Why Private AI Is The Only Way Forward For The USA (...And What Comes Next!)

Transcript

This involves both hardware security obviously which is what exists today and increasing security on model usage and increasing security over how autonomous assets can coordinate within contested environments. >> How does the cryptography fit into the hardware ultimately like defense is a multitude of things. >> How do you see this growing kind of physical and digital space with regards to AI? Obviously the ZK side, the privacy side, the proving side. [music] Welcome back to AI Supercycle, our premier AI show airing every single week

presented by Near. We cover the ins and outs of decentralized AI, privacy, and the future of this massive technology. Near is the blockchain for AI and the execution of AI native apps. You can check out Near's latest AI product at [music] near.ai. Now sit back, relax, and enjoy the show. All righty, guys. Welcome back to AI Supercycle episode number 16, our weekly premier AI show powered by our friends over at Near Crushing with their Near Intense products. They're also their private near AI chat product. One of my favorite chat GBT chat products uhes product. So, Ismmail, welcome to the show.

>> Andy, Robbie, thank you so much for having me back. Love being here always. first uh first show of the new year with our friend Ishmamail. Uh you were part of the recurring cast of characters back in 2025 and feels like we're we're just getting back to it. >> Oh, I know. It's going to be a big year. We talked about everything from AI to ZK last year and I think the AI space at least is not slowing down. >> Yeah, man. It's definitely not, man. Catch us up. >> Yeah. What's the latest? >> What is the latest in the in this intersection between AI and ZK? I know that you guys are doing some interesting work with ZKML as well as just general uh >> uh defense work which is also interesting. Catch us up here.

>> Yeah. So for everyone who doesn't know, Lrangee is a company that builds cryptography ZK FHE consensus everything that's required to make AI safe across dual use technology, both commercial things like crypto and now defense. A big part of the last six months was us taking our industryleading technology from within crypto and finding dual use applications of that with legacy primes with upand cominging defense companies people like Anderell Shield and continuing to purpose that within crypto for tremendous business value.

>> And so my understanding of Ander I remember I saw the the demo of the headset on Joe Rogan and this guy was able to you could see through walls. It was incredible. I can only imagine what you know was happening in in Venezuela when we took Maduro maybe putting some of that technology to work. How how does the cryptography fit into the hardware ultimately like defense is a multitude of things but where does the cryptography fit into the AI weaponry defense stack? >> Yeah, this is a great question. So to take a step back, we've had a thesis for a while that the cryptography that's being developed and financed in crypto is light years ahead of what's being done in any other sector, right? We're

the only industry that throws billions of dollars into research on consensus and ZK and FH and all this stuff, right? Except every single company in crypto is only purposing this innovation into crypto. We've seen the broad variety of other sectors that the companies that can take something that's built in the private sector and purpose that dual use into supporting American war fighters have been able to have tremendous impact and have been able to drive substantial business val uh shareholder value. And we believe that crypto is no different. there's an opportunity to take things in crypto that have secured AI in some of the most uh adversarial environments, right? Securing billions of dollars of assets across consensus, across CK usage, and across AI agents on chain.

And to purpose that to secure AI used in autonomous weaponry and defense relevant programs. And so this involves both hardware security obviously which is what exists today and increasing security on model usage and increasing security over how autonomous assets can coordinate within contested environments. >> So okay so the zk element is important for for these autonomous environments. Why like like why is this property important for this specific niche? >> Yeah. So assume that you have a defense relevant drone that you put into a a battlefield environment, right? That drone has an onboard model or is communicating back to a command and

control system that's running a model to coordinate what that asset does. The question is, is the right model being used? Has that model been tampered with? Has the connection been tampered with? Has the result that is that that that drone is going to use to make a killchain relevant decision, is that coming from the right model? These are all very very important properties when you use AI in defense relevant programs. Our thesis is that the cryptography that's been developed in crypto has a unique opportunity to secure that. >> Got it. >> This is fascinating. This is something that because this is a combination of DPIN but also ZK and some of the crypto cryptography that we've talked about on the show. And so we haven't really broken into this. Was you know you've

been in the cryptography space you know for a long time building lrangege but when did it dawn on you that okay we should be applying our cryptographic principles to this particular sector of the industry how did how did you know how do you break into this this >> you know I I think some of the work and some of the speeches by you know Alex Karp and technological republic or or some of Pauly's podcasts talk about this idea that you have a generation of companies you know the the the food delivery apps, the future of work things, the payment infrastructure that effectively have taken the best American engineers and researchers and have purposed them to do things that are trivial, effectively small quality of life improvements for, you know, the the

top 10% in in and income are in in the United States. And so you can take that technology and then effectively companies like Palanteer and Anderil and Shield have taken that technology from the valley and have purposed that to secure and to build uh new defense programs to extend US uh you know drone and defense and AI dominance. Now crypto is very much in the same position, right? What has the industry put a billion dollars of R&D into ZK4, right? It's to launch more blockchains. It's to have more private blockchains to sell more tokens effectively, right? Which is fine. People make a lot of money selling tokens, but there's an opportunity to take that technology and the exact same way companies like Palunteer and Anderell and Shield have with their

respective industries and to purpose what has been built and financed in crypto into defense. I think the the analogies are very clear at least to me. But the question is, hey, you know, what team has the capability of doing it? We're a research company. We've been developing zero knowledge systems for the better part of four years. Fortunately, you you know or unfortunately you guys have invited us on here for four years to hear us talk about [laughter] it. >> Um and you know I think I think the frontier of that is obviously continuing to to apply this stuff in crypto to generate obviously substantial token holder value as well as finding opportunities outside of crypto and defense is one that is very very important to us as a business to apply our technology to.

>> Yeah. Yeah. So so we're pretty big on the privacy uh part of AI in this show and I I was mentioning uh near.ai AI that their chat product which effectively is like a very private LLM chat product. When it comes to the privacy aspect of ZK, we're seen a bid on Monero and Zcash recently. Um it's you can't even call it a meta. These things have existed for so long. How do you how do you think about the relevance of the private angle with the work that you guys are doing as it applies to whether it be some of your uh decentralized proof networks, deep proof, ZKML and then how that applies to kind of crypto other industries like talk to us about your personal and

Lrange's perspective on the privacy aspect uh as it relates to the ZK work that you're doing. Yeah, I think you guys are ahead of the curve here. As as as usual, the the ability to apply privacy preserving technology onto AI is one of the largest unsolved commercial problems I think within crypto and outside of crypto. The the reality is there's two questions in privacy I think are important. A, how do you keep the model private to protect the weights, the biases, the architecture, and all the business relevant properties for the person who developed that? And two, how do you allow the the data the prompt that people are using as input to generate some inference result to also be private? Um, and these are two

separate problems, right? There's the business side of how do I make sure my IP is not stolen? And you know, let's say you're in defense, it's how do I make sure it's not stolen by China. If you're in the US, it's how do I make sure my, you know, my LLM that I'm using to underwrite insurance data is not stolen by my competitor. And then the other side is how do I ensure that all of my personal data is private? And this is where things like ZK are so important. Deep proof, our flagship product is able to address both of those things, right? The first is you can run inference over a model that's verifiable while keeping that model private. You can ensure if your user that you're sending your request to, I don't know, Google Gemini and the result you're getting back is coming from the right

commitment to that right model. You also using an ancillary technology like NPC or FHE on top of something like deep proof, you can add privacy on top of the input such that the prompt you're sending is not going to be seen in, you know, plain text effectively by the person running that inference for you, which is great, you know, and it makes me feel nice and warm and fuzzy if it's my own personal conversations with chat GPT. But when you talk about enterprise or business context, it's even more important. Havachi is fast and private per [music] trading secured by ZK with Celestia underneath. Go to habachi.xyz to start trading today. Looking for yield that stands out? Infinify gives you just that. Deposit stable coins and

earn far more than the current savings rate. Infinify is a battle tested, reliable defy platform that is ready to go. Head to infinifi.xyz. Holiday is unified crypto payments. Give your users more wherever they are. go to holiday.xyz to never write a smart contract again and give us a an update on on the state of ZK because for so long this was in R&D phase. It was very academic. We had to do hard math in order to you know get the performance to a point where it became it became useful as a a pragmatic business solution. Where are we at today? What has the performance been like over the last few months and and do we still have more room to go or or is

it at the pace and the performance now where you know if we need to prove every you know instance of where a drone is shooting we can get that proof back you know within within a matter of milliseconds in order for the drone to make a decision or we're able to use it in chat bots at scale where you know where we at in the state of ZK. Yeah. I mean, have you guys ever seen that chart where it shows effectively like the hype on stuff and it gets really high and then it goes all the way down and slowly rises back up again? That's what ZK was right for the the early part of last year and then definitely uh in 2024. The the hype on ZK was so high. It was we're going to solve everything with zero knowledge. Every blockchain, every rollup, everything in the world is going

to be proven in ZK and we're going to throw cash at this problem, right? And the industry, as we always do, got over its skis on it, right? Right? This is what happened with ZK. And you realize, hey, this isn't as fast as we hoped it was. Even with GPU acceleration and custom chips, we're three, five years away from this taking over literally all compute. And we're still on that timeline, right? ZK is inordinately faster than it was a year ago. About an order of magnitude. And it's an order of magnitude faster than it was a year before that. An order of magnitude faster than it was before that. There is now a variety of commercial applications that ZK can be purposed for and is being purposed for. things like defense, things like enterprise and obviously things like you know cheap block space on chain that were infeasible 12 to 24

months ago. Now can we prove every inference from you know uh Opus 4.5? Can we improve everything from Gemini 3? Can we prove everything from uh you know any major commercial model with with ZK right now? Probably not. Will we be able to in 12 to 24 months from now? I believe so. Right. We are approaching that point faster than ever before and I personally think it's going to be one of the largest economic opportunities that crypto has seen when everything is becomes effectively ZKI.

>> Yeah. Yeah, man. I mean, I I think that people are going to want this like like like now like people want this privacy aspect now. They want ZK to work now. Um, has it has the has the progress of the actual technology in terms of the proving uh costs and time has it significantly improved? Where are you at with like your your latest 2026 benchmarks? Yeah. So we've announced that we can prove uh Gemma 3, we can prove GPT architectures, we can we can prove um uh a variety of open commercial models uh with deep proof that are large LLMs. Mostly what we've been running is

kind of low uh 10 figure parameter models through our prover um and we can prove about one token per second right right now. um which you know it it is not as fast as we we think we will be able to in a year. We can assume 10 tokens per second in a year and 100 tokens per second in two years. Um but it's substantially faster than it was before. I agree with you that people want things now and that is why as a business there are use cases that are willing to accept latency and willing to pay a little bit more for verifiability and use cases that aren't. Right? If you are getting medical data and you want to ensure that it's private and verifiable so your patient doesn't, you know, have a life-threatening uh adverse effect, then you're willing to pay for that,

right? An extra, you know, 10 seconds and an extra, you know, $5 or $2 is not really going to to make or break you. But, you know, if you're if you're just, you know, vibe coding on claude, we're not at the point that ZK is the right technology to just prove everything out right there, right? You don't really need verifiability. It's maybe a nice to have but doesn't really matter. And so that's where we are with ZK for at least AI super impactful super relevant for a variety of commercial use cases in high value sector things like defense healthcare finance and now approaching on things that are much more consumerry when that's where I think the big frontier will be within 12 24 months.

>> Yeah. What has been the the perception within the defense space when you bring them ZK you know what is are they cursory aware of it? would have been there. Is there much of a learning curve? Because it it it seems to me I we used to couple months ago we would ask you know some of the crypto companies that were diving deeper into payments they were going and pitching banks and getting adoption. It's like okay how do you guys pitch a bank DeFi? and they would kind of walk us through, you know, where their learning know where their their uh knowledge base was and what what other things, you know, they needed to get through and overcome in order to understand really the value proposition here. What what has been the knowledge

base of of those in defense company and and how much have you had to overcome when you're explaining? >> When we first started talking to defense companies, a big part of my worry was they would not take us seriously, right? you know very fortunately we hired a fantastic lead of our defense go to market um who has a significant amount of experience he's a military background and defense GTM background as well who's been architecting a lot of our defense GTM but broadly I was very concerned that we wouldn't be taken seriously because we're a crypto company but it's a credit to the defense industry and the companies that we've spoken with that they're very willing to embrace new technology right the imperative is to win to build the best thing possible and ensure that they can effectuate uh whatever type of mission outcomes they

need to by maximizing security and maximizing the capabilities that are available to American war fighters. And there is an incredible willingness across both uh the department of war, across DARPA, across government agencies and across uh you know defense companies, both primes and more of the upstarts to actually embrace things that are available. Almost all of them have cryptographers on staff. Are they working on things like ZK, FHE, consensus? No, not most of them are not. But they understand what this stuff is. They they have cryptography backgrounds.

They have, you know, very technical teams. They have research labs. And they're very willing to find ways to try to purpose this to see if there is, you know, the potential to augment and improve on the security properties of their systems. >> Yeah. Yeah. That is very interesting, Rob. >> Well, I I I just I find it fascinating. I was reading about this last night and how you know right now in the current administration we're taking a more top down approach and and Ishmail you're absolutely right like the objective is to win. >> Yes. >> Period. And we've heard that from the highest level of the government. And so they've chosen particular sectors where we want to win. AI, defense, uh crypto is another one.

>> Drone dominance. >> Drone dominance. And so we know that there's going to be significant investment not just in financial investment but also time and energy and resources that are deployed in winning in these particular categories. And so you saw this you allocated you know more time into going into this particular direction. Where do you go from here? How do you think we continue to develop K adoption into defense? And then broadly what are there other categories that you're that you're tackling? you guys all in on defense and if so where does defense go from here as it gets more implemented?

>> Yeah, so this is a fantastic set of questions. We're a dual use company, right? So our ZK machine learning technology deep proof is valuable across both defense and commercial. Um so we have a bunch of commercial partners across enterprise finance across health care across uh enterprise AI and then we have a bunch of folks that we're working with in defense. We've announced that we're part of supplier ecosystems with some of the largest primes and we've we've demoed open uh integrations with a variety of defense SDK environments and there's a bunch of stuff we can't talk about because of NDAs and obviously other types of compliance requirements that are in place in defense but we're very very excited about that category as one of our principal expansion areas over the next year. Um now where we go from here I think that there are other

problems in defense that are adjacent to zero knowledge proofs that also have tremendous commercial opportunities. um things like teaming and swarming right in contested environments is very much in our view a consensus and BFT problem right you have 10,000 drones they don't have an operator link and they need to be able to coordinate a mission outcome when at any point in time you know k of those could be shot down or compromised right if I said hey I have a thousand nodes and I want to make sure that a bunch of them might be taken over or bad actors and I want them to reach an agreement over what to do you tell me consensus right that's how we think of it in crypto but that's an open problem in teaming and swarming for defense. Um, and that's an area that that that we as

a business are are taking some steps into right now. >> Trade on anything, economics, weather, [music] sports, crypto, and more. All on couchy.com 24/7 for the world's biggest prediction market. >> Yeet is built for big moments. The most fun you'll ever have with your crypto. Go to yeet.com to yeet into your next game. >> Not your keys, not your coins. from the team that pioneered cold storage. Treasure has just released its new wallet. Guys, if you are still using hot wallets in 2025, [music] 2026, you are missing the entire point. Secure your coins. Get a treasure at treasure.io.

Love that, man. Um, curious from your perspective what what you think the like blockchain kind of intersection here is at play, right? So, kind of thinking that thinking of this AI crypto category again, it was late 2024. we had this agentic kind of uh Twitter account phase that really ran this year or this past year. It kind of came back but hasn't really came back like when you're thinking about uh in the context of this of this kind of AI ZK proving privacy uh lens as well as these kind of extrapolated categories being uh defense drones etc. Where do you see the most applicable blockchain or like use cases for blockchain? We even had uh a bit of

a robotics thesis conversation earlier with Grant from Blockmates about 2026 and he and he kind of said like look I'm yet to really see how uh crypto rails are going to improve anything in robotics. Um how do you see this growing kind of physical and digital space with regards to AI? Obviously the ZK side, the privacy side, the proving side where do you really see this intersection happening? what is that going to look like? I think a lot of investors really want to allocate to this category.

They've they look at Bint Tensor, maybe they look at Lrange, they look at some other assets. Um what are they what are we looking for from the investor perspective as far as where the real use case is for blockchain, what that kind of transition looks like. Um give us a bit of lens from that perspective. Yeah, I the the taxonomy I like to use here is that blockchains and ancillary technology that's financed within the blockchain space are security technologies, right? It allows you to apply new types of security properties on top of uh you know effectively compute right you're able to provide verifiability with ZK in some aspects of privacy you're able to provide privacy

with FHE you're able to provide resilience and livveness guarantees with consensus and cryptocurrency is a mechanism for payment right it's a mechanism to transact and to store value effectively um and account for value and so the answer to where you're able to find value in the intersection of AI and crypto just comes down to where do you need efficient payments and where do you need security or privacy and so where I see those things being is the application of AI in enterprise contexts where there's not counterparty trust that's where security becomes very valuable right you know there there's no counterparty trust across large forms of AI systems used in defense right that's

the zero trust properties in defense so it's useful Right? There's no counterparty trust between a large hospital and potentially a closed source LLM provider who doesn't want their data to be mined. Great application of blockchain technology. Um, and where do you need payments? Well, for large scale agent coordination, I think that's going to be one of the most natural use cases for it. I I think a little bit of this um also data aggregation and data collection is another fantastic use case, right? How do we allow mass payments for data across, you know, thousands of heterogeneous countries and or hundreds of heterogeneous countries and, you know, tens of thousands or millions of heterogeneous actors?

Another fantastic use case. But I think what it all boils down to is where do you need efficient payments and where do you need uh augmented security and privacy? Ishma last thing from my end is um you know just where do you think we are 2026 if you're looking back at the end of the year any any predictions um in the defense space the ZK space or you know crypto as a as a whole um what are you most looking forward to this year and uh and and yeah any thoughts >> what I'm looking forward to is seeing smart people tackle hard problems the big issue that crypto has had specifically towards the you know early to mid part of last year and definitely in 2024 was very very smart people

tackling things that were not very hard because it was profitable to do things that were easy. That's not compelling and that's not a good use of intellectual capital. We're seeing within uh traditional tech a shift away from building, you know, food delivery apps and Slack clones and SAS rappers into people solving hard problems, drone dominance, hypersonic missiles, nuclear fuel, things like this, you know, rare mineral refining, domestic semiconductor production, right? And we're seeing that same trend in crypto. People who are raising capital to do things that have substantive value, things that generate revenue. You know that there's companies that are doing energy infrastructure.

There's there's deepin doing wireless networks. You know, it's companies like us that are focusing on that intersection of defense, AI, and crypto or crypto infrastructure. This is the type of thing that I think will have substantial growth this next year. The easy money has been flushed out post1010. In my view, what's left is allocators who want to deploy directionally into things that have real fundamentals with the capacity to generate meaningful long-term returns. >> Yeah, man. We're calling this neo finance. This intersection this Yeah. Good. Honestly, the the reception's been bullish, so I'm pretty excited. Um, yeah, we're calling this neo finance.

It's this intersection between legacy finance and DeFi. Um, it it's the it's kind of got multiple enemies here. You know, it's just really all about what's been done in the past, whether it be within the industry or just like kind of legacy finance and the way that that that works. Closing out here for for the AI ZK category. What are some of the teams that you've seen pop up in in in the last six six months really or even like even more recently whether you're partnering with them? Teams that you're excited about in this AI crypto space uh or in in in the ZK and defense space that you're working with that you can talk about.

>> Give us a sense of like where we should be paying attention in this sector. >> Yeah, this is a great question. I would say that to start with the amount of risk financing that's moving into ZK and cryptography is an all-time low, right? There's very few early stage companies we see raising with with ZK anymore. It's almost all the incumbents with the AIX crypto intersection. We have a lot of great partners there who I think are fantastic. I think Cent's an amazing team. I think our friends at Mera are a great team. Um I think uh uh I'm trying to think who else. I think Aether is a great product.

Um I think a lot of these GPU marketplaces are interesting. Open Mind I think is a great play in the robotic side. Um all of this stuff I think is very exciting. Now, in defense, I mean, there's just almost too many companies to name, but I think a lot of the work with Ander and Shield, kind of at the top of the space, is really, really good. I think some of the work with Archer is doing with like EV tall powertrains, like they were using and and Omen drone is also very exciting. Obviously, they're they're a dual use company. They're also doing commercial. Um, but we like them a lot. Um and then I think that there's a lot of work being done in defense for lowcost easy to produce domestic uh domestically produced military equipment. Things like

cheap mass production of drones, the US's answer kind of for for DJI drones. And I think that's an area that I'm very excited about. Right? There's this interesting cost trade-off in defense that I always like to talk about where, you know, if you have to spend a million dollars to shoot down a thousand dollar DJI drone, you really are not going to win long term because you can saturate air defenses very fast. It's an economics problem. And so this is why you need to be able to produce largecale uh domestic armaments that are lowcost things like and barracudas. Um, these are, you know, obviously, uh, I think 200,000 a missile and way way cheaper for shooting down lowcost armaments or

lowcost um, stories. >> Wow. Well, thanks for that, man. Uh, I I I hope the investors that are listening that are interested uh, are able to check those out. Ishmael, congrats on the great work with Lrange. Guys, this has been our AI Supercycle episode 16 powered by our friends at Near, the leading private AI chain in crypto.