Now Playing
Advanced Packaging Crunch: TSMC vs Intel, Samsung's HBM4E & FCC Bans
Audio is available on Spreaker — see link below.
AI chip packaging — not transistors — is now the binding constraint on the AI hardware race, as TSMC's CoWoS hits a 78-week bottleneck and Intel's EMIB moment finally arrives. Samsung posts a 1,814% profit surge on HBM demand while the FCC moves to ban Chinese robotics and power inverters from U.S. data centres.
Audio is available on Spreaker — see link below.
TSMC is building something that looks a lot like what Intel has had for years, and that tells you exactly where the AI hardware race now stands. The announcement is a partnership with Kinsus, a Taiwanese substrate specialist, to develop an embedded-bridge packaging architecture.
Intel Foundry's head made that point explicitly this week, stating that chip packaging, not transistors, will define the AI revolution. That's a remarkable framing from a company that spent decades selling itself on silicon leadership.
On the memory side, Samsung posted a one thousand eight hundred and fourteen percent year-on-year increase in operating profit for the second quarter. The driver is AI server demand, specifically HBM and high-density DRAM pulling margins up sharply.
The geopolitical layer shifted this week as well. The FCC moved to ban imports of new Chinese humanoid robots and networked power inverters, citing national security.
Pull back to the macro picture and the numbers are striking. Global semiconductor sales are projected to reach one-point-five trillion dollars by twenty-twenty-six, roughly ninety percent year-on-year growth, driven almost entirely by AI data-centre demand.
The core risk running under all of this is AI return on investment. Hyperscaler capital expenditure at this scale depends on AI model monetisation delivering returns that justify it.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.