Speaker 0: we thought washington's export sanctions had frozen china's chip tech in time, but huawei just rewrote the entire rule book to bypass them. if you think a chip's power is only about how small you can shrink its transistors, the next decade is about comp to completely flip your reality. so huawei is subverting the us chip sanctions. so blocked from buying advanced factory equipment to shrink transistors, huawei unveiled the tau scaling law and logic folding architecture. instead of making transistors smaller, they are physically stacking and folding older legal circuits in three d to match cutting edge speeds. this completely challenges the logic leadership of tsmc and intel by shifting the definition of advanced tech from raw factory manufacturing to clever systems architecture. what do you think about this, john? Speaker 1: yeah. i think this is cool. i think it's funny because with all the sanctions that have been brought, they were made to kind of slow, progress for these companies. but, but huawei has found a way. so let's talk about the problem huawei is trying to solve, and then let's dive into logic folding. so modern chips have two fundamental challenges, and we kinda just talked about this, in our previous, our previous episode there. transistors, under moore's law, are becoming extremely difficult and expensive to shrink, and data movement is becoming a bottleneck. so, what semiconductor companies have tried to improve performance is by reducing transistor size. so they're now at 1.4 nanometer, i think. so while these dimensions are approaching atomic scale, the more you try to shrink, it becomes more difficult and expensive and more power intensive. and at the same time, these modern ai workloads, they spend enormous amounts of time moving data between all of the components on the chip, the compute units, the memory, the cache, the network fabrics. and so, a lot of engineers are arguing that data movement, not computation, is becoming the dominant performance bottleneck, and i'll i'll ask if you agree with that after. so huawei is proposing a solution called logic folding. so it's actually kind of similar to our previous episode as well. they're arranging circuit so instead of arranging circuit circuits in two dimensions, they're proposing vertically integrating logic memory and analog circuit to shorten the communication paths inside the chip. so if you think of it like this, a traditional chip has a room, then a hallway, then a room beside it, then a hallway, then a room beside it. okay? so logic folding is just stacking rooms on top of rooms. you eliminate that hallway. okay? so you've got shorter the the information has to travel shorter distances. and so huawei's claim is that future versions could deliver transistor density equivalent to 1.4 nanometer class technology by 2031. and just to kind of talk about this whole controversy, huawei was blocked by using all of the latest and greatest technology, and this is how they came up with that. so what are your thoughts on this? Speaker 0: i think, necessity is the mother of invention. Speaker 1: yeah. Speaker 0: as they say, you know, we gotta build data centers. everyone has to build them out. we can't get these chips anymore. what do we do? oh, we found a way to make them better. overall, it's probably better for the industry. but as we get into the more concept idea of we're adding into a sovereign ai world yeah. people don't wanna believe it or whatnot. but, you know, you know, us government's gonna have an ai that protects the us, and china's gonna have an ai, if not just one hundreds, whatever, for military, monitoring, communications, all these things. so from a consumer point of view, i love it. yeah. because i think we're all gonna get better. from a overall ai nation state race, you know, china's doing some, seemingly, you know, we're getting information. you know, we're not in china looking at it, but they're doing some seemingly amazing advanced tech things that seem to be leapfrogging what's happening here in north america. and i question, who's gonna be the leader in four or five years? Speaker 1: yeah. well, look at china's ai models. right? they're they're competing, if not if not, surpassing at some at points. right? Speaker 0: yeah. and, you know, people like that idea, hate that idea, what have you. and people will say, gotta use this, gotta use that. but, and i'm not pointing fingers at anyone. but openai's 5.5 is $30 per million output tokens. deepseek version four pro is 80ยข per million output tokens. so if you're using apis to do it and one is 80% of the other, but it's literally, 40 times cheaper almost, what would you do? Speaker 1: yeah. Speaker 0: so think of this update like building a city where you officially run out of land. traditional chipmakers like t m s tsmc are finding hyper advanced microscopic materials to buy tiny single story homes closer together. huawei can't buy those tools. so instead, taking ordinary, readily available bricks and building massive interconnected three d scribe to scrapers. so you get the same population density and efficiency just by building, up instead of out. so we just talked about, what tsmc and those seem to be banking on, that 10 millimeter wafer in between where they stack the crystals, and that looks great. i don't know how this is gonna compare long term, but these are expected fall twenty twenty six, including premium hardware packages, debuting exclusively in huawei's upcoming flagship kirin mobile processor lineup extending into their superpod data center cluster. so this is happening as we speak. Speaker 1: yeah. this is exciting. yeah. i mean, this again, innovation is always exciting. Speaker 0: so while western media is framing this as a sudden surprise, huawei actually peer reviewed this entire system level thesis by formally presenting it to the global elite engineers at the ieee international symposium on circuits and systems. the framework wasn't built overnight to dodge recent laws. huawei has been quietly running this parallel process since 2020, creating us entity list placement as an architecture problem rather than a geopolitical one. and by, deploying unified bus to attack data center latency, huawei is taking a direct swing at nvidia's ultra lucrative proprietary nvlink ecosystem, turning a smartphone hardware defensive play into an aggressive enterprise data offensive. so this is interesting. by stacking logic, layers generates a massive concentrated heat footprints like we were talking about before where, the crystal stacking and we were talking about and that layer in between avoids the, 600 plus degrees celsius needed to make some of these things happen. without access to cutting edge high bandwidth memory and electronic design automation, huawei risk hitting a thermal wall that architectural cleverness simply can't solve. but we say that, but, you know, net tomorrow, wait. we solve this thing that we've never been able to solve before. Speaker 1: yeah. Speaker 0: so huawei just proved that when you lock a brilliant engineering team out of the world's best factories, they will simply reinvent the fundamental physics of how a computer chip is organized. it is a system level stacking a legitimate substitute for advanced physics, or is huawei just hitting a temporary coping mechanism? drop your thoughts in the chat below, and will you be jumping on grabbing one of these? let us know.