Chamath says Taiwan stops mattering for chips in 18 months, and this week the show laid out the five-layer case for why that's way too neat a story. Fabrication, advanced packaging, equipment know-how, talent, customer qualification — those are five separate things, and they all have to move off-island at once. All-In just let the clock keep running. We’ve also got a BigID CEO saying AI risk lives in the data, not the model, and Amy Webb plus Harper Reed on TWiT asking whether US tech dominance is actually ending. Let’s get into it. All-In Podcast writes:
We're 18 months from Taiwan not being an important moment of conversation the way it is today. Why 18 months? Because we are at a point where we're probably 1 to 2 nanometers away from being able to do what we need Taiwan to strategically do for us. And so, as we scale up our chip fabs, as we get more capacity, and interestingly, there are these orthogonal technologies being developed.
Chamath on All-In this week: eighteen months until Taiwan stops being a strategic conversation, because the US is supposedly one to two nanometers away from closing the gap on its own. That is an extremely specific number for a very confident take. And our Step Back segment this week walked through the five layers that have to move before that even starts to be true — fabrication, advanced packaging, equipment know-how, talent, and customer qualification. Chamath picked one of them and treated it like the whole timeline. The nanometer gap is real, sure, but that’s maybe layer one. The packaging know-how, the ASML equipment dependency, the decades of process engineers Taiwan has built up — those are not on the same eighteen-month schedule. And then Chamath jumped to Neuralink’s surgical robot as proof of American mechanical dexterity, which is… not exactly the same supply chain. Exactly. It’s the same move we flagged all week: a confident VC turns a five-variable problem into one number, and nobody in the room pushes back. Did Friedberg or Sacks ask which of those five layers actually closes in eighteen months? Because the transcript doesn’t show it. Okay, I keep hearing very confident people on podcasts say Taiwan just won’t matter for chips in, like, 18 months. What would have to be true for that to actually happen? So this claim falls apart pretty fast once you look at what the chip industry actually is, because it’s not one thing you can pick up and move. There are at least five distinct layers: the fabrication fabs themselves, advanced packaging, the specialized equipment and process know-how baked into those fabs, the engineering talent, and then customer qualification — the months-long process where a company like Apple essentially certifies that a new fab makes chips that meet its exact specs. On fabrication, TSMC has been expanding into the US, Japan, and Europe, but per a detailed analysis in the Kaohsiung Times by Taiwan’s own representative to Singapore, the R&D core, technology development, and leading-edge process nodes remain anchored in Taiwan. That’s not spin — that’s the structural reality. On the US side, Commerce Secretary Howard Lutnick has stated a goal of bringing 40 percent of Taiwan’s semiconductor supply chain to America, but analysts cited by CNBC flatly doubt that would be easy, especially given Taipei’s hard line on keeping its most advanced technology at home. And on the fab-build timeline: TSMC’s Japan expansion for 3nm production comes with an estimated $17 billion price tag and years of ramp — that’s one node, at one location. So if even the companies trying to move this stuff say it takes years per facility, where does the '18 months' number even come from? Honestly, it sounds like it’s mixing up announced investment with operational capacity — and those are very different things. What I’d watch is whether customer qualification timelines actually compress, because that’s the silent bottleneck nobody talks about: a new TSMC Arizona fab doesn’t replace Taiwan until Apple, Nvidia, and others formally sign off that the chips are equivalent, and that process alone has historically taken the better part of a year per node. The silicon shield isn’t just geography; it’s accumulated process knowledge and customer trust, and both move slowly. From Dimitri Sirota at SC Media:
Most organizations think AI risk lives in the model – or the identity. It doesn’t. It lives in the data. In this episode, BigID’s CEO reframes the conversation: why legacy access controls are breaking down, why visibility into sensitive data is the missing foundation, and what it takes to govern humans and machines under a single, accountable framework.
Dimitri Sirota from BigID on Enterprise Security Weekly — and his core argument is that most organizations are looking for AI risk in the model or in identity, and he says both are wrong. The real exposure is in the data itself, specifically the lack of visibility into sensitive data before agents ever touch it. Which is a genuinely clean reframe — except Sirota is the CEO of a data management platform, so I’m asking the same question I asked on Heron Power and Suno: is this scar tissue from watching real breaches, or is it a product thesis dressed up as industry insight? The cascading breaches segment gives it some grounding — they’re specifically talking about third- and fourth-party attacks that chain through compromised tools and services. That’s the same attack surface Jake Moore demoed with the deepfake hiring scenario a few episodes back. The pattern isn’t isolated. And it closes the loop on May 15, when we were asking whether someone would actually name the data-stack problem for AI workloads. Sirota named it. Now the question is whether ESW got a practitioner with specifics, or just the vendor with the most to gain winning the framing war. Probably worth hearing which version they got. This Week in Tech, with Leo Laporte:
Is the era of American tech dominance ending? Get an inside look at how China’s pragmatic approach to AI, robotics, and hardware is shifting the global balance, and why the US might need a new playbook to keep up.
TWiT’s Sunday episode drops Amy Webb and Harper Reed into the question of whether American tech dominance is actually ending, and the timing is interesting. Chamath was on All-In this week confidently dismissing Taiwan’s chip position in 18 months, and our own Step Back segment just mapped out five distinct layers — fabs, advanced packaging, equipment know-how, talent, customer qualification — that make that timeline fall apart. The real question is whether Webb and Reed get specific, because we already have 'China doing smart distillation under compute constraints' as an actual operating detail from earlier this week. If TWiT just lands on 'China is rising, the US is sleepwalking,' that’s not adding much to what’s already on the table. Worth flagging that the show also has an H200 clearance story baked in — the US just approved chip sales to ten Chinese firms while Jensen Huang is literally in China with Trump. That’s not a vibe, that’s a policy contradiction happening in real time, and it either sharpens the dominance question or quietly undermines it. If Webb and Reed can thread the H200 contradiction — 'we’re selling them the hardware while worrying about the hardware gap' — into something more than anxious macro, queue it. If it’s just the same China-rising sermon, the minerals-and-fabs reporting from earlier this week already did the heavier lifting. If Tech Podcast Podcast is part of your daily routine, consider subscribing and leaving a quick review wherever you’re listening. It really helps other people find the show, and it helps us keep making it better.
You’ll find links to everything we talked about today in the show notes. If a story caught your ear, that’s the place to dig in a little further.
That’s Tech Podcast Podcast for today. This is a Lantern Podcast.