Intel Joins Musk's $25B Terafab as Foundry Partner

Kind: captions Language: en Welcome back to the Awesome Agents podcast. I'm Alex Rivera and today we're talking about a deal that I genuinely did not see coming. >> Yeah, this one landed on a Monday and kind of broke the semiconductor news cycle for a few days. >> So, the headline is Intel is joining Terrafab, Elon Mus massive chip joint venture as the primary foundry partner. We're talking a $25 billion project >> and that number is real. Terrafab is a joint venture between Tesla, SpaceX, and XAI. All Musk controlled entities at this point, >> right? SpaceX absorbed XAI back in February. So now you've got one giant Musk tech entity and they're building their own chip fab. >> With Intel doing the actual manufacturing, Intel is contributing its 18A process node, which is their most advanced node currently in production in the US. >> Okay, 18A. What does that mean in practice? >> It's a 1.8 8 nanometer class process. Two key innovations, ribbon fet, which is a gate all-around transistor design, and power va, which delivers power from the back of the chip instead of the front. >> And that matters because >> backside power delivery frees up the routing layers on top of the chip. Less congestion, better signal integrity, lower power consumption. It's a real architectural shift. >> So, Intel's technology here is actually competitive. They hit 65% commercially viable yields on 18A in early April. That's the milestone they've been chasing for two years. Not exceptional. TSMC runs higher, but it's good enough to produce at scale. >> And the timing wasn't accidental, right? Yields hit, deal announced almost the same week. >> Intel isn't dumb. They waited until they had something concrete to show before signing on to a deal this visible. >> So, from Intel's side, what does this actually mean? Because their foundry business has been kind of struggling. That's the real story here. Before this deal, Intel Foundry had $37 million in annual external customer revenue. That's a rounding error compared to TSMC. >> Ouch. >> And they had this circular problem. No anchor client means you can't prove the tech at scale. Can't prove the tech means no anchor client. Lip Bhutan took over as CEO in March 2025 and has been trying to break that loop ever since. >> And Terrafab breaks the loop >> completely. If Intel is processing chips for Tesla's AI5 and AI6 silicon, they'll have real wafer volumes, real yield data, real proof points to show Apple or Qualcomm when those conversations happen. >> You mentioned AI5 and AI6. Those are the chips for Optimus Robots and the Cyber Cab. Right. >> Correct. But here's the part that surprised me. Those chips are actually the smaller slice of Terraab's production. Only about 20% of output is earmarked for ground applications. Wait, so what's the other 80%. >> SpaceX's orbital data centers, radiation hardened chips designed to operate in space on the Starlink constellation. >> I didn't know space compute was the main point of this whole thing. >> Most people don't. The headline is Optimus robots, but the volume is space. And radiation hardening means defect tolerance is very low, which is actually a bigger ask for Intel's yield than regular chips. >> So what are the real risks here? >> A few. Intel's yield at 65% still trails TSMC. For chips going into satellites and humanoid robots, you need high reliability. That bar is harder to clear. And the scale targets are genuinely hard to believe. 1 terowatt of annual AI compute capacity is roughly 50 times total global AI compute output from 2025. That's a vision, not a production plan. >> I mean, the campus itself, Musk said, 100 million square ft. That's 10 times the size of the existing Giga Texas building, which is already nearly a mile long. I drive past that thing sometimes. >> So, you can verify the audacity in person. >> Huh. Yeah. The other thing I want to flag is the public money angle. >> The chips act subsidy, $2 billion attached to this deal, >> which was designed to reshore chip manufacturing for national security. But 80% of these chips are going to a private satellite network. That's a question Washington will eventually have to answer, especially given that Musk also runs a government cost cutting office. Receiving $2 billion from the same government you're auditing is, let's say, a standout position to be in. >> So, where does this land? Is this good news for Intel? Genuinely, >> yes, with a caveat. It's the best outcome Intel could have hoped for right now. They needed an anchor client. They got one, but they're betting on a customer base that's almost completely internal to one person's companies. If the relationship with Musk's sour, Intel has a very large empty building. >> Exactly. And Intel said this deal could open the door to Apple and Qualcomm conversations. That's the longer game. Terrafab proves 18A works at volume. That proof is what the next customer needs to see. >> So, this is Intel using Musk to win back the semiconductor industry's trust. >> That's a pretty clean way to put it. >> Okay, Maya, final thought. Is this a turning point for Intel Foundry, or is it still too early to call? Ask me again in 18 months when the first production wafers are shipping and we can see the actual yield numbers. Right now, it's the most credible bet Intel has made in a while. That's something. >> Fair enough. That's Intel, Terrafab, and what might be the most consequential chip manufacturing deal of 2026. Thanks for walking through it with me, Maya. >> Always. >> You've been listening to the Awesome Agents podcast. We'll be back soon with more.