Local Ai's Biggest Challenge Yet - Ai News Today

Local AI is under its biggest threat yet, and if you are interested in how you don't become the permanent underclass, the GPU poor and the unprivileged of intelligence, then you need to watch this. I also I'm going to talk about what these changes to potentially your accessibility to Chinese open-source models mean at a micro and macro level for you all the way up to Nvidia. We're going to look at what that could mean for the economy as a whole. I'm also going to talk about whether or not this is all just spontaneously happening or whether there is actually a potential for a mythos class geopolitical model behind the scenes to have been orchestrating everything into a very specific potential pipeline for future outcomes that has resulted in this timeline we are living in right here. I hope you're excited. Let's dive into this insane topic, which is by far the most important video I've done on YouTube to date. We're seeing the development and the further progression of what restrictions around local AI and cloud AI on the frontier and downstream of the frontier may look like this week. Today, there is a report out. I'm going to link that. I'm also going to link a document, which is the basis source for much of what that was written about. We're also going to talk about some caveats of what this is, what this is not, why this is happening now, what this points to, which if you think for a second that Dario is coming up with all of this himself. Have you considered that he's probably running mythos class and internally the teams at Anthropic most likely have access to systems that are vastly vastly superior at geopolitical planning. Yes, that's right. Geopolitical planning and manipulation is absolutely something that theoretically could be happening at the highest levels. What does that mean? Geopolitical AGI achieved? God, that's not what I want. Is that what you want? Sound off below. Also, while you're there, take a chance to hit like, subscribe, and thanks to all of our channel members. It is because of you I can remain actually neutral and not a vendor-sourced sponsored channel, which, sad to say, is the norm on YouTube these days. So, open source and open weight Chinese models may be facing restrictions from the Chinese Ministry of Commerce, which is similar to the US Department of Commerce. The reporting around that document that is linked below, which is in Chinese, and you can actually feed, by the way, to your LLM if it is a like Qwen model and have it just make sure you prompt it to explain in English back to you cuz it'll kind of revert back to Chinese at that point. Uh and it can do an excellent job of understanding it, possibly better than Google Translate will be able to do for you, which I think you should definitely consider. This is China. They do not treat national security lightly, and you would not want to fall on the wrong side of this law. It could have insanely severe penalties for national security violations, which is what the subject of much of what is being talked about here is in this document. The actual restrictions on movement of top-tier assets like Chinese researchers and certain pieces of hardware is also covered in this. That's some of the things that are not covered elsewhere, and that is something you should definitely be paying attention to. There is a definite focus on what the government can and cannot do that is included in this document. This actually very much is a very real threat to open-source model releases that are even close to the frontier. There is definitely talk of restrictions, and there is definitely permissioned access that looks like it could be coming to the Chinese ecosystem. That also could limit the ability of US users to access Chinese frontier cloud models. So, there already is limitations happening around Alibaba in particular. This is the parent company of Qwen. Like I mentioned, that Qwen 3.7 never showed up, even though we had some employees indicate that that was probably going to be happening in a near term. We have still not seen it. And if you understand what this document's talking about, that makes perfect sense because the Qwen 3.7 lineup is actually very, very good for what its offerings are. There is a commercialization for commercial aspect kind of approach to this that is definitely going to be something that both the United States and China are going to be very heavily pushing the buttons on to sway it into their favor more strongly. This is even outside of that though. This is looking at things that are downstream of frontier. So, the access to just something like 3.7 may not be a considered frontier model. Definitely is not considered a frontier model. But even that would be downstream. And we see what could be the pullback of that. What we see and versus what is going on is a fraction of what is actually going on. If you look at the series of events, the way things have played out, and the adaptations to messaging, and what has actually transpired, if you look at the actual what's transpired, there is a huge indication in my opinion that we are not dealing with a human-based intelligence. Does that alarm you? That actually does alarm me. A person in control of that system would be able to craft your future from a legislative standpoint and ensconce in themselves and their company things like regulatory capture that may never be overcome. This is actually what is meant, in my opinion, when we talk about the underclass. Those who do not have access to sway the government in their favor so that they can maintain access to things that are provided free and open source. Whereas, unfortunately, the largest companies out there have unlimited unfettered access and probably, in many instances, the only access to your representatives. That should piss you off a lot. We need better representatives. Like, across the board we need better representatives. The impacts of this, however, to you are going to best be stated as looking at this from impact to access to technology, which already this is you already see you don't have Quinn 3.7. So, there's already accesses that are being for either commercial reasons or for geopolitical strategy posturing reasons pulled back and Queen 3.7 is a perfect example of this. Will there be a GLM-6 that's released to the public is a is a great question. And it may not actually end up occurring. That right there is creating a stratification. You are then also further stratified by are you a Chinese citizen if you wanted to access that it has been one of the talks in this paper. So definitely being able to access those may not even be possible as a US citizen. So if there is a divergence and they do pull ahead, this is the stratification of how we could see the race to AGI really really taking that next leap and stride. And one of the things that I have to talk about is the evolution of the conversation around local AI and the stagnant nature of it. And really there's too much talk about product and there's not enough talk about use case enough talk about political implications enough talk about what this means to you if you don't have access to these things personally in your house to run like you want and especially there's huge costs associated with the cloud. Those costs are going to skyrocket. If you are a user of the cloud, your costs are probably going to go up dramatically for access to the things that are fable class for instance. That is something you should be paying attention to. I mean they're already expensive. How much more expensive can they get and you still maintain access? Mm. Mm. It's a very [clears throat] hard question to answer. So there is you know a lot of local AI hardware talk this channel does. So I'm going to really dive deep on this segment here because this probably has the most impact for you my viewers. Share this video out there also by the way. You want to support the channel, share the video out there. Hit subscribe. This channel is wanting to get to 100,000 and trust me there are a lot of people a lot of very very powerful people who do not want this channel to get to 100,000. So hit the freaking subscribe button right now and make sure your friends do also. If there is a pullback, if this does cut off access and releases to open source models that are coming from China, the impact to local AI hardware is going to be it's going to be like very, very big. So, we're going to explore that here. So, I'm going to break this down into two classifications of systems, which is how I evaluate these. We have your AIOs and then we have your DIYs. Your DIYs versus AIOs little box, maybe a DGX Spark, maybe an Apple Studio Mini kind of thing, maybe a Strix Halo. A lot of those things are not good products outside of a local AI use case. So, that actually really significantly impacts the need for those to even exist if there is not a local AI use case. I could not imagine paying $4,000 for a DGX Spark, 5,000, whatever, for a DGX Spark for a use case that was not like a almost 100% local AI focused. I mean, Framework tapped into the Strix Halo at kind of a really great price point and like $2,000 was not a bad deal for those at one point in time. I don't know what it is now, but I know it's not that anymore. Like 128 GB of RAM is not something that most people need outside of AI use cases. Like you you never saw and as a long time home laber, people that really were like, I need to just get like 512 GB unless there was a reason. 128 GB is you need a reason territory. 64 GB, honestly, you can probably handle almost everything under the sun that any home laber could ever throw at it. But home lab as a niche is already dying. Like people are the cloud is winning. Like let me say that also. I don't think I say that enough. Home lab is dying. Local AI is flourishing at the moment, but the cloud is the like ever-present siphon and like it wins for ease of use. And like you end up compromising because of things like privacy. It is easy to make that trade-off. And I cannot say that it's easy to self-host all of your stuff. It is insanely difficult. But definitely, there are rewards to doing it. Your knowledge actually increases substantially of the systems you're hosting when you do that. That is one of the main reasons I urge people that are right now considering learning and upskilling themselves, DIY systems are a really great way. So, AIO systems could be under severe product market fit viability if this happens. They like I just want to be clear. Uh with that Chinese OSS models, open weight models, whatever you want to call them, uh the local AI scene withers. It dies eventually. Uh it you'll have people like me who are like content with Quinn 3.6. Do you remember the video I did recently on Quinn 3.6 where I said I'm basically married to it, I'm happy with it, I'm fine if they don't come out with 3.7. Do you think I said that for no reason? Look back at this channel's history at when I talked about things that were about to happen. I'm not saying I'm from the future, but there's an indication that there might be a part of my brain that gets receptions from the future because I'm pretty good at seeing what's coming down the pipeline. And this is unfortunately probably what is going to come down the pipeline. I'm going to talk about steps you can take to protect yourself and what you can do now today at the end of this. So, make sure you stay tuned for that. And also, make sure you hit the hype. You know how big that hype button is for small creators like me. Hit the hype button on your app right now. That actually gives me a massive boost, and I really, really do need that. Hit that hype. Hit that like. Share this out there. Share this out there is the biggest thing still. >> [snorts] >> If there is this pullback, AIOs are hit tremendously hard because there's not necessarily the next best great thing uh coming out. Like most people that bought an AIO for $4 to $5,000 or a cluster of AIOs for like $20,000 were thinking to themselves, "Woah, I'm going to be able to like use this into the future to be able to do a lot of stuff." Like, you didn't pay that price for it's just like locked in time right here in 2026 when you know, 2027's right around the corner and there's always going to be better. Um but that better may not be accessible to you or me or us. And that is because, man, so many things have gone wrong in especially the open way open source development realms. It's classic, though. It's classic. It's very very predictable, very classic. But definitely, that does dry up the viability of those as a really good product market fit. Nvidia had local gaming. Like, they they had that. Nvidia has, of course, probably the biggest share of local AI users. Having specific boxes for local AI use cases to compete against Apple was like a strategic decision that was a very good decision for them to make at the time. And seeing the evolution of those products into actually more and more capable AI systems is something that I would expect in the next next generation hardware. I don't think they're going to be capped like they kind of are right now. A little more cooling, a lot of bit more processing power and system bandwidth, and you have some actually probably really kick-ass systems that can actually use that 128 GB at its full like potential. The AIO used market could see really good prices. If this happens, I expect there will be a sell-off of people that have AIO machines because they'll be like, "Well, what the F?" I That's like going to happen. If this happens, there will be a sell-off of people that have DGX Sparks. Those DGX Sparks will not command the premium that they have been paid for as they will be essentially a product with a limited lifespan as far as future viability, potentially significantly. If you see the future shaping up like this, it it it has to be something that I share with you. And this is just what I see. It's not to say I can predict the future 100%, but boy, in this case, I feel like I am pretty on point on this. So, who could step into the void of there's no more Chinese OSS models. Cohere, uh you know, maybe I I don't know. Um news research, I think they made a model. Um there might be a few others that have been made out there. Most of them are tunes though, so they kind of build upon that base. If you look at Nvidia, they're probably the best positioned player and there has been recent comments that made me think that they are definitely going to be heading in this direction because they probably have seen these headwinds coming. If you look at the time and talent that it takes, it's not something you can just start today and finish tomorrow. So, training open-source models that are really good takes time from scratch. Uh finding the talent for it. They hopefully are ramping up planning for this because I do expect this to be coming. If Jensen is watching this, by the way, kudos on making some killer hardware, but we definitely need some killer models to make that hardware viable. The AIO use cases, however, outside local AI, like are you paying 3 to 5,000 dollars for one of these AIO machines for gaming? You shouldn't. Uh you might be, but you shouldn't. Are you doing it for home lab reasons? You shouldn't. You could be, but you definitely shouldn't. So, it's very possible that we see those excellent deals on used AIO systems if this happens. So, keep an eye out for that. I know I I love buying used stuff because there's such a reduction in price when you buy right off that like oh, the cutting edge, like huge reduction in price. If you look at just the 7995WX, that's like a $10,000 processor when it was released. If you look at it now, it's like 5, 6,000 dollars on used markets. Do you like huge reduction in things whenever that 9,000 series came out right when that happened. So, the same will probably happen when we see Sparks and DGX kind of systems and possibly strict halo. Uh maybe some of the Macs the Macs have a user base kind of built in. The insanely large VRAM footprint Macs, I don't think have a user base built in and I'm not sure what the use case would be outside of local AI for those. Uh Uh but definitely, there will be pressure in the AIOs. Does this mean you're safe if you're in DIY systems? No, it doesn't. And why? And this one I can draw a pretty clear parallel to what happened one time before during the Ethereum mining craze when a 3090 could be minting you like 20 bucks a day, like each one. And the ramifications of what happened when that went away, Ethereum went completely to proof of stake, these 3090s over there. Uh those 3090s definitely mined Ethereum in the past. And like they're still very, very viable hardware. They didn't need to be cleaned up, like seriously cleaned up and repasted and repadded. But after that, they do their job just fine. So, let's take a brief look at what happened the GPUs at their maximum dip price, which was about $400, which happened for a brief tier period of time, about a month after Ethereum was over and people finally accepted it and came to the realization that it was over. It took those people a very long time. There was a very brief window of time where they were 400. That lasted maybe a month. It really stabilized around 600 pretty rapidly after that because gaming demand for 3090s has still always been there, and 3090s are just kind of a killer GPU no matter what. They really made it an awesome GPU in the Ampere lineup. But definitely, when you look at what 800 leading up to 1,000, which is what we saw as local AI systems became more and more capable as there was more and more demand for them. Now, 1,100 might be the used price range for some of them. Do you think like 1,100 is the highest 3090s have been? No. When they were producing $20 a day each or $25 a day each, like that was the highest price that they've ever been, and they were in the several thousands of dollar range each back then. And that dropped to 400. There is potential ramifications for DIY systems also. However, there is always somebody to buy it. There's some gamer that needs a new GPU. Uh whereas with the AIO systems, that becomes substantially more risky. Like, is a gamer going to buy your DGX Spark for $4,000? You answer that question for me. What does this mean for enterprise systems? Things like B300s. So, in my opinion, this is bad news for Nvidia because there is a need for an open ecosystem that is flourishing. Not just stagnant, not just contracting, of course, but flourishing, actually flourishing to supply the increase in demand that they need for their offerings. Again, we're not just dealing with uh gaming GPUs. We're not just dealing with enterprise systems like B300s. We're now dealing with custom-built AIOs like the DGX Spark. Like, this is a new product feature line that they've got that is appealing to a new demographic, and it is being successfully received by many in that demographic and serving them pretty well. So, I mean, I'm you know my gripes on the DGX Spark. Like, those are historical. You can just go back in the channel and watch all those. But, still, it is meeting market demand. It is a viable product. It is a viable product for that reason. If there is a major pullback, like, that's going to hurt everybody across the board, especially the people that are selling B300s for internal enterprise use cases where they may be setting up and running their own OSS models that are Chinese origin or maybe a Nvidia tune. That happens quite often of a Chinese origin model. Literally, Meta is not even a part of the conversation anymore. You could always hold out hope and be surprised, but I don't think that's going to play out. If you look at the demand, though, for those systems from the hyperscaler perspective, also, it's impacted because a lot of companies will go and take those up from them, also. There's a lot of demand impact for the Chinese ecosystem models because of the cost savings on tokens. And I think that doesn't change. And this is definitely something that, if you look at, can Nvidia survive as well as it is right now in its current format, if it was just selling GPU or you know, the enterprise systems to the hyperscalers for non-OSS needs, it's going to be a smaller amount of B300s that are demanded versus what is currently in demand. That that that that that could have huge implications. Like I I can't even begin to state the implications. People in finance could understand that much better than I can. I'll just tell you big. Weaker demand in hyperscalers can like they can't just pick up excess demand and just be like let's add on extra things, especially given the current environment around data centers and data center public pushback and pullbacks of actual products that are projects that are now happening in data centers as a result of that. So, this could definitely damper the very aggressive, very it's evolving and getting faster hardware development life cycles that we're seeing because you need the product to be generating a reason to buy it each time and each time it needs to be generating like warm money. So, currently there is no national restriction yet on your access to hugging face models. There's no ID requirements, download what you want. I urge you by the way, this is not the first time I've said this. Like go back a couple months. I urge you to download your models locally. And if you see something that has come on the the hugging face repo that you're interested in, consider downloading it now because it it may not be accessible easily. I know of course people are like, we'll set up the hugging bay and like I love the idea and stuff like that. But you've got to also factor in like there there is the potential for onerous oversight I if there if this is a master class in winning, like who is this all who's all this information good for? I can only think of like two two companies and like that's it. Both of those happen to just be the closed model ecosystem out there that are the largest players in the United States. Damn. There is no yet release of open weights restrictions and law. So, there is not like regulation that says, "Hey, by the way, you can't do that yet." Now, they can always flex national security. By the way, national security is what the Chinese are going to flex also, it looks like from that document. But, they can always flex national security at any point in time, like they did recently with Fable Mythos, and say, "Pull that sucker back, or else." And the or else is like, "You're not going to be allowed to do business. We're going to pull the plug at the like data center level." This is the cloud, by the way. This is the cloud. This is why I've for all these years been like pretty anti-cloud, is because you can actually do that in a data center setting insanely well. Regulatory capture is like the playground of of the cloud. Currently, not specific regulation around it. That doesn't mean that there couldn't be. As a matter of fact, if we see specific regulation around it, it's going to be represented by the special interests of the moneyed peoples who are not probably interested in open-source stuff, and they're probably going to be writing laws against you trying to access open-source stuff. Does that make sense? It doesn't make sense to me, either, but that's probably what's in the pipeline. Should you be pissed? You should be like yelling right now at your computer screen, honestly. Was there always going to be unlimited open-source models released? That's a very fair question. That's a very good question. That's a question I struggle with. So, why is this all happening now? This is probably weeks behind, possibly months behind what's been going on behind doors. And like, you're going to see not what they see. You're going to be the recipient of the information, not the creator of the new cycle. That is your place as a consumer in this society. So, expect that there's a lot happening and already happened and already in the works that you don't know about. There are tons of think tanks that are working on legislation for this right now. It is very probably very bad, most of that legislation, for most people that are trying to be independent, self-sovereign in what their data ownership looks like, not tied to the cloud, and not tied to a monthly subscription. So, you you really can't have this both ways. Like I I've said this before. There's a reason I don't have an Anthropic account. I don't have an Anthropic account. There's a reason I don't have an OpenAI account. So, do you think this is all part of a plan? Let me ask you. This is this Do you think this is all just random action reaction kind of happening? Or do you think that there's a possible geopolitical master model behind all of this running on servers, fleets of servers right now, providing the best intelligence around strategic development for a couple of companies, maybe one in particular. I think there is a very large likelihood that there is. And we may never know about it. Wow. That's pretty crazy. So, I look forward to reading what you say. Everybody buckle up. We're in for a long ride as far as local AI, as far as cloud AI, as far as what comes down the pipeline. 2026 is going to be interesting, but just wait. 2027's the year it really all pops off. Hit like, hit subscribe. Thanks to all of our channel members. Like let me say this. It would not happen the way it happens. I would not say the things I can say because you can't get that next sponsorship if you talk frankly, honestly about a bad product. I A lot of people try to have it both ways and say that it's fine. It it really isn't. And you should not trust things that come out of the mouth of people who literally have sponsors that are related to that sponsor. They have interests to protect, their own financial interests. I have your interest to protect, my interest to protect, but that is not a sponsor's interest ever. I work off an affiliate revenue model for how the channel gets funded outside of that. Like I find a good idea that's attractive, I present it to you. If it actually is, it may sell. If it's not, it may not. And if you are interested in budget getting into local AI, which I would urge I would definitely contract my spending on local AI hardware at this point because it is yeah. Do Do you think anybody else that's selling DGX Sparks or getting DGX Sparks or anything like that? Do you think they're going to give you advice like you should contract your spending on local AI hardware? No. Of course they're not. But because of my members, I can say what actually is going on, which is a very rare thing on YouTube nowadays. And so if you value this channel, consider joining up as a member. If you can't join as a member, consider sharing this out there. Consider getting your friends to join also and get them onto this message. This message is important. This is going to shape the future and this is your insight that you're not going to get anywhere else. No other channel's going to present what I have presented to you today. And if I disappear, you better believe there's not anybody to fill the void. I hope you have a good one. DSP out.