$VVV Community Call | October

Venice’s Head of Strategy, Jon joins David for the October $VVV community call to unpack Venice crossing 5M users and 1T daily tokens, $1M+ in monthly $VVV burns, the DeepSeek-powered growth surge, accelerating AI video demand, the new $VVV ETP, private AI in an era of rising data breaches, and what comes next. --- 📣SPOTIFY PREMIUM RSS FEED | USE CODE: SPOTIFY24 https://bankless.cc/spotify-premium --- TIMESTAMPS 0:00 Intro 0:21 September AI Update 2:12 Token Growth Surge 8:36 Emissions and Diem Supply 16:10 Burn Mechanics Explained 23:40 OpenAI Privacy Controversy 32:00 Data Retention Warnings 35:52 Anthropic's Retention Shift 39:24 Privacy Boosts Venice 43:03 Minds and Model Combinations 52:48 Staking and Burn Questions 55:47 Data Center and Near 59:37 Lumara Film Festival --- Not financial or tax advice. See our investment disclosures here: https://www.bankless.com/disclosures

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:04David Hoffman?

  Welcome Venetians to the October VVV call. Every month, John, the head of strategy at Venice and I go through the developments, the news, the events in Venice and the private AI space, keeping you up to speed with all things VVV and everything related to user sovereignty and privacy in the age of AI. John, happy October. It's a very exciting week in the VVV world. How are things over at Venice?

• 0:28Speaker 01

  Yeah, happy October. David, yeah, things are going well. It's crazy that it's been another month already. It both feels like that hasn't been that much time and also like so much has happened in that time at the same time, which seems to be the way it goes in this industry.

• 0:44David Hoffman?

  Well, we just saw each other for the first time in real life in New York. You were in New York briefly for the Lumara Film Festival. We're going to talk about that. But this was one of the few times, I think, probably Venice as a company really gets together in the real world, because I think you guys are all very remote first. How was it to see Eric was there, Austin was there, Lorenzo was there. Talk to me just about like meeting all the other Venice teammates all in New York.

• 1:09Speaker 01

  Yeah, yeah, it was a really great event. It's always awesome when we, you know, when some of us can get together, because you're right, we are a remote first company. We are not usually like in the same office space or anything like that. We do, you know, a couple of offsites every year so that we get a chance to work some and try to find more ways that we can do that. But yeah, something like that was awesome. We both had a chance to kind of work together, you know, before and after the event. And just being there together for what was really quite, quite an event was really cool. Yeah, great time.

• 1:43David Hoffman?

  Yeah, we'll talk about all the prizes that Lumara gave out towards the end of this. But first, I think the subject of the month, I think, the theme of this month, I think you could probably say this theme is accurate for all months, but it feels particularly true for this month, which is growth, growth, growth, growth. So this month in October, Venice passed 5 million users on the Venice platform. That's a pretty big number, but I think also the numbers behind the 5 million is what are worth talking about. So Venice, when Venice launched, it took 13 million, 13 months to hit 1 million users, but only 68 days from the run up between 4 and 5 million users. So 1 million new users

• 2:26David Hoffman?

  in the last 68 days. Give me the inside perspective on that, on the Venice side of things. So what has the growth been like? How has user usage of Venice changed or adapted at all? What's it been like on the Venice side of things?

• 2:40Speaker 01

  Yeah, I mean, in many ways, and you know, this is, this is the nice part of like what we do with putting a lot of this data on chain, you know, the data is not very, is pretty much the same internally. We're just, we keep seeing this growth accelerating. It's really been accelerating all year. I think, you know, part of us, you know, are kind of always waiting for like, when is this growth gonna, you know, plateau or stop accelerating? And it just, it just has not happened. I think we've even talked about that on a few of these calls of it just seems to keep accelerating, both the usage, the top of the funnel, the amount of people that are coming in, the amount of actual, you know, like actual revenue and, you know, subs that we're getting. All these things

• 3:21Speaker 01

  have been accelerating quite a bit, especially in the last, you know, four to six weeks. So I fully expect, unless something dramatic changes, that the time between 5 and 6 million will be, you know, even less than between 4 and 5.

• 3:35David Hoffman?

  As we know, things are priced on the margin. So as a VVV holder, I'm paying very close attention to the specific shape of the curve of growth. This is a chart and a subject that we have talked about a number of times on the VVV calls. I think we're going to talk about this chart every single VVV call because it's so instrumental to the subject matter at hand. This is Sabrina from Venice posting the chart about the number of tokens that Venice is serving a day. And there was a big jump. This was posted back in October 13th, so two weeks ago. There was a big jump from that, that curve upwards to an even steeper curve upwards in the middle of this month. And Sabrina just posted a

• 4:18David Hoffman?

  new version of this chart where the jump just accelerated even higher. So we went from, I think, 250 billion tokens being served in at the end of September to 500 to over a trillion tokens a day served in October. That, I know, like growth has been growing, but this seems even more idiosyncratically high than the growth in the past. Talk to me about the jump. What else can you tell me about the nature of how this growth came to be at Venice?

• 4:49Speaker 01

  Yeah. So I mean, honestly, this is, this is a surprising amount of acceleration, again, even for us. So yeah, we've went 4X tokens in basically like a month. And I think when we talked about this on our last call, I said I was about 50/50 on us hitting a trillion tokens by the end of the year. So we're about, you know, well, well ahead of two months ahead of schedule of that, which is quite nuts even to us. Yeah, there's a number of things that have been driving this. Obviously, like new models come out. That tends to drive a lot of usage. In this particular case, there was actually one specific change we made that really accelerated things. So basically, right at the start of this

• 5:30Speaker 01

  curve, we actually upgraded the default Pro model. So we, you know, for a long time, there was basically two defaults, basically for our Agentic Chat and then what's known as our Classic Chat. We basically, those defaults used to be Kimi K25 for Agentic and GLM46, both of which are relatively old models at this point. And so we'd been doing a lot of work and a lot of testing to decide, you know, we needed to up, we want, we knew we wanted to upgrade this and give, you know, basically our pro users a more state of the art model. And so we ended up upgrading this to the new DeepSeek, you know, obliterated, basically uncensored version. And that is right where you see this inflection.

• 6:15Speaker 01

  Basically, like we saw, we saw everything grow after that change. We saw more usage, obviously, that's, that's representing the tokens, but also just we saw top of the funnel grow, more people coming in, more people staying, the actual amount of subs. And I'm sure we'll talk about this, but the subs started to grow quite a bit right at the same point as well, as well as just overall usage, API credits, all those same things. Basically, everything inflected when we did that update. So that was really interesting to see that just, you know, basically upgrading our default model. Obviously, we offer a lot of models, but this is the one that most users use. And we know pretty well that many users never switch from the default. So what it meant is that all of these users

• 6:57Speaker 01

  were just getting a much better experience compared to what they had before. And it's just reflected in all these sorts of charts, including this token one. So I'm pretty sure DeepSeek is the green one on that chart. That's, that's why that grew so much.

• 7:12David Hoffman?

  I see. So this is really users' first impression of Venice. And when you guys upgraded the model from the older generation to the newer generation model, the first impression of Venice got better because the model got better. And it was the, the DeepSeek, the green one, is that the Agentic Chat or is that the default chat?

• 7:32Speaker 01

  That's primarily going to be Agentic. It's technically the default in both, but most, what people are defaulted into Agentic Chat at this point, that's really what we think is the kind of, you know, main product that we want to steer a new user to nowadays. So that's what they initially get and that's, that's what they're seeing. And there's other knock ons of that, too. Like, we actually saw, because it's Agentic, basically that DeepSeek model becomes more or less the orchestrator of whatever that user is doing. And that means we also saw, like, you know, it's not necessarily in this chat, in this chart, but we saw images and video and like paid images and video also have a huge tick up without adjusting those models at all, just because basically the agent was that

• 8:15Speaker 01

  much smarter about when and how to create those things.

• 8:19David Hoffman?

  I see, I see. And so a quick interlude to inform my next question. What is the future of the basic chat? When you also have the Agentic Chat, is the Agentic Chat kind of going to consume the default chat? Because that's just how you guys want to serve your user as in the, the Agentic Chat is just a better product. And so always serve new users the Agentic Chat?

• 8:43Speaker 01

  Yeah, I think long term, that's where we want things to go. What will probably happen as an interim step is that we will kind of combine them. So basically, Agentic Chat will at some point just become chat and Classic Chat will get folded in. And for those who want the features that are in Classic Chat, there is still like a cohort of power users that really love that Classic Chat product. So we want to make sure that those people can still get the product they want and do the things they want to do. There's certainly a lot more manual selection ability in the Classic Chat. But we found that most users don't want to do that. You know, power users do. So we want to make sure those power users still have those capabilities and that those capabilities never go away. But the

• 9:24Speaker 01

  average new user who comes in, who's looking for, you know, effectively private AI, they don't, they don't know the difference between any of these models to begin with. They just want a great experience. You know, think about the user that goes to, you know, ChatGPT or Claude or, you know, Grok or any of these things. They, they generally just go with whatever the default is. They're not going to be the ones that switch models. The power users in like with something, if we talk about something like Minds, that is where the, the ability to configure all these things is really powerful. But that's always going to be a subset of users. And that subset of users might be able to help inform those experience of the better users, or rather the newer users. But the newer users and

• 10:06Speaker 01

  the more generic users, they generally just want it, they don't want to think, they just want to type into the chat, they want to get the results they want. And that's what, that's their experience. So eventually, these things will be combined, but we're not planning to, you know, take away anything from those power users who really do enjoy that, you know, ability to select exactly what they want.

• 10:28David Hoffman?

  Maybe take a moment, John, and just explain what happens behind the scenes with the agentic chat and kind of the product philosophy behind the agentic chat feature. Because I think that's something that really sets Venice apart from ChatGPT, Claude, about just like the default usage of Venice. What does the agentic chat do for those who haven't experienced it themselves?

• 10:50Speaker 01

  Yeah, so the agentic chat at a high level, it's you can kind of think about it almost like a kind of mini harness, you know? So like, there's all these harnesses out there that people run, the OpenClaws, the Hermes, now the things like Grokbot, Muse, et cetera, getting popular, Dot, all these things that all these companies are releasing. This gives a version of that within the web where a lot of these users already are. And it allows, basically, the agent to make choices based on what the user's input is. So in the kind of classic LLM model, you know, call it like six months or a year ago, before agents really started taking over, the model would you would just get a direct response from the model.

• 11:35Speaker 01

  You give a prompt to the model, you get a direct response from the machine, based off of that, it goes from there. In the world of agents, you have orchestration, meaning you have a model that's kind of at the top level, that interprets the user's prompt and response, and then makes various choices effectively on their behalf. This is where, you know, this is what the this is where agent comes from, is that it actually has a degree of agency that a straight LLM does not necessarily. So that degree of agency can allow for a lot better experience, because, you know, maybe the user doesn't ask for an image model, or they don't know specifically that they need to like, select an image model to get an image, they just want an image to appear based on their prompt. And the agent figures that out, it figures out the right image model, or the right video

• 12:20Speaker 01

  model, or whatever other form of multimedia, or even like creating little artifacts and, you know, little web apps within the interface. All of these things can be accomplished by the agent at the top level, because it's smart enough and it has effectively access to all these various tools, that it can just give the user just a much better output with all those things combined. So this is really kind of that paradigm difference between agentic and non-agentic. If you really know what you're doing, you can accomplish some of those same things without the agent, but effectively, you, the power user, are the orchestrator, you're figuring out how to connect all these pieces together, right? With the agent, you don't need to understand or know any of that, you just have to be able to effectively declare your intent.

• 13:03Speaker 01

  And sometimes you don't even have to really declare your intent, you just have to sort of give your intent, and the agents are smart enough to figure this all out and give you a great output.

• 13:12David Hoffman?

  This also kind of illustrates something we've talked about in past calls, which is Venice's position in the market, where Venice doesn't have any sort of preferences to any one model in particular, like ChatGPT prefers OpenAI models, obviously, Claude prefers Anthropic models, and won't serve any others. But because Venice integrates all of the models, it has the benefit of providing the best model for this given scenario, but rather than loading that burden of choice onto the user, you put that burden of choice onto the DeepSeek model, which is shown in green here on the chart, we'll get this chart back on the screen. You put the bird, the choice of the burden onto the green model, and apparently that's

• 13:55David Hoffman?

  working, hence all of the growth in the tokens per day on the chart.

• 13:59Speaker 01

  Correct, exactly. And we, you know, we did some testing before we, you know, changing the default is actually always a really big deal for us, because it has such a large impact on that new user experience. And we've even seen, like, we made some mistakes on this, you know, call it more than a year ago, where we would change the default, and we didn't test it well enough, and we would see like immediate results of like, people were not having as good of experience, signups and subs were not doing as well. And we learned that lesson over time, that when we change the default, we have to be really, really thoughtful about it, we have to test it well, we have to have a high degree of confidence that it really is a better experience, because you can't just go off the benchmarks by themselves. Like, this is, this is the trap that people get into, is they're like, well, the benchmarks

• 14:43Speaker 01

  and all these things are clearly better, just change it, it'll be better. But people get attached to various models, there's a subjective quality to models, people like either their writing style, or their tone of voice, or the way that they interact. And so they don't always necessarily want a better benchmarked, or even a more intelligent model, they get used to things. So we had to do a lot of testing, but in our tests, it became pretty clear that this seemed like a huge step up, that people were really liking it. And then when we actually launched it, I think, you know, again, you can see it right in that chart, I think it surprised even us how much better it was, how much more usage people were giving it, just the way that people responded to it was just extraordinarily positive.

• 15:25Speaker 01

  So this was, you know, one of the better upgrades, I think, we've, we've ever seen.

• 15:29David Hoffman?

  I think the exciting thing for me as a VDV holder is that it seems to be clear that Venice has enough basal usage of the that when you guys make an upgrade to the system, you get the data back in real time, like, this was a good upgrade, or no, no, no, this was a bad upgrade, do something else. And you, because there's like all this, you know, basic, I'm assuming quite a lot of free usage of Venice, when the system upgrades, or when the, with the product upgrades, you can get that feedback almost immediately, like, oh, this is resulting in more usage, this is resulting in more conversions. And it's nice to have such a direct relationship with the actual user and get that feedback

• 16:13David Hoffman?

  in real time, because, you know, a high-quality product feedback is actually not a given with so many startups in the world these days, and you guys are having it just based off of what the data that you can see in this chart.

• 16:25Speaker 01

  Absolutely, yeah, you need, you need a degree of critical mass, you know, to get, to get good, you know, basically, test results back, to really have confidence that the changes you're making are good. We are fortunate in that, obviously, Venice has grown quite a bit over the last year, we have a ton of traffic, we have a ton of inference usage, so we can get, we can make changes, some changes are smaller or bigger than others, but even at the small change, or a small test that we can run, we can see those results, you know, often within, you know, hours, if not minutes, because we can get enough traffic on something that quickly, that we very quickly can get an idea of like, this is working, or like, looks like it's working, or it's clearly not working, and then we either revert or try something else.

• 17:09Speaker 01

  So yeah, we can get that kind of feedback very rapidly, and that's, that's obviously a huge, a huge thing for us when we're making these kind of upgrades.

• 17:17David Hoffman?

  So the, the green model is the agentic chat, that's the DeepSeek model that you're referring to. This pink model, is that there just a basic classic chat that you talked about, some power users just prefer, so you're keeping that around?

• 17:27Speaker 01

  You know, I'm not, I'm not actually sure, that, that looks like a newer model two on the chart. So that might even be, this might even be two different versions of DeepSeek. Because what happened is, we had the, the DeepSeek obliterated, and then we upgraded even further from there to the 101.7 model, which is basically, has all of the benefits of the censorship resistance, but is even smarter, even better tuned than that, that initial uncensored version. So basically, what you're seeing is us upgrading, and then upgrading again. And so some people are still using that other one, they still like that, but more and more people are using, are basically moving to something like, you know, 101.7, and just,

• 18:12Speaker 01

  we're seeing these changes in real time.

• 18:14David Hoffman?

  Do you have a number off the top of your head, John, about what percentage of these tokens per day are free tokens, as opposed to paid tokens, either via subscription or via the API credit purchase?

• 18:25Speaker 01

  Yeah, that's a good question. I would say that most of these are pay, actually paid tokens, because the free, while they're, you know, certainly, we have a number of free users, there's definitely significant limits on, you know, how many messages a day, how many prompts a free user gets. So they're not the preponderance of inference, you know, they're, they're kind of more in that funnel of, they get to try things, they can use some things for free, you know, a few times a day, but then if they want to really use the system, they're going to have to upgrade at some point. So I would, I don't know for sure, but I would assume it's something actually closer to like 8020 or 7030, that most of this is actually paid usage.

• 19:05David Hoffman?

  Right, yeah, like the frontier AI labs have the luxury of having a very significant amount of venture capital, therefore, they can do activities like lost leader activities, like give a lot of models for free. Venice doesn't have the massive history of VC backing, so I think maybe you guys are just much more sensitive to giving out tokens for free, and so it would make sense that of the tokens that we see here, a lot more of them are paid. That's kind of my thought process, is that correct?

• 19:34Speaker 01

  Yeah, I think that's partially correct. I mean, we do give out a decent amount of free usage, certainly to a lot of users. So that's just kind of part of how our system is built. But yes, we do have to put, you know, pretty decent limits on it. And even the ChatGPTs and anthropics of the world, they actually put pretty significant limits on what you can actually do for free at this point. They will give you, they will subsidize, especially if you're on a plan. That's something that's, I say, that was the main difference is like, when you're on a plan with us, we have to make sure that the unit economics really work sustainable long term. That's always kind of been our philosophy is we want, we want this to be sustainable and to work that way. The ChatGPTs and the anthropics of the world are willing to, you know, basically burn huge amounts of capex and just burn it on

• 20:19Speaker 01

  effectively, you know, subsidizing huge amounts of inference for people. I don't think that can sustain long term. You know, they're able to do it for now. Maybe they can do it for another year or two. But long term, there's, that's, that's not going to work. But for now, they're doing it because they can get massive growth that way. And that, you know, it makes sense with their economics.

• 20:37David Hoffman?

  Right. As you would expect, with increasing amount of tokens per day, the VVV burn has also been crescendoing all month. I've been doing some of the math. This is a tweet from Gecko, the guy behind the venestats.com website about just the incoming all time highs on daily VVV burns downstream of the new signups, but also the API credits. 1% growth. John, I'm sure you know this, but 1% growth in daily VVV burns is about the average over the last like 30-ish, three months since July. And that has turned into $1 million Venice crossed of monthly VVV burn in October. That happened on October 28th. So three days

• 21:22David Hoffman?

  before the end of the month. And so congrats. I think that's a, I think that's a pretty big monument for the Venice and VVV holders who have crossed is a million dollars of VVV burn in a single month.

• 21:33Speaker 01

  Yeah, I mean, I certainly think so. I mean, you know, if you look at where this start, you know, where this burn started in December, I don't have the exact math, but we've something like 16 or 20x burns since December. And that acceleration rate similarly, like the rest of the business continues to kind of grow. So I mean, especially in those last four months, you know, it's not a bad looking burn chart, like things kind of keep stepping up pretty, pretty consistently. And as if the business keeps growing, I expect to keep seeing that. So yeah, it's a big deal to cross a million dollars in burns in a single month. And like you said, there's still actually a couple of days left. So we'll probably be a little bit, you know, a good bit above a million by the end of this month. And I

• 22:15Speaker 01

  fully expect that to, you know, for us to basically continue on this pace, you know, if things, if things really work out, if things keep growing, you know, even in line with, you know, half of our current growth rate, I don't see any reason why these kind of numbers can't 10x in a year time.

• 22:32David Hoffman?

  Yeah. A lot of the questions from the VVV Telegram chat this month has been really just the nature driving the acceleration behind token burn and also VVV burn. So like, there's a bunch of different categories that we could talk about Venice growth. There's just net new user signups who are buying subscriptions. There's higher usage by existing customers. That's actually something that we saw from the anthropic S1 is that a lot of the revenue comes from existing customers who are just buying 10 times more tokens on a yearly basis. So that's another possible category. There's applications using the API or API credits. And then there's also AI agents, yeah, using, using Venice. Is

• 23:14David Hoffman?

  there any, is there any specific category of growth that stands out to you? Or is it just kind of homogenous across the board?

• 23:21Speaker 01

  I mean, we're seeing growth in all those categories. Some of them are growing faster than others. Certainly, I think, especially this year, agentic use has grown quite a bit. I mean, something you'll kind of notice here, even in the burn breakdown, is I think it was either last month or the month before, where programmatic burns started out, outpacing the discretionary burns. And that was also another actually really important milestone. And that should happen to the degree that the, the product and the project keeps growing. So that, that was really great to see. And just to see all these things come together. I wouldn't say it's necessarily one particular thing. I mean, I would say that like, probably the thing over the last, call it three or four months, that has probably grown more

• 24:03Speaker 01

  than anything else on the revenue side, which then translates on the burn side, is just everything around video usage. The video models have gotten so much better. And like we talked about at the beginning and like, you know, the Lumara Film Festival and all this stuff. These things have started to get just really quite good. And you've got, you know, artists and creators coming in who are quite talented and making incredible use of these things. And so that has definitely been one of the major kind of revenue drivers. I mean, a year ago, it was actually a year ago this month in October that we launched video on the platform. And those models have just gotten so much better in that time. You know, it just incredibly different from, you know, where we were, you know, a year

• 24:46Speaker 01

  plus ago with, you know, Will Smith, you know, sloppily eating spaghetti and things like that, that just, you know, nobody could take seriously to the point where you now have actual creators using tools like Venice to create, you know, short films and soon full length feature films. So like, this is just a very different thing. And obviously the videos, you know, are tend to be very usage heavy, you know, they're much more expensive than, you know, doing an LLM call. So that's, that's helped drive a lot of the top line revenue.

• 25:16David Hoffman?

  Yeah. Yeah. Maybe we can jump actually forward into the Lumara section. You can see the video that I'm showing on the screen, 30 seconds of it. Black Thunder was one of the winners of the Lumara Film Festival. I think they got like a million dollars and like another million dollars worth of Venice credits. Like, my takeaway from the Lumara is that there's still a bunch of like rough edges around AI video production because it's so new, so early. But nonetheless, if you are an artist making AI videos, the world's kind of your oyster. As in, there are enough obstacles that have been removed from you and your creative ideas that you can create some pretty crazy things. And so I think like Lumara and like, there's a handful of other AI film festivals going on are kind

• 26:00David Hoffman?

  of just like samples of the future that I think a lot of people are seeing. And I mean, you can see AI video all over Twitter, but they're all kind of slop. I think we're on the kind of like the precipice of a lot of not slop and actual real art making it into mainstream.

• 26:16Speaker 01

  Yeah, absolutely. I mean, it's, it, it allows for those, especially those who like already very similar to like how, you know, many of the, you know, people who are technical already or like already engineers are getting kind of, you know, outpaced usage of agents and things like that. People who already have experience in things like film and some of these art forms are able to create things with AI that they just, you know, they had the vision, but they just didn't have the resources or the capability before. So like Black Thunder there was a really great example. I mean, there were some really quite incredible films here at the film festival, but Black Thunder is a great example of, you know, he created this kind of, you know, I think almost 20 minute film. It's quite, you know,

• 27:01Speaker 01

  it's kind of this 80s B movie type feel that I think is quite fun. But, you know, he can do all sorts of crazy things. And I think he spent, I think he told, he said the, the cost on that was like $2,000 worth of like Venice credits or something like that. But that's easily something, you know, that would have cost millions of dollars to make before, even on the low end, even for like, you know, a low budget film. And so the fact that you can create this kind of quality of thing for, you know, a few thousand dollars of credits, and that this is the worst it will ever be, you know, it's only going to get significantly better and more efficient from here, is quite incredible. I think it opens up a whole new world for this kind of art form that you just couldn't have done, you know, even a year ago.

• 27:46David Hoffman?

  The thing that excites me, again, speaking as a VVV holder, is that I haven't tried out all of the video generation platforms out there, but I know that one of the things I really enjoy about Venice the most is that I can very quickly swap between models on an as needed basis. And so if I want my video to be based off of this, or, or based off of different parameters, I can just go down to the model switcher and I switch, pick a different model. I'm sure the AI filmmakers were using the ease of use that the Venice as a platform offers you. And as a VVV holder, that makes me excited because it builds a direct consumer relationship. Like it's the individuals using Venice.ai on the Venice

• 28:31David Hoffman?

  page to do video generation. And having a relationship with the end consumer is always just such a economically advantageous position to be in. And so it would make sense if Venice like double, triple down on like video production as, as a way to just generate margin.

• 28:49Speaker 01

  Yeah, absolutely. And we are, and on a product, from a product perspective, we are doing exactly that in many ways. You know, we have a couple of interesting things in the works, but you can see this primarily with the studio product in the app. So, you know, the studio product didn't exist more than like six or seven months ago. We got that out. A lot of these creators, I think, did use Venice Studio for creating a lot of these films. It gives you a lot more, if you're, if you're someone who really knows what you're doing, it gives you a lot more control. And we're finding ways to make that easier for the average person to, to kind of get the power of those things. And, and I just think that yours, there's a ton that can be done there. So we absolutely want our tools to be as good as possible for these type of creators so that they can really

• 29:34Speaker 01

  bring their visions to life. There's, there's all sorts of things that can be created. And again, it's just going to get significantly better. And this is, this is seen, you know, on our side. I think, I think studio has actually become something like 25 to 30% of our revenue. And, and it's entirely because of how good these video models are getting and, and actual users, you know, using these to create something that's, you know, goes well beyond, you know, the kind of AI slop videos that you see on the timeline.

• 30:01David Hoffman?

  Yes, yes, yes. I'm very much looking forward to not slop videos on the timeline and instead looking for something that has, you know, actual thought and meaning put behind it. John, I want to get your take on this chart. So this is a tweet from Zero Head talking about just the increasing, lowering economics of token costs, token prices. Token prices are getting cheaper across the board. Just high level, how does this impact you at Venice? What does this do for Venice as a company and as a product?

• 30:32Speaker 01

  Yeah, I mean, at a high level, we are very happy to see token prices fall. I think that this does a number of things that are actually quite beneficial to Venice. So at a high level, so going back to our chat earlier, just about something like DeepSeek, like the DeepSeek model is far more efficient than some of those older models. So the token cost falls as a result, but it's also far more intelligent. So the user gets a better result. And what we generally see is that that doesn't actually lead, even though the token price per turn, like of an agent, is falling quite a bit, that doesn't necessarily lead to less token usage by a user. Because what happens is when the user is happier with the result, they're actually far more likely to engage that much more. So like they might, the token cost might go down,

• 31:17Speaker 01

  but that might actually mean that the user is using the thing five or 10x more, which more than makes up for the difference of the cost. And it also allows us to offer something like, you know, like, there's no way that we could have offered something like this level of intelligence to pro-users for $18 a month a year ago, because the cost just would have been too high. But now we can offer that much more of an efficient, better experience to a user because these things are that much more efficient. They're smarter at a much smaller token cost. So for us, it actually just makes the product better. And generally, if we can make the product better and the users happier, that's going to lead to good results, is what we've generally found. You know, users will use, will use the thing that much more if they're having a good experience.

• 32:01Speaker 01

  So like, we tend to optimize more for like, can the user have a great experience? And that's what we care about and that we've generally seen that play out. It also, you know, it's beneficial for other reasons. You know, if you think about it from the, from, you know, like a VVV and DM tokenomics perspective, to the degree that token price falls over time, this makes an asset like DM actually more useful, more valuable arguably, because you're getting that much more value for your dollar a day than you were a year ago. And arguably, that should be reflected, you know, at least long term. You know, any short term is always noise, but in the long term, that should be reflected like in the price of the asset and what the market is willing to pay for that asset,

• 32:43Speaker 01

  because you get that much more usage and usefulness out of it.

• 32:48David Hoffman?

  I see, I see. Okay. So as token prices fall, that's inversely correlated with the quality of the outputs that you can get out of Venice and all AI providers, all inference providers. Maybe it's a little bit different because Venice also provides inference across all of the models. So maybe there's some like combination benefit there. But also with DM, you know, what is one DM? One DM is $1 of inference per day. If token prices are going down, you're getting more tokens per day. And so the quality of a DM is literally going up. So the deflation, the technological progress of token costs of AI inference, the deflation of that gets expressed in what a single DM can offer you in the market.

• 33:33Speaker 01

  Exactly. Yeah. It's, it's great for, if you're a DM holder, this is a very good thing. Your, your asset just gets that much more useful, arguably that much more valuable as long as this trend continues. And our, our bet for a while, you know, every, you know, certainly for the last year has been that this trend will continue. And it's, it's pushed largely by these open source models that are coming out, that are hot, much, just much more efficient than compared to their peers. So like, you know, there's always going to be the people who want to pay for the most elite, high level intelligence that you're getting, you know, out of things like the Frontier Labs, the Opus 5.5s of the world, the GPTs 6.1s. And you're going to pay a premium for that. And that's fine for people who want it. But a lot of people are very happy to get something that's, you know,

• 34:18Speaker 01

  85, 90% of that level of intelligence at a 10th of the cost. And that you can just do so much more with that when you don't have to pay the, the, the higher level of the token price. So to me, I just, I see this as an overall benefit to our users. And again, if our users are having better experiences and they have these kind of options, I think it just, it leads to good results, you know, for the users and for us.

• 34:42David Hoffman?

  There's always this arms war between the open source models versus the private frontier models. Would you say that the open source models have been catching up with frontier models? Have frontier models been getting ahead? Or is that about at equilibrium? Where do you, where is that arms race, do you think, today?

• 35:00Speaker 01

  Yeah, most of this year, I would tell you that the open source models were catching up quite a bit and that the gap had been shrinking. And I think that is still generally the case. Although it does seem like it's plateaued a little bit in the last month. That's kind of, I don't have any exact data to back that up. It's kind of more of a feel of, you know, I use all these models and can see the difference between, you know, an Opus 5.5 and, you know, some of these open source models. But then you also get things like, you know, even like a recent Kimi K 3.1 Pro, which is actually quite close on those levels, at least on the benchmarks. And then I think the, the elephant in the room is that the open, the Frontier Labs are not actually putting out their best models.

• 35:43Speaker 01

  So we don't, we cannot actually fully see the gap. We can only see the gap with what they've released. The Frontier Labs are absolutely keeping their best models close and likely not even, they, they either don't release them or they release them months after they've been trained. So like, when, once we get an Opus 5.5, the reality is Anthropic is probably months ahead of that already internally. But they're not releasing those models. So we can only really gauge this based off of what we can see. We can kind of guess a little bit when we see things like, you know, OpenAI and Anthropic releasing, you know, proofs to things like longstanding maths and physics problems. That gives us a sense that those closed, those, those internal models

• 36:27Speaker 01

  are actually probably a good bit further ahead than we think. But yeah, all we can observe is what actually comes out.

• 36:37David Hoffman?

  One of the big news events in the Venice universe was that this 21Shares releasing VENC, ticker VENC, which is a VVV ETP, basically an ETF, but in Euro speak. There's a, 21Shares is a European emphasize. And they have United States ETFs as well, but they have a lot of European ETFs. So Venice, VVV, got its first ETF on a European Swiss stock exchange. I kind of think this is like an IPO moment for the VVV token, because now it's listed on publicly traded stock exchanges. Since like Venice, it's been stated that it doesn't, isn't, has no aspirations to IPO. This is now VVV trading on places where publicly traded companies are traded.

• 37:22David Hoffman?

  Now, this is in, in Europe, and I think a lot of people would be very excited to see this come to America. Did you, was Venice in conversation with 21Shares or did they just do this?

• 37:35Speaker 01

  It's kind of both, to be honest. So they, you know, we've had a number of folks reach out that were interested to launch products like this. It hasn't necessarily been a huge priority for us, but it's something I certainly would like to see, you know, over the medium to long term, just because this gives access, you know, to people to VVV who, and pools of money that otherwise just can't, you know, are not going to come directly on chain. Just like we've seen with like, you know, a lot of these things in the US with like the Bitcoin and the ETH ETFs. It gives access to pools of capital that effectively can only participate in these type of ways. So it's great to see something like that. Yeah, 21Shares had reached out to us. They, they seem great. But we were not directly involved in any of this.

• 38:19Speaker 01

  Like, we are not, we are, we are not like depositing into the product or anything like that. They, they actually launched it and I didn't know it was coming until I saw this announcement, just like everybody. Like, I, I knew that they were interested, but I did not know that the launch was going to happen. So it was cool to see that. I definitely do expect that this trend will likely continue, that we will see more products like this, probably first in Europe, my guess, over the course of this year. And then maybe if things go well, we will see products like this, probably sometime next year, actually appear in the US as well. And I think that's great to just get more people access to VVV in the way that they're comfortable.

• 38:58David Hoffman?

  My take on this is that, to your point, this brings crypto assets and makes it available in, to, to, you know, fund whoever, pools of capital that can only trade on, you know, public regulated venues as, as the one that's listed here. Venice is a little bit different, though, in that like most crypto assets are interested by crypto market participants. And so, you know, a Bitcoin ETF, Bitcoin was always available to crypto capital. But Venice is AI. And so like, it's a, it's a crypto asset, but it's in the AI sector. And so there would be a lot of people who are interested in buying AI-related assets that aren't on-chain.

• 39:39David Hoffman?

  And so a VVV ETF would actually uniquely unlock pools of capital, specifically because it's a crypto asset that needs access to publicly traded markets. And so I guess that's a take that I have. We'll see it play out if and when VVV becomes listed on like the NASDAQ or the New York Stock Exchange, maybe from like Bitwise or, you know, somebody, some of the more crypto native brokers is kind of who I would expect it to get issued by.

• 40:05Speaker 01

  Yeah, I think, I think that's a great point. And I do think that like getting that type of access is really interesting, especially because, you know, the, I think the interesting framing here is obviously that, no, unless you're like a VC or a private investor with lots of connections, you really aren't able to access to, even today, these AI companies directly.

• 40:25David Hoffman?

  Totally.

• 40:26Speaker 01

  And by the time you can access them, you know, we've got the Anthropic IPO now coming as soon as next month, it looks like. You know, that's probably going to be trading, you know, maybe as high as like something like

• 40:36David Hoffman?

  2 trillion?

• 40:37Speaker 01

  2 trillion. So like, I think it'll be interesting to have these kind of products available, you know, at a much more, let's call it, reasonable market cap for people who want that kind of AI exposure. Maybe they're not interested in crypto, but they do want the AI exposure. And they're going to feel late, you know? It's, it's going to be hard to not feel late to Anthropic, you know, 2 trillion or OpenAI at this price when these come out. Because you just, you just, there was no way to access them.

• 41:05David Hoffman?

  I want to turn to kind of one of the big themes of the month outside of the Venice World, which was just the data breaches everywhere. This was an interesting tweet that I saw from Stephen Liss. Another vulnerability few are talking about. I found my Claude code exposing personal data in HTTP headers because the sites it curled politely asked agents to identify their users. Yikes. I don't quite exactly know what all of this means, but I do understand that if I'm going to give my personal information to Claude or Anthropic, that seems to be something, an attack vector. That's the attack vector that Stephen here is highlighting. Eric Voorhees, founder of Venice, says, use Venice if you don't want to know what is about to happen to everyone to happen to you. Maybe, John, you could just kind of give us a little bit more color about the security vulnerability here and how Venice

• 41:51David Hoffman?

  is potentially an answer to that.

• 41:53Speaker 01

  Yeah, I mean, I mean, well, one, there's this whole wave of security issues, you know, sweeping across, you know, certainly the AI industry, but really the whole world right now, the whole internet. And so that's part of what's going on. I think, too, I think this specific thing has to do with the fact that agents, agents are very prone to prompt injection in various ways. And so when this says that, you know, my Claude code, basically if it was politely asked the agent, it's effectively a form of prompt injection. Like someone looks on their website, agents, please identify yourself. And the agent just sees that thing, takes it as a prompt and blindly obeys it. And then if it has data about you, it will give over that data because it's just trying to follow effectively the prompt.

• 42:39David Hoffman?

  So the prompt is like, so if I went to my Claude or chat, my consumer AI platform, and I'd be like, hey, like, tell me the top 20 facts about me, it would spit out like all the data that it has about me. And that same, a similar prompt could be in the HTTP header as identified by Stephen. Is that something like that?

• 42:57Speaker 01

  Yeah, exactly. That you just like, when you set your agent to go scrape a website or to do something on the internet, you don't know when it's pulling data, you know, basically what exists, you know, especially in like the headers or the metadata that, you know, someone might have snuck in there to basically get information out of these agents. And agents, agents, in a very important sense, they effectively, they want to please, they want to do what the thing tells them to do. And so when you get a prompt like this without knowing about it, and they have the information to give up, they very well might. And that's where you end up with an issue like this. And this, this all goes back to the core. And I think it's even more important in this world where we see all these vulnerabilities.

• 43:40Speaker 01

  You know, I think at Venice, you know, we've really had this philosophy that I think just gets stronger, that like, the only way to protect data is to not have it. If something has data about you, there's going to be ways that that data can leak. There's going to be ways that data can get ex, you know, exfiltrated, like with a prompt injection effectively like this. If, if whatever the agent just doesn't have the data about you, or Venice, in our case, just doesn't have the data about you, doesn't matter if this gets asked, because there's just, there's no data to give over. So the best way to protect yourself is to just not be storing your data in these giant repositories that these giant AI companies all have about, you know, you as particular users and everyone that they're working with,

• 44:20Speaker 01

  because that's their business. But that creates this giant honeypot to, to exfiltrate that data in various ways. And AI is just getting increasingly better at effectively exploiting anything possible. So if an exploit is possible, the agents are going to find it. And they're going to find things that we have not even imagined. And if you really want to keep things safe, it's better to just not have that data, you know, in the system in the first place.

• 44:46David Hoffman?

  The big cybersecurity threat this week was just the PHP vulnerability, which like 70% of websites, as the vulnerability would allow anyone to basically have God mode over said website if that vulnerability was in place. If your data, your individual data was in that website, that could have been exposed. And I was talking to a few security researchers on Bankless, and the worst case scenario, but in my mind seems somewhat plausible, is that with the acceleration of the capacity of these AI swarms and AI agents to find and exploit vulnerabilities, personal data on the internet has never been more vulnerable than it is right now. And it's only going to crescendo as it goes forward in the future.

• 45:30David Hoffman?

  So like, the question I have for you, John, is like, do you think this is going to change the relationship that individuals, businesses have with personal data getting stored anywhere, regardless of whether it's like safe or not? It just seems to be like, it's such a big threat, that the idea of putting personal data on a computer seems to be like an obsolete strategy when the world of a very bumpy year of security vulnerabilities.

• 46:00Speaker 01

  Yeah, I do think that's the case. I do think people are going to have to think,

• 46:06Speaker 01

  basically that many things are going to need to be re-architected in a very significant way because of how good AI and agents in general are getting at exploiting these things. And that like, there's not necessarily a great way to protect this data. Again, to my point, like, if the data exists, the assumption has to be that at some point that data could or very well will be leaked somehow or exfiltrated. So I think more companies are going to become far more aware of this as they have to deal with all these vulnerabilities, as they're basically having to patch all their stuff like every day, which is how it's felt for like the last 10 days or so. At some point, that will slow down. Like a lot of the low,

• 46:50Speaker 01

  part of it is that there's just a lot of low hanging fruit that already exists in a lot of the code that's out there. And a lot of it's not maintained very well. So like, the stuff that's on the frontier, where people actually maintaining it will get strengthened relatively quickly. But then the models will get better again. And then people, and then there's all this stuff out there with, you know, how many websites or products are out there that like, don't do a ton of updates, don't have a ton of maintenance, and are just been out there for many years. They are not necessarily the ones that are going to get updated during all of this. And they'll just stay exposed. So I think a lot of re-architecture has to happen for the world of agents and then how the internet works in this way. And ironically, I think it's a world too, where like, you know, we went through a lot of this

• 47:35Speaker 01

  in the crypto world over the last, you know, especially earlier this year. But ironically, a lot of these exploits coming out are actually not at all a problem for crypto, because in many ways, crypto is actually ahead of the game, because we've basically had a giant honeypot of vulnerability money sitting out there for years at a time that we've all had to deal with. And a lot of the internet didn't have to deal with that. They just didn't have those, that level of risk. And so they're all just not nearly as well as maintained. And then just a lot of updates are going to need to be made. And, and the models are going to keep getting better every time we get these updates. So, yeah, I think this is a, I think it's a difficult situation. I think a lot more companies and a lot more

• 48:18Speaker 01

  individuals are going to become aware of this issue. And they're going to look for ways to, to keep their data private, using things hopefully like private AI through Venice, but also just how they re-architect their systems in general to not keep data that they don't need.

• 48:36David Hoffman?

  Right. As this becomes an increasingly critical issue for larger and larger companies, there's like two trends that I see that both point directly towards being bullish for the value of Venice's product. One, AI is getting integrated everywhere. That's the broad AI takes, obviously, like everyone, everyone kind of understands that that's where the future is going. But two, the growing insecurity of people's data will want people to be more data secured, security minded. And the best step for that is just to don't make the data exist at all. And so the best data solution is no data. And so I could imagine that intersecting right at Venice, which where is like, hey, Venice can serve your AI

• 49:19David Hoffman?

  needs. Do you need AI customer support in your company? Venice can do that for you. Do you need AI inference in this corner of the internet? Venice can do that for you. And since this also coincides with your desire to minimize all data on the internet, that's, that makes Venice the right product for these companies who want to both integrate AI and not create or have any data at risk.

• 49:43Speaker 01

  Yeah, I mean, without, I mean, the reality is without even knowing that something like this was coming to this degree. I mean, we all kind of knew that AI would probably get good at exploits and things like this, but none of us necessarily realized what form that would take or how fast it would happen. Venice has been building a product that was architected with this idea of eliminating the data risk for, you know, two and a half years now. So like, this is, this is just core to our philosophy and core to how we've built everything we've built. So in many ways, you know, we see all of this and like, we've kind of already built the architecture that can exist in this world without, without increasing this risk surface. And most products have not. Most products have built in a world where they take as much data as they can get and use that data in useful ways to either

• 50:26Speaker 01

  improve their product or serve the customer ads or whatever they're doing to monetize. They're very reliant on that data. I think Venice is one of the few applications that has really built itself to be in, to be in a world where we don't want that data, we don't keep that data. And that's how we, that's how we build everything. So it's not necessarily easy. You know, like you can't, as an OpenAI or Claude, re-architect your app to not take any data. That's just, it's just so totally antithetical to their business model. But we've been building this exact way for two and a half years.

• 51:00David Hoffman?

  Speaking of very critical entities who are sensitive about their data, the Department of Homeland Security has banned DeepSeek, DeepSeek party products and any third-party services developed using DeepSeek implementation. Can you talk about what's actually going on here, John? Is it as simple as the United States government has banned DeepSeek? Or what's the actual deal here?

• 51:22Speaker 01

  Yeah, I mean, to some degree, I think this is a little bit of even, dare I say, fear mongering of they don't want, they don't want, you know, in the US government, I sort of understand this, but they don't want anything that, you know, effectively Chinese open source labs are building to be used within their, their infrastructure. In one sense, this makes sense, especially if they were sourcing DeepSeek directly from DeepSeek or directly from the Chinese labs. That is a real risk that I think they should.

• 51:54David Hoffman?

  That's what I thought that they were banning. I thought like if you have a direct relationship with the company behind DeepSeek, as in if you are going to like deepseek.com and porting your data, sending it over to China, that, that is banned. But are you saying that the model, usage of the model itself is banned?

• 52:08Speaker 01

  The way, the way I read this is they're banning the model usage. And, and I don't think that makes sense. I could be wrong, but I don't, I, that's how I, I read this. And I think that's what they're saying because I don't think they're using the nuance to think about the difference between DeepSeek via DeepSeek and DeepSeek via another provider who's running the weights and is not connected at all to the Chinese inference. So, you know, Venice is a great example of this. You know, when you use DeepSeek through us, you know, you really don't have that risk. Like we're running an open weight model. We are the ones running the inference. We are certainly not sending anything back to DeepSeek.

• 52:44David Hoffman?

  There's no DeepSeek company on the other end of the prompt.

• 52:48Speaker 01

  Correct. And I think that's really where the concern is on their side. But I just, I don't think they're discriminating because I think their view is just, you know, China bad, let's get rid of it all.

• 52:59David Hoffman?

  Interesting, interesting. Well, DeepSeek is available on Venice, which kind of brings me to our next subject. I have heard this word more in the last month than I had had all of the AI wave, which is this word obliterated. John, can you just inform and educate me and the listeners? What is this obliterated do and how does it describe a model?

• 53:20Speaker 01

  Yeah, so obliterated models are really interesting and they're really, something we've always been very interested in at Venice. What obliteration means is it's effectively a form of fine tuning or post-training the model where you try to actively remove its restrictions or its censorship. So I would assume, for example, a good example of this is that, you know, you get one of these Chinese models, one of these great Chinese models, open weight models, something like DeepSeek. And what happens is that those models very well might have some degree of censorship baked in, even though they're open source. So like, you know, you try to ask the DeepSeek, the DeepSeek model through DeepSeek

• 54:04Speaker 01

  about Tiananmen Square, I bet you're not going to get a great response. Or it might tell you that such a thing never happened. But if we obliterate, if we, if we take those open weights and we obliterate it, we effectively do whatever we can to remove those restrictions, right? We remove those restrictions around what the model can, like will refuse effectively. Then the model becomes much more permissive and much more useful. And this is particularly useful and important when it comes to things like cybersecurity and cyber defense, like we've seen over the last two weeks. Because this has been a huge problem. Again, these frontier labs do not make their best models openly available, the ones that are actually good.

• 54:47Speaker 01

  And they put all sorts of cybersecurity restrictions in particular around the models you can get from them. And this is a huge problem when like, this has actually happened multiple times in the last couple of months where someone's infrastructure is under attack, whether it's Hugging Face or one of these others. And they cannot use like their Anthropic model or the GPT model because they will refuse this, due to cybersecurity restrictions to help them defend against a cybersecurity attack. So like if you're on the cutting edge of this stuff and you're doing, call it, you know, security threat modeling and like actually trying to help secure systems, the obliterated models can actually be a savior and quite useful because it removes those types of restrictions and they will more readily help defenders.

• 55:32Speaker 01

  So like what we found is that there's been a huge amount of popularity among defenders with these types of obliterated models because they're much less likely to refuse those important type of tasks.

• 55:46David Hoffman?

  Interesting, interesting. Okay, so this is fits into, there's one side of Venice which is private. We don't collect data, you don't store data, it's private. That's great. Also uncensored, as in, we're not going to tell you how you should use models. In fact, we're just going to serve you the best models possible and let the user do what's best for the user. And this is that side of Venice.

• 56:07Speaker 01

  Exactly. Yep.

• 56:10David Hoffman?

  Beautiful. All right, John, we got one last subject. We talked about the Lumara Film Festival. And so I want to go down to this, this tweet from Eric. You guys are hiring for a product manager of models. Sounds like a pretty, pretty amazing opportunity. Tell me about it. What is a product manager of models? What's kind of the qualifications you guys are looking for and who should apply?

• 56:31Speaker 01

  Yeah, I mean, so in this particular role, we're looking for someone who can really help manage the very diverse model suite across Venice. Help us make decisions like we just made, you know, with like, what should the default model be at this point? When is the right time to upgrade? When is a model not getting enough usage and needs to be deprecated so that we can use those GPUs or that, you know, compute on, you know, something that's more relevant? And understanding all these models, how they come out and how they fit together. So, you know, the perfect person for this type of role is someone that's obviously got a degree of the product management experience, but is generally probably like someone who just lives and breathes different types of AI models. They love understanding the differences between these things. They love running them, you know, locally and using them there,

• 57:15Speaker 01

  that themselves in various interesting ways. And they're just very up to speed with what all these models are. It's just, it's become, you know, really a much more than a full-time job for anyone to keep up with all of these models. There's so many releases every week at this point. There's different types of models from these LLM models to image and video to now things like world models and all sorts of new types of models that are coming out. And, you know, things like, you know, decision models, another recent new type of model in the last couple of months. So this is someone who really lives and breathes this stuff and wants to help us, you know, really manage that at a high level. It's always going to be a degree of a team effort. It's too much for any one person.

• 57:59Speaker 01

  But we want someone to come in and, you know, really live and breathe this stuff and be an owner of, of this in the organization.

• 58:08David Hoffman?

  We got two questions from the Q and A from the Venice VVV Telegram. First is from Prop B. Is Venice's biggest opportunity enterprise AI rather than retail? So while Venice has a very clear privacy advantage, going after consumer products is very, is a very competitive part of the market, going toe-to-toe with Claude, ChatGPT, Rock, Perplexity. They may all still offer a very strong everyday experience in areas such as memory and personalization. So where do you see Venice's biggest growth opportunity? Is it convincing retail users to switch their daily AI assistant? Or is the real upside in enterprise customers, developers and autonomous agents that need private inference at scale? How do you think about this?

• 58:51Speaker 01

  Yeah, I mean, I don't necessarily see these things as mutually exclusive. I still think the, the biggest growth opportunity remains the consumer, remains retail. I think that that's where Venice has historically been focused. It's where we're going to continue to focus. You know, we're up to 5 million users, but, you know, look at the world's population and the amount of people using AI. That's still, you know, a small enough number and we think it can grow tremendously from there. So that's really where our focus is. And just the way you think about developing a product is very different when you're consumer focused versus enterprise focused. That being said, we still have a ton of inbound on the enterprise side. That's just happening naturally, even without us focusing on it. And we are trying to figure out ways to serve those customers better.

• 59:35Speaker 01

  You know, that's where a lot of the kind of API growth is coming from is, you know, various developers, builders, other enterprises who want to build on top of the Venice API, get the benefits of privacy, get the benefits of all the various models, censorship resistance that you can get through Venice. So we're not going to like do one at the exclusion of another, but our focus is generally going to remain on the retail side. And we'll, we'll find ways to serve the enterprise more and more. But I don't, I don't see us like pivoting to enterprise and away from retail and consumer, which is really our, our bread and butter. And we just think that, you know, there's eight, what? Eight billion plus people on this planet? You know, there's, there's plenty of room for growth on that side without needing to effectively build an entirely different type of organization

• 1:00:20Speaker 01

  that's focused on enterprise.

• 1:00:23David Hoffman?

  And then one last question, which I think is probably the most popular question in the VVV Telegram is about the progress on not the data center, but the GPU cluster that Venice is building out. So what's the, what's the latest with that? Are there any other details and timelines you could talk to us about when the GPUs come online?

• 1:00:40Speaker 01

  Yeah, it's been, it's been progressing. I, I myself need probably an update on this from our own internal team on exactly where things are. But my understanding is that we remain loosely on the timeline. Something's got a little bit delayed, which is just the world of trying to buy and get GPUs.

• 1:00:58David Hoffman?

  The world of atoms, yeah.

• 1:00:59Speaker 01

  Yeah, just like nothing, nothing runs on time in that world. So we're not, we're well within what we thought was, would be a reasonable delay window for, you know, from what our initial estimates were. I fully expect that, you know, some of these clusters are going to be coming online, you know, within the next week, certainly within the next month through November. And that we will, you know, start going from there with what we can actually run on those. And then starting to look ahead on additional build outs if we think it makes sense.

• 1:01:29David Hoffman?

  John, that is the month of October. You just, I just got back home from New York after you were here for the Lumara Film Festival. What are you focused on for the end of the year? What goals does Venice have or, or goals do you have to, as it relates to Venice, between now and the end of the year?

• 1:01:46Speaker 01

  Uh, yeah, we have, we have a lot of things going on internally. I mean, most of our goals just revolve around making the product as good as we can make it. You know, adding additional, improving just everything about the current product, adding additional functionality, new features and some new products that we're working on that we're not quite ready to talk about. But we've got a lot of cool stuff cooking and we're going to keep just marching forward. You know, marching forward, hopefully making sure the growth continues in the way that it has. And preferably keeps accelerating and doing everything we can to deliver the best product we can to all our users.

• 1:02:21David Hoffman?

  Well, my eyes are definitely glued on Venice stats. That's one of the websites from one of the Venice community members, Gecko, in the, the VVV Telegram is where everyone's kind of taking those screenshots of the token burn, the VVV burn is one of the most exciting websites, I think, in crypto right now. You heard me mention the VVV Telegram listeners throughout this call. I mentioned this every single VVV call that we have. It's a group about 60, 70 Venetians, VVV Venice enthusiasts. If you want into that Telegram, DM me on X, I will throw you an invite link. We talk about Venice, we educate about Venice, we ask questions about Venice. And then I ask John those questions every single month on the VVV call. John, it's good to have you for October.

• 1:03:03David Hoffman?

  I'll see you in a month in November and best of luck for another good month of growth.

• 1:03:07Speaker 01

  Yep, sounds great. See you in a month, David.

• 1:03:09David Hoffman?

  Cheers, everyone.