week. Hash price on the US dollar denominated, doing a little bit better with bitcoin recovering, but still a long way to go before miners are really raking it in.
- [03:39]
D
Yeah.
- [03:39]
C
With Texas 4CP in effect, it's not a surprise we've seen hash rate stall that, you know, eventually when the difficulty adjustment hits, then we see, yeah, hash price sometimes recover. Bitcoin price has recovered slightly. Bit 63, 64k if we're lucky. It's weird to say that that is now like it feels it's very normalized all of a sudden, but not gonna look at that too closely. We've just got a little bit of reprieve in that we're just not plummeting further right now for hash price. So shout out to all the people who are sticking in there and mining bitcoin. Still, our heart goes out. All right, we have Frank Holmes, executive chairman of Hive Digital in the audience. Will bring them up here shortly after a word from the conference happening next week.
- [04:40]
B
If bitcoin's actually the best money and
- [04:42]
A
it's the thing that people should accumulate is the best risk adjusted asset, I'd lose zero sleep about whether or not
- [04:47]
B
that's going to happen.
- [04:48]
A
I just ask the question of when.
- [04:53]
B
It's literally matrix math that you're running
- [04:55]
D
on large pieces of data, the bitcoin
- [04:57]
E
miners can absorb that energy.
- [04:59]
C
And in many ways, this feels like
- [05:00]
B
a second bite at the apple to
- [05:02]
C
build a new Internet.
- [05:12]
B
All righty, Charlie, let's get the one, the only Frank Bulldog. Holmes up on the screen. Frank, welcome back to the show, sir.
- [05:22]
A
It's great to be with y'.
- [05:22]
C
All.
- [05:24]
B
Great to be with you too, Frank. We've got a lot of ground to cover. We're going to be focusing on what else but HIVE'S AI buildouts. And I would love to start with Bowdoin Sweden, because that has been the biggest or hottest topic so far on Yalls AI revamps or recently rather given the news in late June. And could you give us an update on that deal specifically if you can at all, where the definitive agreement stands today for a tenant and whether or not Hive is waiting to put in a GPU order until that is finalized?
- [06:00]
A
Well, I think the big part, the order that has to go in when you're going from tier one to tier three is you have to go buy these lead items and a lot of them are year out. They're air conditioning basically. And for simple math you can look how big this build out is if a gigawatt electricity will consume 50,000 tons of copper, just copper. So then you look at copper making an all time high recently. So you can see that where is the supply coming from? From old mines and their degrade is going down. It's having this AI boom and data centers is having this sort of collateral looking at everywhere else. And lithium batteries, lithium is one of the best commodities also because you need to when you do a tier 3, backup energy systems in case the electricity comes down. So when we look at something like why tier one is so attractive for the hyperscalers is because the permitting has already been done. You built the substation, you have all the wiring that's there. You laid out a cement floor, they may have to redo the cement because the new clusters are so heavy and the ceiling may not be so high, high enough for the new structure. But you have 90% of the shell and electricity is there. But you have to bring in more air conditioning because the brain, when you build a brain, it gives off so much heat. And so therefore you see on 100 megawatts, almost 40 megawatts is going to be H vac air conditioning, cooling of some form. Whereas in bitcoin mining, well, what we do in Montreal, we recycle it and we save the pool maker beside us $300,000 a year on his electrical bill for heating his building, which is five times our size. You're not going to be doing that for the AI world as of yet right now. So when we look at Sweden, and that's where that was one of the largest data centers, mining Ethereum. And it was a very profitable business, but it was gpu. So we became experts early being the first crypto mining company. But being big in Ethereum taught us about GPU chips. And for your listeners, a Asic Chip is like driving a jeep or a Bronco. These are resilient. The mechanics are everywhere. They know how to take it apart and fix it. When you start going into GPU chips, you're now got a Porsche. Oh, better engineers, you need some. They're not everywhere to fix them. And now we go into Nvidia GPU chips. Oh, we've gone from Porsche to Ferrari, we've gone to Bugattis now. I mean like the cost of these chips aren't $4,000. They're all up on the air conditioning. I talked of at the beginning $70,000. That's. So when you look at Cambridge University, their great study that build out in bitcoin mining, it's a million dollars a megawatt. But it's the most important part is the shell has been built. The electricity, the permitting, everything's in place. The hyperscalers are looking at three to five years starting from scratch to go buy the, get the power, get the land, get the permitting. And nowadays you get these cave people. You know what cave people stand for? Citizens against virtually everything. And they come in and they reverse your contract. So now the big data center guys are getting really worried that they get approval and then income, the cave people. And they upset and give me all this FUD they put out. But you and I, the three of us here and all your listeners, we're used to dealing with fud. We're resilient. It's like the Monty Python movie. You know what army headbutt you and cut it off. Because we're used to confronting the FUD. Well, the, the FUD that's coming with the cave people. It's really unsettling right now. But they don't have to worry about when you've already built out the data center. Tier one data center. The fast track to tier three, your big spend is going to be the chips. They're really, really expensive. About 70% of that spend is the cost of the chips. And then you're going to have the H vac and all the 30 to 40% is going to be the need for air conditioning. So when we look at Sweden, we have electricity, we have a beautiful beautiful building. And how fast could you convert that to tier 3? Really fast. But we didn't own the land and we just bought the land. And that was really, I think key element in the announcement of having a contract from another hyperscaler to come in that wants that, that energy that we have and the data center. So that's where we are with Bowdoin. And you know the frustrating part I share with you, I said we have to get on a. As a. And this would be great for y' all to be able to educate people. Do you know what we paid in taxes in just in a short period in Paraguay for the government? Over $50 million. What do we pay in taxes in Canada? And they're anti bitcoin mining over 100 million of which they're supposed to give us back 50. So therefore, net we give them 50 million and they're trying to make up things of the new rules. They have not government approved the agency to say, oh, bitcoin mining is bad. So therefore you can't get your depreciation and you we're not going to have to pay you back what you owe you. That that's what we deal with. Same thing in Sweden. We did the calculation in Sweden. We paid that government $50 million. So anyone ever tells you that this is a nefarious business, it's all fud so and so coming back to it, it's on track to become, you know, the. A pure hyperscaler AI factory in Northern Sweden.
- [12:16]
B
So going to some of the other builds as well. You mentioned lead times for some of the materials. Obviously this is super complex. Going to tier 3 from tier 1. Curious if you have any updates for the merit facility as well as I understand there's the first installation phase of 2300 GPUs and then a later installation of roughly 2100. Is that still on track for late 2026 or early 2027?
- [12:43]
A
It's so exciting for us. You know, we still are the most efficient miners. We function over nine time zones, three continents, enough to do translations in five languages. We're not just in the state of Texas. Some people are only in one state and they're still not as efficient. So when we talk about that efficiency of running a data center, I believe the capital market. So we're the first Canadian company to be able to do this transformation of saying Canada has these tax laws. You can't do a convert without having a coupon on it. Well, we found a way to go zero coupon. And we. And so we're very innovative and creative with that. What that did is it ignited us to give us that cash, zero cost of capital to go and put the down payment for the GPU chip. So when you're buying the GPU chips a year ago was very difficult to fund them. Today it's not Today there's more and more investors like BlackRock has a whole fund just lending against GPU chips. The depreciation of a ASIC chip is a lot shorter than now. A GPU chip is six years. So therefore it has a longer life to it. But you get your money back over three to five years. But it has this longevity to it. So I think from that end, the capital markets for funding, the growth of AI is evolving very quickly with many different pockets of wealth. But for us it's been the growth of convertible debs that have gone from 100 billion to 500 billion. And it's all based off of like a quadratic formula they use to go short, long, short, long. It creates more price discovery and liquidity. But for us, what's been really helpful is that we take that money and like buying a house, you put down 20%. Well, we're going to put down 20% and we're going to get 80% lending. And where all of a sudden those contracts are funded. So what does that mean? $100 million is like $500 million. Does that help you?
- [14:39]
B
So yeah, I guess. My next question is going on the financing angle. Kind of a two part question here. How will you leverage the roughly 130 million that y' all just raised out of the convertible note and two for the Toronto data center, which is going to be massive. There's a roughly 3.5 billion dollar Canadian US dollar investment involved in that. Where will you seek to tap forward financing for that build?
- [15:06]
A
Well, it's a great question, but you know, when we put it together, we immediately got offered to three times the real estate value. You know, it took us 18 months to consolidate 25 acres. A touchdown bass from the airport and halfway between a corridor between the smartest universities in the country. One where Ethereum came out of Waterloo University and now the noble lord University of Toronto in between. Those 60 kilometers is tremendous brain power. And now the country under Carney has a minister of AI. And that's very, very different to have someone just dedicated to it. And the province premier is giving money to universities that come with AI curriculum now. So you'll get $5,000 a student if you are teaching this AI. So we feel that we have the right politics now. Two years ago, no, it was broken in Canada and it's coming together. Same thing happening in Paraguay. And now the Central bank of Paraguay is characterizing bitcoin mining as an export, which is phenomenal because it is. And now we proved out with the University of New York that they can run off our a 40s that are down in, in Paraguay, 5,000 miles away, they can create models that distance inexpensively. So that means that Paraguay is able to take their electrons and export them as compute power. And, and the first part that came up only this couple weeks ago was hashing power is being exported. So it goes to a, a pool in America which is socks one and two compliant, which is all great, but we basically sell to them our hashing power and they give us our Bitcoin every day. And the government is recognizing that. I think these are really quickly happening dynamics that are creating greater value. Now in Canada and countries like Paraguay and Sweden, they're very, very focused on what they have to do is a fast track, the AI build out.
- [17:24]
B
So Frank, that actually leads into my closing question pretty well. Talking about paraguay, I believe Hive has 400 megawatts in, under management in Paraguay currently. Are there any plans to convert that to AI and hpc? And if not, what are some of the hang ups that would that are getting in the way of turning that into an AI and HPC factory? Or is there just no desire to do that because it's such prime real estate for bitcoin mining?
- [17:51]
A
The country originally had very short term energy contracts which make it very difficult to attract US capital. They're looking for much longer, 10, 15 year. The longer the energy contract, the better it is to fit into a net present value model and feel safe about that source of cash flow. President Pena, who went to the University of Columbia in New York work, worked in Washington D.C. for the World Bank IMF. He has all these relationships. So he understands this. And they've announced about three months ago an AI, four months ago, an AI strategy that they will offer up to 15 year contracts, stable electricity. And then it got the politics back and forth just like we have getting the Clarity act done. But this other act will get done faster probably there than the Clarity act will get done here. And, and that is coming out with, and we're expecting this in the next couple weeks, that 15 year contract and what the commitments are. So they're there and we're building out 100 megawatt, a 300 additional expansion substation. We can grow from there. We've been land banking like we did in Canada, buying as much land around our areas as we can. And that's, and we've seen that take place. And now when they made their announcement about this AI strategy of long term energy contracts, the Taiwanese government is there, they're very close to Taiwan and they export a lot of food to Taiwan. Well, all of a Sudden, now they're looking there. And so I think that eventually down the road, the opportunity for hive and Hives buzz is about a gigawatt of Runway potential in Paraguay. What we know, we have contracted now that 100 megawatts that we're going to do the bitcoin mining, it is going to be repurposed for AI and so that is very positive. And we have people knocking on the door saying, look at it. How fast can you build there? We can build there faster than we can build anywhere else. And so that's what's really important to recognize. And the government has a surplus electricity. They love the idea of the concept of exporting electrons, not just wheat and beef and corn. So it's very exciting to be there. And the country loves America. That's also very important from a geopolitical risk analysis. They're very pro America. And what every bitcoin would like to hear do you know, they never had income taxes until 2009, and it's only 10%. So you have a lot of Europeans now seeing moving there from Spain, going there because. Because they could turn around and be what they like to call what I know from my jets ETF, there's this whole trade of 50 million people. You're moving around, and you can run your program from Paraguay. You can do your coding from Paraguay. Well, the tax rate is so much less. And then you can fly anywhere. So that transient coder, strategic consultant. These are big numbers of filling up the airplanes. But now we're seeing these other benefits happen in countries like Asuncion.
- [21:20]
B
Well, Frank, thank you so much, man, for the update. We'll be keeping an eye on developments throughout the year. And I do just have to say, now that y' all have got that office in San Antonio, y' all suit you. I'm happy to hear it. Entering in the vernacular. So anyway, hope everything's going well and we'll see you later on.
- [21:38]
A
Will you be in Dallas next week?
- [21:41]
B
We won't be, no. Unfortunately.
- [21:43]
C
We have a live. We have a podcast to run. Frank also, great looking hat. Very clean. So.
- [21:48]
A
Oh, we got a new one for. For the show.
- [21:51]
C
The hat.
- [21:52]
A
New hat. It's. It's the silver, and that's for Buzz. So, AI, powdered blue swag is. Is silver, but for bitcoin, it's gold.
- [22:02]
C
Sweet. Hot.
- [22:04]
A
Do some swag.
- [22:05]
C
Yeah. Yeah.
- [22:06]
B
Thank you, man. Hey, appreciate you. I have to have you back on in the latter half of the year when we have more developments.
- [22:17]
C
Love, Frank. Super cheerful guy. Love hear him talking About Broncos as a six Porsches.
- [22:25]
B
That was really fun. Also like articulating the sovereign individual thesis there.
- [22:30]
C
Towards the end, man was about to give us a treatise on on flag theory which maybe for another episode. Anyway, we've got Alessandro in the wings. We're going to go back to Latin America, talk more about Paraguay and other Latin countries and bitcoin mining. But before Alessandra, a word from our sponsor, CleanSpark.
- [22:50]
B
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 bitcoin@cleanspark.com all right, let's get the sultan up here. About time, Sultan.
- [23:26]
C
Bienvenidos Alessandro. Welcome. You gotta unmute really quick.
- [23:31]
D
Howdy folks.
- [23:32]
C
Hey, good to be here.
- [23:34]
D
Was it was about time.
- [23:36]
B
Yeah, it was about time. Great to have you on here man. I thank you for joining. Got a lot of ground to cover. I, I think that to set the stage and to let people know you, you do business development in South America for Luxor and you have this great bitcoin mining around the world specifically focusing on South America. And I believe right now you have articles out on Brazil, Venezuela, Paraguay and kind of like an overview of Latin America as a whole. So I think that's a good place to start with regards to Latin America. Where are the clearest opportunities right now and what are some of the more underappreciated jurisdictions? And then we'll move into Brazil, Paraguay and Venezuela individually.
- [24:17]
D
Yeah, I think look, if when we're speaking about Latin America specifically, Paraguay always comes out as the highlight. You know, paraguide is currently the fourth market with, with the most hash rate in the world. And so they, they grew one extra hash from 43x a hash to 44 in in Q2 of this year. Year over year the growth has been more like 25%. But when you look at the rest of the markets in Latin America, I mean besides Paraguay, the second biggest one is Venezuela. Coming up at it grew 1 extra hash as well, but from 5 to 6 extra hash. So Paraguay, Bolivia and Venezuela were among the few countries that grew their hash rate activity during the last quarter. So again it's interesting in terms of the opportunity look at Paraguay. It seems to be this market that opened up to miners got a little bit more difficult in terms of regulate from a regulatory standpoint over the years. Miners started with like a 3 cent kilowatt hour cost back in 2020 when they opened up to mining from the operators on the ground. Now nowadays in Paraguay what we're being told is the mid, the small, the midsize operator is being charged more like 6.8 kilowatt hour. 6 kilowatt hour dollars per kilowatt hour. I'm sorry, so you, you. They were hiked. Their, their electricity costs were hiked basically 100 over the course of four years. So what you're seeing in Paraguay specifically is a concentration and this is a good follow up to, to your Frank Holmes interview, but it's a concentration of the big player. So you got Penguin, you got Boyden, you could hive in Pair White and then a few others like I think social input group that bought the Paso Bay site from Bit Farms last year. So but other than that, I mean those four, the estimates are there's like a gigawatt worth of mining activity right now in Paraguay. 700, 750ish megawatts out of that gigawatt. Are those four players.
- [26:30]
A
Yeah.
- [26:30]
B
Interesting. I want to return to Paraguay here in a second as to like to kind of explore what the ceiling on that capacity is, especially in the context of the AI debate. But I want to go ahead and go back really quickly to y' all's hash rate index, hash rate heat map and return to what you said about Venezuela kind of coming out of the back of the pack. It used to be one of the premier places to mine in Latin America. Now with the regime change, it seems like interest is flooding back in. And I can't help but look at this within the context of Brazil just to the south and Venezuela leapfrog Brazil according to Yalls data. I'm curious as to how that's possible considering Brazil shares the Typhoo Dam with Paraguay. They are flush with hydroelectric energy. They're an oil and gas powerhouse. Why aren't we seeing more bitcoin mining take off in this country? And then I want to shift focus to why Venezuela is doing so well.
- [27:25]
D
Yeah. So Brazil is an interesting plaza. It also comes down to a regulatory conversation specifically I would like say twofold. So the custom side of things. So the importing of, of hardware. First of all, it is illegal to import use any use thing into Brazil. So the, the conversation of trying to import use a 6 and see Brazil kind of like Venezuela has worked over the, over the past decade as like a ASIC cemetery where all the use AS six that are unprofitable in the US the rest of the world fly over to Venezuela. That consideration in Brazil is particularly difficult. It's not impossible. But the ton of paperwork that you need to get, that you need to get in order to import use ASICS is just material, materially considerable. Okay, so the second one is the how the energy market is structured in Brazil. So Brazil has basically at a broad level two different energy markets. Okay. So the regulator energy market and the free energy market, they're, they're starting to deregulate the market for large load consumers. So essentially miners and data centers because the government is starting to understand the trend. So basically now as a bitcoin miner, starting in August of this year, you will be able to negotiate energy contracts directly with the energy producers. Before you would only fall under the regime of the regulated energy market. And so you would effectively pay like 19 cents a kilowatt hour, 15 cents a kilowatt hour. So it just makes mining economics impossible. So the concentration of mining in Brazil is particularly spreaded and among like small to mid sized miners.
- [29:15]
C
So let me jump in and ask about Venezuela. Kind of your, your home, home turf there. The US just enacted a regime change and Venezuela had a lot of bitcoin mining before that. Has the regime change altered the landscape? Does it make it more, Is it make it, does it provide stability for bitcoin miners or is there other benefits that or I don't know, what's your thought? You're like the person to ask about this.
- [29:49]
D
Yeah, I, I appreciate it, Charlie. So look, it's interesting because the password of the again grew 20 during Q2 while the government is rationalizing electricity at a nation state level. So how do you, how do you grow hash rate while you're rationalizing? Well, the, the honest answer is the biggest mining player in Venezuela is the state. So in that regard it may be like a picture similar to Russia now in the generate the all, all of our generation assets and the distribution lines in Venezuela haven't been maintained over the past two decades. Okay. One of the things that the US is is pushing for that will hopefully open up the market is the privatization of the electrical grid. Because essentially up until like a month ago, the only one that can supply electricity in Venezuela was the state. And your auto generation and auto consumption was limited to up to 2 megawatts. Okay. There used to be a cryptocurrency regulator in Minnesota, the Suna crypt, which also regulated mining activity in the country. So you needed to get a license to legally operate in the country. That regulator is currently being restructured. So speaking from a Venezuelan law perspective, there is no way for you to legally be authorized to do this mining activity as a private individual. So most of the 6x a hash that we're seeing, we can estimate that it's like 90 related to the state. Now what the US is helping us do though is open up the market and privatize different industry. Because when we're specifically talking about Venezuela and this is not only a thing particularly to the mining industry, all the industries are fully owned or semi owned by the state in Venezuela. And that's what the US is trying to push through an OFAC and the treasury framework where basically there's now this like by law system in Venezuela where you can be approved as an American entity by OFAC to generate, consume, distribute and sell electricity and other stuff like oil and gas related purpose related activities. But there seems to be a push from the U. S regulator to open up to privatize the electrical grid and also open up mining. There's some discussions to try to like from my understanding try to include the importation, commercialization and operation or in operating basically bitcoin mining hardware and AI gpus and like AI hardware inside Venezuela approved by like the ofac.
- [32:39]
B
So kind of on the same vein Alessandro, but like less on the bitcoin mining side of things. Just curious, as a Venezuelan yourself, what changes are we actually seeing in the country now that it's been regime changed? I think a lot of Americans are kind of in the dark on this and depending on your political allegiance like you're going to get certain news versus others for how it's going. What, what's your honest assessment of like what has or hasn't changed since Maduro was taken out of office?
- [33:10]
D
Yeah. So from the data points, oil production has been hiked. That is a fact. We went from like before the U S blockade, Venezuela was producing on average like 700,000 barrels per day or so. During the blockade, the sea blockade, it went down to 300000 barrels because basically the nation was unable to export anything. And so it went to from a bottom of 300k barrels per day to almost 1 million barrels per day. So that's been improved. One other thing that the US has been helping with is pushing for a better, a freer currency exchange market.
- [33:53]
C
Right.
- [33:53]
D
So like let the dollar float a little bit more freely and give more access to dollars inside the banking system. Now that has helped in turn with to ease a little bit of the inflation pressures. But still the biggest question in terms of like a, let's say rebuilding the nation and rebuilding all of our political institutions and financial institutions etc is the rule of law. We, we could still argue that the same regime is governing Venezuela, but now there's a referee and it's the U.S. the good thing though, it looks like from this week the US is helping us restate the original Congress that used to work in Venezuela. Basically the Venezuelan regime created like their own congress and that's how they get to approve anything they want. And there seems to be like the creation of a roadmap to transition back to our democratic institutions. And that's going to be the key part to really talk about like a change of things.
- [34:55]
B
Yeah, I appreciate that because again we just don't have a good lens into these things and I don't trust most of our media.
- [35:01]
C
We're off to find a different war now all of a sudden. So yeah, yeah.
- [35:06]
B
People's attention spans or that of peanuts. Alessandro, last question really quickly moving from Venezuela back to Paraguay as you mentioned and as the hash rate heat map here shows 43 exahashes about 4.3% of the global bitcoin market mining market share. Do you see that as being capped? Are we at a ceiling for this and what could disrupt it including AI? I mean are there actual talks within some of the miners there that they're going to try to do AI and HPC builds on these sites like we've seen in the US Is that less feasible in Paraguay? For a number of reasons. What's your sense on where the ball is moving in this area?
- [35:45]
D
Yeah, so I think there's. Paraguay is the hottest country in South America. When the summer hits you could hit. It could hit up to 40 Celsius degrees. So it is, it is insane from a cooling perspective that will force anybody that is trying to do AI operations in Paraguay to consider leak liquid cooling for their facilities. I think that the ones leading this are hive. Specifically I visited their Iguazu site and from what it looks like that will be one of the Primus locations in terms of like and seeing Pair White do some. Anything related to AI data centers. I think the other hard part is connectivity. Right. So the fiber. What's, what's, what's the, what's the state of fiber connections inside of Paraguay? And it's pretty precarious. Okay. So it's almost like you know these, these mining operators in these countries. As Frank said, they're big land owners now. They fully integrated with the market because they build like a local base of employees that makes almost the company really feel like from Paraguay. If you visit their sites in Paraguay, for example, and they have to talk directly to like the, the telecom, so like Tigo etc to pass like specifically fiber lines directly to their site. So these are big, big conversations that necessarily have to involve the state and the state needs to incentivize it. I, I hear that Penguin group is also trying to get into AI in Paraguay. So it could be like a race between Hive and Penguin in that regard. But then these players are immensely integrated into the local market in the sense of like Hive is helping and the electrical grid operator digitalize their dispatch signals because when, whenever they have to curtail. Right now they're being told to control like over a WhatsApp group, for example, everything is analog in the electrical grid in Paraguay and Hive is helping them digitalize that. So these are just like broader and bigger conversations in terms of AI. Potentially a prime market for AI is Brazil, because there's already a connection from the north part of Brazil to the US and Florida. So that could help for like inference loads etc and people trying to consider AI data centers. But again, in terms of Brazil, the big question will be how will the government incentivize the importation of hardware? Because right now, pretty much at a general level, whatever electronic you want to import in Brazil, you're talking about 30 to 100% taxes. Wow.
- [38:31]
B
And yeah, going back to the grid infrastructure, the optic that's going to be laid, I imagine it's kind of hard to cut through the jungle to do something like that. So, I mean, I don't know how that works, but I would imagine it's kind of unwieldy, but we'll be keeping an eye on things as they develop down there. Alessandro Sultan, my man, thank you for joining. This was great. We'll have to get you back on later this year as well. Appreciate the insights and shout out to you and all the Venezuelans out there navigating difficult times, no doubt.
- [39:02]
D
I appreciate you guys and I look forward to hosting you at some point down in Caracas, Venezuela. The time will come.
- [39:09]
C
I love it.
- [39:09]
B
Brother would love it. Hey, appreciate you, Alessandro. Have a great weekend.
- [39:13]
C
Thank you. Love that guy. He is the Sultan, one of the
- [39:20]
B
most interesting folks I've met in Bitcoin in the sense of like very multicultural. I don't want to get too much into his personal background, but the dude speaks Portuguese, Italian, English and Spanish. I mean that's. These are the types of people that you meet in Bitcoin, man. It's one of the things I love about it.
- [39:36]
C
Yeah, love, Alessandro. We are now, man. This is the international episode.
- [39:43]
D
This is the international.
- [39:44]
C
We're just doing the world tour, buddy. I know this. We're gonna go world 2. We're going to Asia now. We're gonna go talk about Xi jinping and Kimmy K3. But before that, a word from our sponsor, Luxor.
- [40:05]
B
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's an automated profitability engine that runs every five minutes and tests your fleet's power settings against live energy and hash rate markets. ERCOT back tests show 10% improved profitability with intelligent mining over binary mining. Commander Pro is a hundred dollars per megawatt or a 25 basis point pool fee adder and you can try it free for 60 days. So if you want to learn more, go to Luxor Tech forward slash commander to get started. All right, Charlie, I guess we start with G and then move on to Kimmy. Or you want to do Kimmy first and then G because it's.
- [40:51]
C
Let's start with Xi first. I've got this CNBC article. So basically last night US time, Xi Jinping prime something or other of China gave a long talk at the China AI Summit and basically outlaid a number of things that the four pillars of AI like the four pillars of how China views the next several years and the next meta in. In AI. I'm actually. Because you had prepped this, I'm gonna, I'm gonna tag the. You to explain the four pillars here.
- [41:37]
B
Yeah, I found this to be extremely fascinating, especially in the context. I mean there's no, there's no accident that this is coming right after Kimmy K3 was launched, I don't think. But the four pillars are as such. The first, openness and win win innovation. Second one, security and human control. Three, civilizational inclusiveness. Four, multilateral governance through the UN. I'll take these really quickly, one by one. For the first one, Xi has called for an open source collaboration and sharing as the foundation of AI development, framing AI as a new engine of world economic growth. It's kind of a counter to the U.S. export controls and chip restrictions. China here trying to position itself as the open alternative to American technological gatekeeping. The second is kind of straightforward. Xi said AI must be a trusted tool for humanity and called for laws, monitoring and emergency response systems. Basically, let's make sure that we actually use this to our advantage and have safety rails in place if needed. Civilizational inclusiveness AI should not, quote, erode or undermine the diversity of world civilizations. End quote. And this plays into some of the other parts of the speech that were aimed at making sure that AI is developed in the Global South. And then four this is a huge one Multilateral governance through the un basically positioning the United nations or pushing them for a consensus based global governance network, which some could read as undercutting the US's attempts to lead this conversation. The most interesting part about this for me Charlie, kind of the I would say the crystallization of these pillars and to actual direct action is what Xi wants to do with China with regards to the Global South. As part of these pillars, China is making three specific commitments over the next five years, 5,000 AI training and seminar opportunities for developing countries International AI application cooperation centers with the Arab League, African Nations, Asian CELAC and Shanghai Cooperations organization and with BRICS and 30 countries will be enabled to use AI or China's AI powered meteorological warning system. Mazu the way I can't help but read this is almost like belt and road coming for AI in the sense that China is going to try to make inroads in these countries and help develop their internal AI infrastructure or their internal AI expertise largely based on the Chinese toolkit that they have available through these open source models.
- [44:14]
C
And on the really in this talk from Xi came on the heels of the Chinese Kimmy K3 rollout which kind of took the not kind of has been taking the Internet and the markets by storm causing what appears to be kind of a repricing of like multiple sectors within a within AI. And why is that? And that is because the model is really really good. Here's the benchmarks. Basically it's really good at coding beating out GPT from OpenAI and Fable on some coding benchmarks. Apparently it's really really good at front end. Depending on what type of linguistic runs you're doing it can be pretty compelling and it's quite cheaper. So if you haven't been following the Frontier labs in the US OpenAI anthropic tend to their models are very expensive and the Chinese models generally are way cheaper to use. And there's a bunch of reasons for this.
- [45:17]
A
I was about to say why is
- [45:18]
C
that Charlie, why is that?
- [45:19]
B
I would also push back because isn't the API for Kimmy pretty well this
- [45:24]
C
is considerably expensive and I actually will ask our next guest about this. You'll probably have some insight on this. But basically we I we have we are rife with takes and actually I don't think we have time to go through all of them. But here's from Gavin baker quote Kimmy K3 may be an important inflection point for AI. Potentially negative for anthropic and OpenAI while being net positive for essentially every other company in the world. It's a very good time to be a token consumer. So if you are a user you probably love to see this because now all the labs are competing against each other. OpenAI and Anthropic are toe to toe with various types of like usage resets. Anthropic continues extending their fable, their fable access. They anticipate releasing more models. August apparently going to be big, rumored to be big for new frontier models. Again we'll ask our next guest about this but at the same time
- [46:27]
D
it
- [46:27]
C
is creating a bit of maybe another deep seek moment where NEO clouds and the price of compute may be being repriced. And this is on the heels of the meta neocloud pivot. So is everything overpriced? Are tokens way cheaper than ever than the market thinks?
- [46:47]
B
It depends on whose hands on the scale.
- [46:49]
A
Right.
- [46:49]
B
I mean I think one of the fears is that similarly to China's domestic manufacturing base, this, the Chinese state could be subsidizing the production of or in the inference costs of these models.
- [47:02]
A
Right.
- [47:02]
B
And if that's the case then OpenAI, Anthropic etc can't really compete with that piggy bank.
- [47:09]
C
And this is you can say subsidation. We can. A lot of people point to like distillation where the American labs do all the legwork.
- [47:17]
B
I think yeah, and, and yeah, I think both, you know, are arguments I've seen levied so far with regards to this and this one's coming from Nick Carter where he says from an ecological perspective Western frontier models are a common pool resource that Chinese open weight models exploit. They are finite because if you exploit them too hard they lose the incentive and resources to train new models and everyone loses. I think that's you know, kind of at the heart of a lot of people's fears here. And, and one of the questions going forward in this debate versus like the open weight models and the closed, closed door ones that were that are being built here in the US is what's the end game like if you exploit them? If these Chinese models train on them to such an extent that they are so good that they end up like killing the host so to speak, then those models will then have to rely on their own Training.
- [48:07]
A
Right.
- [48:07]
B
And but then maybe at that point it doesn't matter because there's so much market share with these models that they can do what they please. A lot of questions with regards to geopolitical positioning on this as well. And you know, I do think that Kimmy K3 dropping shortly before or in the window of OpenAI and specifically anthropic is OpenAI is pushing their to next year, but Anthropic seeking an ipo. The timing is really interesting.
- [48:35]
C
One more thing here which is this tweet from Negligible Capital pointing out, I think a dynamic a lot of people haven't seen as everyone kind of just reacts to the Kimmy announcement and saying oh it's all the game has totally changed. Negligible Capital identifies that Alibaba who owns 36% of Moonshot, which is the company that develops Kimi, dropped 4% overnight. So there might be a little bit more to Kimi than it's just cheaper, better and all the frontier models overpriced. So as they say, Alibama now owns 35% of an AI company that's highly competitive with Anthropic and OpenAI but that company is currently valued around 3% of their values. Either Moonshot is grossly undervalued or Anthropic in AI OpenAI are grossly overvalued. Both things cannot be true. I would like to say there may be more information that we don't know. Most certainly on the heels of that let's get into. We've got our next guest in the wings, Carmen Lee who will bring of compute. Yeah. Who will bring up right after a word from our sponsor Lygos.
- [49:53]
B
Bitcoin's recovering but lending desks that are playing funny games with your Bitcoin could still be under pressure if they have given away the bag and have more obligations than Bitcoin on hand. Do yourself a favor. If you're using a Bitcoin backed lender, don't look any further than Lygos Finance. They are our preferred Bitcoin backed lender. Here at Blockspace they use Bitcoin native smart contracts to make sure that you are always in control of your keys. That's right, with Lygos Finance your keys never leave your self custodied wallet. With Lygos Finance there are no funny games. No rehabothecation, no bridging and no wrapping. Just competitive rates as low as 8.5 to 10% APR and self custody Bitcoin lending. Go to Lygos Finance to learn More. All right, Charlie, let's bring Carmen Lee up here. Big news hitting today. Carmen, welcome to the show. Thank you for joining.
- [50:48]
E
Thank you for having me. I also have a lot of joy just to listen to you guys, you know, talking back and forth about Kimi. Obviously I wake, waking up seven o' clock I was like what's happening?
- [50:58]
A
Right?
- [50:59]
E
Probably nuts. My Twitter went nuts and the people setting my so around two companies. Compute Change is linking data. We're going to talk about compute change today. But slink and data indices went all over the place. People, financial times, everybody start freaking out. So yes, it's pretty, pretty exciting. One of the things I when I was hearing you guys just now in the backstage is, you know, we need to talk instead of looking at models only, which is I'm sure this model, you know, scalability is an issue. Is a huge model who can serve this efficiently. Right. Effectively. It's one question number two is who is accruing like the end benefits. Right. May not be the Chinese Kimmies of the world, maybe the neoclass. Your point token factories. Right. We have to see where the value get created eventually. Right. I think that's another unknowns at this point point. But yeah. So glad to be on the show.
- [51:45]
B
Yeah, I mean lots to talk about. Wait, hang on, Charlie, hang on. I know, I know you're gunning, I know you're gunning. So my host Charlie will have a lot of good questions for you with regards to this news. As you said, things moving so quickly. It seems like every single day something breaks in this industry and then the next day everyone thinks we're so back. But I want to first I want to front load this segment with the news that y' all released today at Compute Exchange for this secondary market for used and refurbished GPUs. Can you give us a quick background on this product and why you all decided to release it?
- [52:17]
E
Yeah. So we have self launched. So computechange is for zero, right. Compute Change is a marketplace for compute resources procurement. So we do longer term contract forward contracts on both GPU and tokens separately. We also do secondary transactions, own GPU servers. You can buy and sell secondhand servers basically. Right. We soft launched the server secondary transactions early this year and we see tremendous demand supply side. We were extremely excited. One of the things that I always want to push back to industry is if you look at any Neocloud so anyone have servers on their balance sheet, they always discounted to zero at year end of year five. The assume worth nothing which you can argue is More conservative estimates, but same time. Is that true? Is that right? We all know people using L40s today hyperscalers still charge you 40 cents per hour for L40s, clearly it's more than 5, 8, whatever years old. There's a value. If you look at any capital assets valuation, how do you do this? Number one, you can discount a future cash flow to back then back to time zero to understand how much money this particular asset can generate for you down the line. Or together with the SPA demand supply curve, which is secondary markets, how much people transact today? For example, the H100 servers average this year, the average server price is to increase by 10 more than 10%. So instead of depreciating, right. It appreciates for 10% this year, right? So imagine like you are American airline and then people tells you hey, your aircraft worth nothing by end of year five. That will completely screw up their balance sheet, right? So similarly we starting saying that hey, the server is worth Nothing. Even the B200, B300 by year five. I don't know, is that right? Right. So based on what we see, I think that definitely with your value and our compute exchange product really proved that there's very active transaction happening for secondary markets. So it's a choice, right? You can either say, hey, I want the latest and greatest. I'm going to put on PO with Dell supermarkets around the world and wait for six, nine months depends the discount you get and then bring that life. Or guess what? I have my power ready, I have my colocation ready. Give me the H200 H100, I can turn on live next week. That's fantastic. Right? So we do believe give people the option and doesn't translate to all the latest workflow can be deployed with the latest chips. Right. There's a long tail of a useful life for the chips. So that's what we start.
- [55:02]
B
Yeah, I'm glad that you kind of mentioned this up front because it leads into my next question. Depreciation has almost become this shibboleth or mantra for the short sellers. They're just constantly banging the drums saying that the house of cards will come down because these GPUs won't be worth anything in four to five years. But as you just mentioned, that's not even necessarily true. I'm curious why you think that the rental rates for some of these and the value for some of these GPUs, even the older ones have been so sticky and what the end users are deploying them for in the secondary market.
- [55:33]
E
So I'M always curious to see the unwritten database, right? Data based sort of forecasting like. Okay, you tell me. Worth nothing. What is worth? Why do you say that? Right? What data points do you like? Do you think you're dumping servers on the street? Just get rid of them. I would do that trade every day with you. You know, give me a 100. Sure. I would take that from you. It's a garbage, right? It's garbage to you. It's $0 in your balance sheet, right? So I don't think that's true. Based on the data we see from secondary transactions from the forward contracts, we see people put out a 1/ hundreds which happening every day today, right. That shows you get the value at least market consensus point of view for this particular asset class. And it's demand supply driven. And what's interesting about servers, right? And you can look at CPUs. CPU will last long time and they're cheaper, whatever, right? But then gpu, well last a period, a good period of time depends on is it going to be useful for the latest and greatest KME model. Probably not, right? Even the B200 can barely support that, right? That's a separate conversation. But the one thing I don't think should people forget is you have the server layer. You have the model layer which can get larger and larger. But you also have the software layer in between the optimization, the quantization. If that layer also improve scale up potentially you can put larger model to a slightly older chips and make it usable. There's a lot of variables along the way to figure out the supply and demand curve for the servers. It's not a static view. Back to your question. Workflow depends. Obvious traditional ML machine learning. And if you do some simple OSS summarization, you can use very old chips. Even 509 probably will work. Or you go something really crazy. The latest and greatest kid me. You probably need B300 for that. It really depends on workflow and a software layer. How that change?
- [57:24]
B
Last question on this front for me and then I'll toss it to Charlie here. You know, as we see the GPU hardware procurement market evolve, I'm curious and I could be wrong about this, but moving away from. Or there are now more options to move away from direct OEM procurement to secondary markets. What does that do to change the market dynamics? Does this in any way, do you think this will impact Nvidia's ability to price it? Is that already happening? How can these secondary markets impact the primary and maybe create an equilibrium for price on these GPUs.
- [58:01]
E
So I will draw comparables with other assets, right? There are plenty of other assets. Even cars, right? You can brand new cars, think about Tesla. Or you can buy from your friend which he has a Carfax for, right? If the car is at certain conditions, you're happy to buy it from him or someone else, right? Or brand new. It's your choice. As clients or customers, we want to have the choice. Again depends on your workflow to you want latest or do you want a second hand. But ideally you should know what's the benchmark is you want a transparency. One thing why compute change is so critical here is we provide transparency. It's like the car. We provide Carfax for servers. Right. It's not. You're buying something for $30,000 and you pray that thing works. You know exactly what you're going to get right from the specs. Gpu, cpu, ram, all the machine server information which is on paper shouldn't change. And the performances, the flops, the memory bandwidth, the throughputs and the thermal degradation. You need that information. You mentioned in the UID linked. So you know, you test against this particular server, you know there's 8 GPU. What the. Did the 8 GPU change at that CPU? What CPU is that right? So you need all the information. Just imagine carfax before secondary transactions. So we provide that in addition to liquidity so you know what you're getting. Right. So that's very critical piece. And back to Nvidia. I think it's great for them to see their assets last longer than people presume lasts. I think in their mind, hey, they know they can satisfy the demand for their GPUs. Rather have their clients shifting to different provider, different design houses move off their ecosystem. They rather see clients buying their refurbished GPUs. Right. Keep the clients within the ecosystem.
- [59:48]
D
Right.
- [59:48]
E
I would think like it's being in the in anyone's interest to do so. And then we do amd secondary transaction too. Right. We just don't. We don't just do one chips.
- [59:57]
C
Okay, let's talk about the news of the day. We scheduled you for this before. I think the Kimmy, that's right. You're like, you seem like the perfect person to talk about it. You put out this tweet this morning. Quote Kimmy definitely surprised the market. It's also not the easiest model to serve at scale. We'll see what happens. The market freaked out over the past 24:48 hours. But you might have a little insight to the actual practical engineering challenges of serving this model. What do you mean with this tweet and how does this relate to the anthropic OpenAI? Kimmy?
- [60:37]
E
Right, so if you think about the spot market, right, so let's say if you're 100% cloud today and now you think you have an option to switch some of them to Kimi, let's say, right, so you want to switch but who is going to support you? Obviously you're not going to use Kimi, the API. You want American based new cloud provider service to you, then rely on how they're going to serve you. Are they ready to scale up that particular model? Obviously Kimi is very unique this way because they're huge, right? It's a, a big model, not like a regular smaller size model which takes time for any neocloud provider hyperscaler to get ready for that. And do they have the compute power to serve that? Let's assume if everybody right in the US want to say everybody can switch. Can they literally support that with the compute power they have? So yes, open source models will put pressure, pricing pressure on the premium labs, but same time compute, the GPU layer is always constrained. Assume notice get built tomorrow, right. Then even though there's increasing demand for anything, we may not be able to satisfy them. As a result, the open source model pricing might not be lower. Right. The assumption saying the open source model price will be lower is you're given going to be elastic GPU supply to support a deployment. That's assumption. We don't know that. Right.
- [62:00]
C
So do you see that the mismatch between open source like huge parameter models and the actual NEO cloud, the physical GPU layer itself, do you see this as potentially creating long term headwinds for open cheaper models? Long term, do you think that perhaps because it takes a long time to refactor and build out entirely new data centers to service these models, is this like a long, is like a structural headwind against Chinese or just larger, cheaper open models being served?
- [62:39]
E
I mean I won't pretend I'm expert in the software optimization quantization layer. So depends, right. I have seen tremendous improvement in that layer. Make me excited about the longevity of GPUs different older chips.
- [62:52]
B
Right.
- [62:52]
E
But again a lot of people working on that, I think that potentially be a critical piece to unlock all those kind of open source models. But yeah, if you even believe, forget about that layer, right? You only have the server like GPU layer and then the model layer, then yeah. So in your mind be like hey why do I ever buy a one hundreds right? You always want the latest and greatest. And the GB is about Rubens because give you the most flexibility right to host those models then those newer servers will have a lot of pricing power.
- [63:26]
C
So zooming even further out, you tweeted this last week and this gets into Xi Jinping in the China AI summit yesterday quote it's interesting to watching both sides converge. The US is debating restrictions on Chinese AI models while China is reportedly considering limiting overseas access to its most advanced open weight models. You say humanity has made this choice before we accepted curtailed development of nuclear technology in pursuit of greater safety. If both US and China are converging on treating frontier AI as a strategic asset, the challenge is finding the balance. Especially in the context of Xi's comments last night and outlaying this plan which we already kind of knew that's what he was going to say would how do you frame this in the context of safety competition and you bring up the the nuclear analogy here.
- [64:28]
E
I always like to see composition. So you can pretty much tell I'm a very pure capital capitalist, right? I really believe in free markets. My trading background I'm creating exchange indexes companies companies on the side I think have choices is always good for markets and competition is good made the best product win, right? So whenever I see two oligopoly arrive the same conclusion that worries me, right? I'd rather they fight and have different point of view, which is good. I mean you want to hear different voices. If they all arrive to the same conclusion, if they agree with each other, there's no room for other voices, right? Imagine if US and China agree on something, which country is going to go up and say we disagree with you? Who has the voices and the power that's in my mind. It's scary, right? You see two strongest players agree a grade and that would translate to either they have bigger enemy which we don't want to see. I don't even want to know what that would be, right? Or I don't know, it's just not going to be good environment. I want to see a vibrant ecosystem from the design chip design houses layer to the model layer. I think that will benefit everybody. So you may have choices, right? And then we have power of negotiation versus non. So when they decide on sounds like similar path for different reasons but nevertheless the same path that would lead to you and me have very limited choices right at the end.
- [66:05]
C
And we've certainly seen that for the end consumer us the competition both between OpenAI Anthropic and then the open Chinese weights to the US Frontier Labs. It's produced a very fun time to be a token consumer. I just keep getting resets left and right for all of my various models and it rules. Carmen, any last thoughts? Compute landscape? What should I look forward to over the course of the summer as GPU prices continue to oscillate?
- [66:42]
E
Great. Yeah. So can be exchange. We do forward contracts and longer term contracts for GPU and also token layer. One misconception is you can only do token prices on demand. It's not true. You can do longer term. You can lock in token prices as your economic impact inputs. Right. Your production inputs. So happy to explore. Feel free to reach out to us. We're very excited about confusion. We love one note two notes deals as much as we love 2000, 3000 nodes. We about provide equal access to the
- [67:08]
B
market forward markets for tokens that we're going to have to have you back on to talk about that because I have so many freaking questions.
- [67:15]
E
It is so fun. It is so fun.
- [67:18]
B
Carmen, hey, thank you so much for joining. Congrats on the news and yeah, we'll have to get you back on again sometime soon. This was, was. This was great.
- [67:25]
E
Thank you so much for having me. Bye.
- [67:27]
C
Thank you for your time. Definitely got to have her back. Talk about foreign markets.
- [67:34]
B
Yeah, I feel like that was the cliffhanger. I mean and you know me, I love talking about the derivatives landscape for all of these things because they're just. The car is being built while people are driving it.
- [67:45]
A
Right.
- [67:46]
C
Yeah. And. And I got to say I've always kind of had a soft spot for tokens. Now I can finally. Now they're finally actually useful. Not exactly how I imagined it. So thank you so much for listening to another week of Block Space Live brought to you by CleanSpark. We do this every weekday at 1pm Eastern featuring quick hits on AI data centers, markets and merging technology. If you're lucky. We'll talk about some bitcoin. Block Space also has other media. The full suite and catalog of everything we produce on our website at Blockspace Media. That's not dot com, that's Blockspace Media. Check it out and leave us a review on our podcast, whatever platform you're listening on. Thank you so much. We'll see you Monday. I'm Charlie. I'm Colin and we'll sign off for the weekend.