In 5 Years, 90% of What You Use AI For Will Run on Your Smartphone | Paolo Ardoino, Tether

AI compute paradigm shifts

0:00If in 2026 people can already run local models on their smartphones to solve only 50% of the use cases of AI in five years, 95% of the population are able to run 90% of the use cases of AI on a smartphone. They're building all of these data centers and it's not at all clear that the compute paradigm that we're currently operating under is going to continue in which case maybe you don't need all those data centers. So if you don't understand how your AI works

0:31is not you becoming more intelligent is someone else becoming more intelligent with your data? The big transformational market is enterprise and government use and those will likely run on data centers because problems that they're solving the speed at which they have to solve them are going to require massive compute. Is that possible that there'll be both?

Tether's digital dollar success

0:56Why don't we start by having you introduce yourself to listeners about how you came to to be involved with Tether and what Tether is and why it's important which a lot of people including we don't quite understand. Look, Tether was a company, is a company born in 2014. Simple idea, digital dollar. There are plenty of digital dollars but only USD, our digital dollar, was able to achieve the holy grail

1:35of distribution and financial inclusion and impact in the world. Over the last 12 years our company built the most used digital dollar in the world with 573 million users growing by 30 plus million users per quarter. I call it the biggest financial inclusion success story in the history of humanity. Sometimes my

2:05blood boils because how is possible that a little company like Tether was able to achieve more for financial inclusion than all NGOs and charities and whatnot for the last 50 years? And we did it in a very simple way. We used new technologies like blockchain to make the dollar accessible. And you asked me before, well, you know, why not just use the regular dollar? Well, the reality is that for

2:38four or five billion people, so more than half of the population in the world, they cannot have access to that regular dollar. They don't have access to basic financial services. The people that are unbanked, the number of people that are unbanked, the word is just enormous. And it's not like they are unbanked because they are bad people. They are unbanked because they are too poor for being of interest of the banking system. They live in countries where their national currency is devaluating so fast against the US dollar.

3:14Think about Argentina. The Argentinian peso lost 94.5% of its value against the US dollar in the last five years. The Turkish lira lost 81% of its value against the US dollar in the last five years. The Venezuelan Bolivar lost 99.8% of its value against the US dollar. It could go on. Yeah. No, no, I understand all that. And you said poor people, but certainly the people who are trading and holding Tether are not,

3:50by and large, not poor people. I mean, you need a level of sophistication and infrastructure. You know, you need a stable internet connection. You need, you know, so, but I have a side and I want to get in an argument about that, but, but, but the, We exactly can. I mean, there are people in Central South America, they have a stable connection. They, like, think about Argentina or Turkish. Turkey is, the digital penetration is incredible.

4:24Even Nigeria. Nigeria has 300 million people or Philippines, huge countries, huge populations. They all have a smartphone. So they all have digital wallets. So the reality actually is that the digital penetration for these new technologies for, and these digital dollars is much lower in Europe and the United States. Why? Because they don't need it. But actually, I truly, of course, if you, if you talk about certain remote villages in Africa, well,

4:57we can talk a lot about it. We do some incredible things to bring connectivity and energy and access to the dollar there. We could spend hours and would be, it's the most exciting thing in the world for me. But the reality is that many of these populations have, you know, the, the, one of the common denominators of the emerging markets is that they have more youth. They have, so the youth has digital means or is, is understands, you know, technology faster, of course, than the older generations.

5:31And they are more prone to change and also have access to the, again, to smartphones. Like the average cost of a smartphone or for Africa is $80. Of course, you cannot play the big, the, the best games in the world, but they have browsing. They have like Opera, that is a very public, is a public company. It's one of the most known browsers in the world that has 300 million users in Africa alone. Right? So we partner with them, by the way. So as one of the main partners, but bottom line is

6:02what we learned with the, with the USDT over the last 12 years is that we can truly disintermediate. So if finance didn't perform well in certain parts of the world, and if, if, if so many hundreds of millions of people were left behind, it was for multiple reasons because there were not enough incentives to solve that problem.

6:33And we proved with Tether that we could solve that problem or help to solve that problem. We are still early, but still we are doing a good job and be very profitable doing so. And we are not taking money from anyone. Actually, well, we are taking money we're earning on the interest rate of the United States. That's it. We are not charging any transaction fee. Sending dollars for the poorest people in the world costs zero in terms of transaction fees. We make the money again on the interest rates of the dollars we keep in the bank account.

7:04So USDT for us, and also for us internally, was the demonstration that the, in the world of finance, there are too many intermediaries.

7:19That they all need to make money and they all need to make money on transaction fees or milking the last mile. And with USDT, we proved the opposite. We proved that we could be very successful, changing the equation and removing all the intermediaries and creating a free product, basically, like just a dollar, simple as that. And we could find a way to be profitable that no one else in the world figured out before us

7:50with the lowest risk possible. Because, you know, again, we hold the treasury bills in a bank account. So simple as that.

Internet and protocol intermediation

7:57But the most important learning point of that is that not only finance is heavily intermediated, is overly intermediated. The entire technology is overly intermediated. The, you know, internet, I feel internet was born to be peer-to-peer, point-to-point. When you connect your computer to the internet or when you connect your smartphone to the internet, you get an IP address. And that IP address is basically your home address and is, you know, well, there are some nuances to it,

8:33but it is kind of unique to you, could be unique to you. That was the original promise of internet. Let's connect people. And suddenly, internet changed. And instead of let's connect people directly, became let's route their connections through data centers. Why? Because also in that case, the intermediaries created the gravitational force so that being intermediaries of people's connections, i.e. people data, they could earn money hosting, holding,

9:09and intermediating people's data. And that is a very huge problem in a growing society for many reasons. Of course, you have data privacy and like all the things we know. But there are some issues that not many think about. If you have a person living in Rome, like imagine like you are a person living in Rome and you're, you know, it's very likely that your family lives in Rome. And you send a message, you know, I send a message. Let's say I live in Rome.

9:40My family is in Rome. My mom lives in Rome. I send a message on WhatsApp to my mom. That message goes from Rome to Frankfurt or Ireland and back to Rome. Is that good? Is that normal? That adds enormous amount of latency. But even so, that is just a message. Every photo, every video, everything that I send goes to Ireland from Rome, Ireland, Rome.

10:10Imagine how much all the governments spent in internet infrastructure in the last 20 years to route packets data that is completely unnecessary when internet was built to be point-to-point and peer-to-peer. So that thing is an unnecessary intermediation that is only justified by the broken incentives of internet. So what we realized at Tether is that we could, but there was already a solution also

10:43when it comes to data sharing, how we could build a protocol over internet that was, that could scale to tens of millions of people. Because of course the classic answer is, oh, but how we can, how we can make internet usable if there are no data centers? Well, there is, you know, in early 2000, a new protocol was born, was called BitTorrent. It was the iteration of all the file sharing protocols and was so great that it could scale

11:18to hundreds of millions of users and exabytes and exabytes and exabytes of data in a perfect way. And so technically, we had already our answer. We could build, we could reuse structures similar to the BitTorrent protocol and to disintermediate servers and data centers in many, many applications. That's also part of the things that we are doing at Tether, happy to elaborate more. But this is basically the spirit that we have in us. So we learn the concept of disintermediation from finance and what we did to the financial world.

11:53And by the way, now everyone talks about stable coins, everyone, every, even banks now are talking about stable coins because they understand the power. And so now that everyone understands the power of stable coins, we looked around and we thought, what else we can disintermediate? Because clearly we are on the right track and that's why we built Hall Punch as a protocol and then QVAC as an AI, the same concept applied to AI.

QVAC and edge computing

12:18I mean, that's fascinating. And maybe I should have you on just to talk about stable coins. But what I really wanted to talk about is your quantum verse automatic computer.

12:32And so that's a research initiative, right?

12:37Can you talk about what that is? And you were alluding to it a minute ago that there is a path down which we can deliver the power of AI without having to go through massive data centers. So, yeah. Can you talk about that? Explain what QVAC, for short, what that is? Sure. So, I'm a big sci-fi fan and just, you know, for the listeners, if they, for any reason,

13:16they would, they want to read tonight an interesting short story. It's 14 pages. I think it's the most beautiful short story I've ever written. It was written by Isaac Asimov, 1956. It's called The Last Question. To me, it's beautiful because it's a story of science, AI, physics, religion, universe, philosophy, all come together in 14 pages. And it's a story of humanity every hundreds of years and dozens of years and millions of years,

13:48always perfecting this automatic computer, make it better, more powerful. And every single time humanity was asking one single question to this computer, to this AI, it was the last question, the most complex, it is the most complex question that you can ask and trying to get an answer for. That question was how entropy may be reversed, so how we can stop the universe from dying.

14:19And to me, that is very intriguing, it's very exciting. But, you know, bringing back that huge complex question to Earth means how we can stop society from dying, how we can stop Earth from dying, how we can make society stable. And society can be made stable going back to and connecting to my previous part is through technology and through connecting society directly without intermediaries. And having, you know, society was built for the last 5,000 years.

14:55In a society that was peer-to-peer, has always been peer-to-peer. People were meeting the streets, were talking, people were using cash or salt or coins to interact peer-to-peer. Only in the last 50 years, information and money became fully intermediated. So we built an entire society over the last 5,000 years that was built with a completely different construct and we changed it completely in the last 50 years. So we don't know how society will grow,

15:29in which direction, how the decisions of intermediation that we made in the last 50 years, how they will affect society in long run. So when I go back to the sci-fi story, I think how we can truly learn from that. And if we want to achieve an AI that is even able to answer the most complex question of the universe, that should be an AI that is part of the fabric of the universe itself. It's almost

16:01like a new element of the periodic table. An AI that is so intelligent, that knows everything, that can scale and be distributed to the four corners of the universe, cannot sit in a data center on earth, cannot belong to one person. Even just an AI, how we can plan to have an AI on Mars if the data center is on earth. The time, the latency between Mars and earth is too long.

16:33And then we add another planet and another planet. Of course, we are years away from that or decades away from that. But still, I think that we should design technology to scale with a scale of humanity and with our ambition in terms of expanding ourselves in other universe and so on. If we don't do that, we are just here for quick gains. We are not here to build a product for the safety and stability of humanity. We are just here to build a product that makes a couple of companies rich. And I became very

17:06bearish with data centers, also for multiple reasons. There are some beautiful stories. There was this guy in Australia, this entrepreneur that was able to find the cure of the cancer of his dog using ChatGPT and AlphaFold. It's a beautiful story. It's a great thing. But the reality is that

17:36between 95 and 99% of the people in the world, they will never do that. They use AI for basic things, for search or to take a photo of a grocery list or the grocery recipe and have at the end of the month, some accounting. They want some translations. They want some education. They want things that a model, an AI model that runs on a smartphone or on a cheap laptop can already do today.

18:06So you have 95%, let's say 90% of the population that ask simple questions. They are not researchers. They have other problems in their day-to-day lives. They need to go from point A to point B. They need to optimize their taxes. But these are very, very simple things that simple models or models can run on a smartphone can do already. A few days ago, Tether launched, you know, there is a big race to medical

18:45health AI models. And one of the most popular ones is the flagship, which is considered state-of-the-art model, was MedGemma from Google, MedGemma 27 billion parameters. Our team was able to produce a 4 billion parameters model that was exceeding the performance and the accuracy of the 27 billion parameter model of MedGemma. So 4 billion model, I mean, 4 billion parameters model means that it can

19:17run on a smartphone, can run on a smartphone of a good smartphone. And we also have 1.7 billion video parameters model that can run on a smartphone that is an average smartphone for Africa. So, and this is today, it's 2026, it's May 2026. So if today, I believe that today with the QVAC, we've created a platform, an open source platform, because open source is very important, is, you know, so that people don't have to trust that they can verify the platform themselves.

19:48If in 2026, people can already run local models on their smartphones to solve only 50% of the use cases of AI, of the use cases, the normal use cases of AI, the use cases of AI that everyone, all the normal people would run. In three years, I'm sure we can bring that percentage to 70%. So in three years, 70% of the use cases of AI can run on laptops and smartphones. And in five years, it will go to 90%. So if in five years, 95% of the population are able to run 90% of the use cases

20:27of AI on a smartphone, then why the hell we are building tens of gigawatts of data centers when, you know, and the subscriptions to these AI models in centralized data centers cannot be justified because then the subscription could cost millions, not hundreds of dollars. And maybe it's still great. I mean, a smartphone will never be able to find the cure. One single smartphone will never be able to find the cure of the cancer of a dog. And that is extremely important. I'm not saying that

21:01this is not important. I think medicine will have huge breakthroughs thanks to AI, but it will become a niche. What will run on centralized data centers on these behemoths will be a niche, will cost billions and billions of dollars, and will be probably subsidized by governments because that is of fundamental importance. But all the normal people, the hundreds of millions of people, the billions of people will use AI as part of their day-to-day lives through their smartphones to maintain privacy

21:35or their own data. But look at Apple. Every year they're releasing the M3, the M4, M5 GPU of their iPhone 16, 17, and now 18. And every year these GPUs can run these LAMA models two times faster than the previous iteration. So again, that's why in five years what we run on the smartphones will be so good that we will forget about paying a subscription to OpenAI.

Enterprise versus consumer compute

22:09Yeah. And that's fascinating. And as I think I mentioned to you, I'm interested in the potential overcapacity of data centers that's being created. But while you're talking about consumer use of AI, the big transformational market is enterprise and government use. And those

22:44will likely run on data centers because the problems that they're solving, the speed at which they have to solve them, uh, are going to require a massive compute. Uh, so is, is that possible that, uh, uh, that, uh, there'll be both, I mean that, that there will be enough demand to take up the, the data center capacity, uh, you know, to run governments, to run, uh, the world economies

23:22and that personal use, yeah, will migrate to, to the edge. I mean, you know, coming from, this is a very interesting question, right? Coming from the

23:33Bitcoin world, you might know that Bitcoin was first, Bitcoin mining was first running on CPUs, then moved to GPUs and then moved to ASICs. AI started from CPUs, now is on GPUs. And last year, the first ASICs of 4AI were born so that right now, if you have a good GPU, you can run Llama 3.2 at 150 tokens per second. The ASICs 4AI were able to run

24:08Llama 3.2 at 17,000 tokens per second, right? So I think there is a world where

24:18even if you are a bank, let's say a big bank, you could buy in five years, 10, 15 powerful ASICs, spend $50,000, $100,000, but you can run an AI cluster that is extremely, extremely powerful through the ASICs. So the ASICs will bring efficiency or energy consumption now down to 98%. So you can run much faster models with much less energy directly on site and with the benefit of

24:53the fact that you can keep your secret sauce for you because you probably, you know, the more the time will pass, the more companies will realize that, you know, someone else is training data, their model on someone else's data, right? So I think over time, of course, now only centralized data center had the capacity to create very cool stuff. And over time, I think that, you know, many banks, they run their own data centers, right? So they, and the more the models

25:24will become good, the more, even on an enterprise, I think that they will have their own dedicated hardware. And so they will be also because think about it, like you don't want every Italian bank, they will, sorry to make all references to Italy, but I'm by, by, by, I'm Italian by, by birth. But will, will all Italian banks run on a US based infrastructure? Right? Probably not. And so will

25:56all the Italian public administration run on a US based infrastructure? Probably not. Right? So, and the same thing as like all French, like, so there is, there is already a big push. If you, if you read one of the coolest new product lines for data centers is, or cloud providers is actually the,

26:21the sovereign cloud means that you go into a country, you install, you build a small data center directly for the country, deploy there your software. And so even for AI providers, they, anyway, they will need to install and, and, and directly on, on premises, certain capacity to run models there. So I don't think we are going to have to see this huge concentration for too, too much or long

26:52time. Yeah. Uh, well, we can talk about the, the over building. I mean, uh, uh, but, but let's talk

BitNet and local fine-tuning

27:02about, uh, the quantum versus automatic computer. Uh, it's built around this BitNet LoRa framework. Uh, can you talk about that? Describe what that is for listeners? A lot of listeners will not know what BitNet is. A lot of listeners will not know what a LoRa is, uh, and, and why that was a foundational breakthrough for, for what you're doing. In order to run on smartphones, we need, of course, to ensure

27:35we, in order to have a platform that is able and capable to scale and evolve on smartphones, you want to do two things. The first one is inference. The process of inference is, you know, you ask something to a model, you get a reply. And so that computation is called inference. The second part that is very important is how the model can learn from you. Imagine this, imagine if you could have a model on your laptop, on your smartphone, that could read privately through all your emails and design and

28:14learn how to respond in the best Craig's way. So you want to do that. You might want to do that, but you might, you also don't want to do that if it entails to send all your emails to someone else to train the model. So we wanted to make sure that if I want to have Paolo's assistant that responds in my own way, you know, making, you know, some, you know, starting the emails in a certain way as I do

28:47and so on. Like you, I needed to be able to do that running directly on my laptop and learning directly from me and maintaining 100% of my privacy. Same with documents, same with, you know, how I start the company documents, my strategy company documents, everything that I do. I want something that is completely customizable and adaptable to me. That process is called fine tuning. And usually the best way, best technique to do fine tuning is LoRa, that is low rank adaptation. So

29:21basically you take an existing model and you adjust only a portion of the weights based and you fine tune them in order to be respectful and slightly adjust those weights so that they could respect and customize based on your own behavior. And so we understood that, of course, inference was already a proven task to do on local devices. But with Qvac, we took the Lama CPP, that is probably the most known

29:56inference engine for open source AI, and we built on top of it and we optimized so much that now can scale to basically every single consumer GPU. And on top of that, the other breakthrough that was done by Microsoft was called BitNet. The problem is that BitNet is a one bit model. So usually models have different levels of quantization. They use either 32 bits or 16 bits or 8 bits for representing the weights.

30:32Microsoft came up with this BitNet. So one bit weight model. The problem there is that it was not suitable to run on consumer devices. So technically it was what would have been the perfect solution to run on consumer devices, but was not suitable because it was too much relying on NVIDIA, high-end GPUs. And so we modified first BitNet to make it adaptable to run on any consumer device, like any consumer GPU, like the Snapdragon GPUs that you find on the

31:10Samsung phones or the Adreno GPUs or the Apple GPUs. Second, we also created in Qvac a common fine-tuning, LoRa fine-tuning framework so that not only for the BitNet models, but any model that we support, and we support hundreds of models, we give to the developer the same framework to fine-tune any model on all the consumer GPUs. So you have your Lama 3.2 model, you have the Medjammer, you have Qwenn, you have all these

31:48models that are supported by our Qvac platform. Now you can run fine-tuning directly on your own device. So I can have my Paolo's assistant, you can have your Craig's assistant, and that will continue to maintain 100% of the privacy. How far along are you in this research? I must say that I'm very impressed by our team. Well, I read them, but they are doing incredibly well. I mean, and the thing is that we

32:24work, you know, the beauty of open source, I come from the huge appreciation for people like Lino Storvalds or Richard Stallman. They are the fathers of the open source and the beauty of working with highest levels of transparency. So when we get out with a claim, it's for everyone up there in open source. Everyone can check the code. Everyone can challenge us. That's, I think, is the most beautiful

32:54thing. I believe, you know, the crypto industry, the Bitcoin industry came up with this motto,

33:02not your keys, not your coins. So if you don't hold the private keys to your Bitcoin, those are not your, really, you're not Bitcoin. Well, I would say not your AI, not your intelligence. So if you don't understand how your AI works, is not you becoming more intelligent, is someone else becoming more intelligent with your data? And so we built, we are very far along. Now QVAC supports OCR, so basically

33:35recognition of images and texting images, supports text-to-speech, speech-to-text, supports standard AI chats, like, you know, what you are used to with ChatGPT, support medical models, support financial models, support so many different models, plus has the ability to also interact with other devices at the same time, supported for, supports delegated inference. So you can, from your smartphone, you can use your laptop to run the heaviest of the tasks, is becoming a very complete product.

34:10And we support image generation, video generation. So there is so much that the team on a weekly basis is rolling out. So I'm very excited. On top of that, we are going to roll out very soon. Our AI assistant will be, again, fully open source for everyone to actually own their own intelligence.

Open source developer ecosystem

34:28Yeah, I mean, that's fascinating. And I was looking at some of the models that you've trained to date. You have this MedSci, is that right? I think there's a four billion parameter

34:47version that can run entirely on smartphones. So when, when these models, how are you going to distribute them or deploy them? First of all, we want the QVAC platform is made so that it's built as a software development kit so that every developer, so I want to empower every developer to integrate these AI tools directly into their existing applications. So if you have, let's say that you

35:19are building, you know, a new version of Uber or you're building a mapping application or you're building a financial application, a training application or whatever you want, or like a cooking application or whatever, you should be able to, with QVAC, you take, you integrate QVAC within your application. It gives you already all the primitives to support text-based AI, like chat mode, you can,

35:50or like translation, transcription, voice recognition, voice modification, everything you want.

36:01We created basically the Lego for AI. I like to describe it in that way. I'm a big fan of Lego products, but it's really the Lego blocks so that whatever application you're building, you can take the Lego block that is needed for you and integrate in your existing application. So that truly, back to the sci-fi story that I like so much, AI can be part of the fabric of the universe. Then in order to showcase how exactly we work and how, you know, we wanted to build our own consumer application, it's called

36:38QVAC AI Assistant, will be released in the next couple of months and will be open source as well. And we'll show how all these different tools and Lego blocks can come together. Almost like, you know, you build, I wasn't making the wrong analogy with the Death Star because I just finished to build that on the Lego side, but so it's not, I don't want to build a Death Star, of course, or AI, but the, you know, you have all the schemas and from Lego from, and they tell you where to put each block, right? So we want to do that to showcase exactly what in our opinion is the best

37:14outcome you can reach, but we give you all the different blocks. So if you want, you can take and build something else and it's all completed up to you. You said the QVAC Assistant will be direct to consumer when it reaches general availability, is that right? And it'll, what, be an app that you can download onto your phone or laptop? Exactly that will be an app that will work on Linux, Windows,

37:44Mac, iOS, Android, name it, and on servers can work everywhere. And yeah, we, and again, will be fully open source. And so that clips the tether to the cloud. I mean, you're now independent of the cloud if you have that compute capability on your laptop. That's fascinating. And so what's your

38:14so you're going to release that? What about the developer uptake and, and has there been

38:23a lot of, uh, third party testing, uh, to see how effective the models are or how competitive they are?

38:34We have been collaborating with, and we are in direct collaborations with, uh, with many, uh, companies that are building on this technology already. Keep in mind that we open source it less than two months ago, and yet there are quite some companies that are building on it and every day more, uh, are, are looking at it. But there are two particularly that are very interesting. You know, the other big part of, uh, let's say the tech revolution will come with robots. And that is where also things can become

39:06scary. I believe that, uh, for multiple reasons, of course, but I think in robots, imagine having like a robot that can only take a decision if they are connected to a cloud and to a centralized data center.

39:24Of course, there is a privacy issue. There is a control issue, many issues, but on top of that, there is a latency issue. If you have like your smart car and it will only break and stop only because you know, the, the image that is seen is analyzed by a data center. Well, that can go wrong in many different ways. So you want, and the same thing with a, with a robot, like if let's say you have a robot that is helping, I don't know, uh, a child, a kid, you want that robot to understand exactly and

39:58react in the microsecond or millisecond cannot wait for packets to go to Ireland and back if you are in Italy, right? So the, even robotics will move, will need to have a GPU or an ESIC in their brain to be able to analyze, analyze information immediately, locally. So I think it's, it's becoming very obvious, at least to us, that is going to be, um, is going to be the case. And so, um, yeah, we have these two

40:32robotics companies that are testing how QVAC is performing in the, the brain of the robot, robots they're building. So we, we are going to hopefully to showcase something very soon. Uh, it's, it's very exciting. So developers are now integrating this into applications and that's the ambition really to, to build a developer community to, uh, to, uh, to, to disseminate this into, uh, into the, uh, personal device world or the edge device world. Um, and, and just on the sort of larger

41:11question of what, because I've talked to other people that are working on smaller models that can operate at the edge. Um, what, what do you think about the, I don't know, what is it this year? 600 billion, 700 billion in capital investment by AI companies, uh, to build data, primarily to build data centers. What, what's going to happen with all that capacity?

Data center financial risks

41:43Uh, I think that, um, so I see a few problems there. Um, it's quite interesting because if you think about it, the majority without naming names, but, um, many, many of these, um, data centers are built by third parties that are basically then sign up contracts for off takes. So it's not that the, so of course the, the, the big AI companies have their own data centers, but the majority of this

42:16expansion in the next years will happen where they go to a third party and they say, oh, I need this data center with this capacity. Can you build it for me? And I will of course sign a lease, but see, you know, someone else will foot the bill for that. So they are reinfencing themselves from the risk of building their own data centers. And also they're reinfencing themselves from the financial risk. If tomorrow they will find that the world will find

42:49out that, um, the data centers are not needed anymore. But on top of that, I believe the other issue is that, um, right now, and this is a general, is an interesting moment in time, right? In 2026, or early 2027, many of these, um, yeah, big AI companies will go public. And it's clear that many of these big AI companies are subsidizing the cost of their subscriptions. So a $200 subscription will cost

43:23from $1,000 to $5,000. That's the reality. Why they do it? Because they are private companies. So they can, they can do it, kind of hiding it. Second, because they can, they need to show growth. So if they have to subsidize, let's say that you are like a big AI company, that's again, not the name names. And your valuation is let's say $100 billion. And let's say that you can subsidize for $5 billion

43:58the cost to bring on the next 10 million users or like, I don't know, 50 million users. And if you do that, your valuation go to $150 billion. So you spend $5 billion, but your valuation went to $50 billion. And you can do that during a repeat, but then the company will become public. Retail will buy the company. And you cannot, I mean, if you are like a public company, it's much harder to subsidize the cost because people will see it through. And so someone else, you know, probably retail will foot the bill on that.

44:34So I'm slightly scared for, for the financial engineering that is happening. Um, I think we'll,

44:44I don't know how it will happen. I, I mean, as Tether, we just can just build an alternative.

44:51Yeah. Um, yeah. Uh, so, um, and when you say you're worried about that, just the, that, that there's going to be this big financial hole that eventually has to be backfilled by governments or, or, uh, could be, and there'll be all of this infrastructure that sits unused. I mean, there's one in Utah that the campus is projected to be twice the size of Manhattan. I'm sure you've heard that, uh,

45:29which is crazy. It's almost funny that, you know, the entire thing. So we are trying to create, right. We are trying to recreate the human brain. We did us. So our human brain consume the same electricity of a potato clock. And we are planning to spend 10 gigawatt to recreate the human brain. Sure. You can say that, of course there is more memory. There is more, much more, but we are doing something wrong.

46:00I think, I think the, our approach to intelligence is quite wrong. We will need, probably we need better research. We need better AI. So imagine like, think about this question, right. So if you really, really want to solve the problem of AI long term. And let's say that you had $1 billion, would you invest $1 billion to hire 2000 AI researchers, or would you spend the entire billion dollar in chips, in AI chips? I would hire researchers. The researchers will yield more over

46:36time. The chips will yield less over time. So our, our, the approach of AI is a financial game in this moment. And something has to go wrong to have a, I think, in a sort of reset and start from, I believe, a better approach to, to research more down to earth, but something that understands that if the brain is so good, because our brain is so cool and good, probably in, if we are trying to recreate it with one, 10

47:06million times the energy, we are, we are wrong. Yeah. Although, you know, it's a little bit like big

Monolithic versus small models

47:15pharma, you know, the actual drugs they produce, uh, don't cost that much, but the, the, the whole research pipeline behind developing those drugs costs a lot of money and something like QVAC. I mean, these are in effect distilled models, right? From foundation models, but you need the foundation model first before you can distill it. The future may be in, in these edge models, but you had to have the,

47:55the massive investment, uh, to get there. Uh, I, I guess the question is going forward, whether you, you know, once you have the foundation, uh, whether then, uh, these smaller models will proliferate and make the foundation models obsolete at some point. You need more to the foundational model. You need the knowledge. And so, but you have the good thing you have internet. And so internet is full of knowledge. So of course, I'm not saying that it is, is zero cost to build a model, even an edge model, but the

48:33reality is, so I think the majority of the problems that, so I think that now there are companies working on trillion parameters models that are basically becoming monolithic models. I believe that monolithic models are not skippable and are not the right answer moving forward. I think that the right answer moving forward are very hundreds or tens of thousands or millions of very small models that are able to interact with each other with highest efficiency. So that you, you don't have to retrain the entire

49:10model. You can just return the, you know, if you have a small, let's say in a very simple way, if you have a model that is expert in physics and another one that is expert in chemics, you want, you can train a little bit more than one in chemics to become better rather than having to go through the entire monolithic, uh, retraining of the model. And of course, you know, you could say, well, if you have the, in the ideal world, an entire, the monolithic model has the highest efficiency, has the highest accuracy and so on, but that is not sustainable and will eventually create my opinion issues. And I think

49:45that the world will converge in having millions of small models rather than one single large model. So that's why the cost can also be reduced. And also you can fine tune, you know, task specific models, um, with the, for, for local experiences. So over time, the ability and efficiency of local GPUs will be so high that you don't need one huge model to solve all the problems of the world. You can have, you can just easily fine tune a a model for what you need in that specific moment in time.

Tether's four business verticals

50:19Yeah. Uh, the, um, uh, how is Tether organized? I mean, because this is a research initiative under Tether data. Is that right? So Tether has four silos. Oh, sorry, please go ahead. Yeah. No, no, that's exactly what I'm asking. Go ahead. So we have four silos or four verticals, more than silos. Uh, we have Tether finance, the stable coins, easily put. Then we have Tether energy. We are building some interesting energy avenues in

50:53Africa. Basically in the most remote villages in Africa, we are building kiosks with solar panels on top and rechargeable butters inside. Um, we have already more than one dozen kiosks and more than one million users of those butters. So it's, it's actually decentralized energy. I really like it. It's probably one of the companies I, one of the subsidiaries that I love the most, um, that we're building. And, um, then we have, uh, the other, the third vertical is telecommunications. So we built

51:30peer to peer communication protocols that can scale to billions of people or billions of machines and billions of AI agents without any data center. Uh, we took the idea from the BitTorrent protocol and we perfected, we built it, we adjusted, we changed it to make it extremely efficient, but also not only good for file sharing, but for any type of communication. And fourth, we have this as vertical, we have this AI platform. So these are, yeah, four verticals. Yeah. And, and, and which, uh, the, I mean, this, the, this sounds promising this, uh, this QVAC, the, the

52:10tethered data vertical is where are you spending most of your time?

52:17I'm, I was obsessed by everything we do. So I, I don't have, so I try, if I do something, I try to do it well. So I dedicate almost like even amount of that, probably energy is the part where I spend less, but AI, well, finance and AI are the, my top two priorities, because I believe that

52:40if we want to have a stable society, we cannot allow 4 billion people that are the ones that don't have access to basic financial services, also to not have access to basic intelligence services. So I think society will have a hard time in the next years or decade, if we increase the gap that is already existing on the financial side, adding intelligence gap.

Decentralized energy in Africa

53:07Uh, this is, this is really interesting. Um, well, let me ask you before we move on, although the, this is really about, uh, the, the QVAC initiative, but on the energy, uh, that's a fascinating idea, decentralized energy. Um, and this is all solar that you're pursuing? Uh, we serve the, so we installed these costs in the central, you know, center corridor of Africa.

53:40Uh, you know, that is, is very dark at night because people, that is where, you know, it's the poorest part of the world population. And so we, you know, there are two, sometimes you talk to someone that says, oh, we can solve Africa's energy problem with some nuclear plants and long distance distribution lines. That is crazy. It will never work. But what we wanted to do is to have an approach where we can build these kiosks. And, uh, so they're like sharks there. We put solar panels on top and a few

54:14thousand rechargeable batteries inside. And so for a couple of dollars per month, you can recharge the battery four times, or you just bring back the battery to the, to the kiosk and you swap it four times. And so we wanted to do that. We wanted to start from the most difficult part of the world. And so with 1000 kiosks, we have more than 1 million users, 1.3 million users now. And this, you know, because these people have smartphones, by the way, these people have like light bulbs, but they try,

54:46they have some weak batteries. They, they, they don't last long and they try to recharge them. Maybe they go to, you know, but it's people like kids cannot, not study at night. They don't have, they don't have, uh, if you don't have stable electricity, it's, it's very hard to be part of a civilization, right? So we, we wanted to do that. And by the way, the company, I, this is not charity is a, is a very well-designed, very lean approach development that brings amazing utility to

55:19these populations and, and have a very reasonable cost. The average salary in those regions is $80 per month. So you can, you can spend two, $3 to have electricity at home. These are, these batteries are 145, but 145 watt batteries, very powerful. So they can run like, of course they cannot run a fridge yet. Right. So this is too expensive, but at least you have light during the night you have, like you can recharge smartphones. And so that, that is the first approach. And we know that we can scale to

55:51100,000 kiosks in the next five, six years will bring energy to 30 million households. So, and, and that, my opinion will help a lot and is, is expensive, but not that expensive compared to many other things that are, that are going on in the world. And we believe that is going to be very important for Tether because it's a, that is our population, right? Servicing these people, um, is, is, uh, you know, it's, is our user base. Yeah. That's, uh, and is, is this, uh, uh, open source design that someone

56:29can build or do you have kits or, and do you market them in the U S there's a lot of places in U S with, that are far from the grid? I mean, we, we never looked honestly at U S at U S or Europe. We started in, in Africa. Um, the, I think we could, I think it would be a good idea to open source the kits. I think it's a great idea, actually. I should bring it back to the team. Um, because, you know, it's the, we, we designed, we designed the batteries. They have, of course, in order to avoid any, any stealing of the batteries, we made a special firmware so that the batteries can only

57:04be recharged in the kiosk. I mean, the team managing that, uh, that company is incredible. It's truly incredible. They, they, they are, I think, achieving something that no one else was able to do. So in, in, again, the poorest part of the world, but I like your idea of, uh, or the suggestion to make it open source, I think is, uh, is great for humanity.

Global operations and travel

57:25Okay. Well, um, Paula, this has really been fascinating. Um, and, and I'm gonna follow this, uh, QVAC, uh, more, more closely now. Um, yeah. And I'd like to have you on again, of, uh, there are a lot of questions that we could pursue. Where are you based? I'm, um, um, based all around the world, uh, mostly in El Salvador, uh, because it's, uh, um,

57:57it's an, I love the place. It's, uh, it's a beautiful place. And also it's close to Central South America where there is a lot of, uh, of our user base. Um, but then I travel, I travel so much. Yeah. And I see outside the window behind you, it looks like a mountain. What's, what's that? You know, now I, I came to, in Europe, in Switzerland, I needed to meet with a family. I still have family in, uh, in, uh, in, in Europe. So, uh, from time to time, I, I come, of course, still very attached to them. Okay. Great. Paolo.