Now Playing
MediaTek Stays with TSMC, HBM4E Arrives Early & Nvidia's $150B Taiwan Bet
Audio is available on Spreaker — see link below.
Samsung's foundry bid collapses as MediaTek reaffirms TSMC across every node, while Samsung memory surprises with HBM4E samples six months ahead of rivals. Nvidia commits $150B to Taiwan and AMD follows — here's what it means for the AI infrastructure stack.
Audio is available on Spreaker — see link below.
Samsung's push to win MediaTek's foundry business just failed publicly, and that outcome tells you more about TSMC's competitive position than almost anything else happening in chips right now. At MediaTek's shareholder meeting on May twenty-ninth, CEO Rick Tsai confirmed the company is staying with TSMC across the full node range from twelve nanometers down to one-point-four nanometers.
The important distinction is that MediaTek's rejection of Samsung's foundry pitch doesn't mean MediaTek is giving TSMC a free pass on packaging. On the same day, MediaTek's senior VP confirmed they're now supporting both TSMC's CoWoS and Intel's EMIB packaging technologies.
While Samsung's foundry unit absorbed another setback, its memory division moved in the opposite direction. Samsung began HBM4E sample shipments on May twenty-ninth, roughly six months ahead of where SK hynix and Micron are tracking.
The larger backdrop for all of this is the capital concentration happening in Taiwan. Nvidia has committed up to one hundred fifty billion dollars in annual Taiwan spending, expanding its partner ecosystem from fifty to one hundred fifty companies.
On the policy side, the EU's draft Chips Act two-point-zero targets one hundred twenty billion euros in public-private investment by twenty-thirty-five, anchored by a thirty billion euro proposal for a three-nanometer foundry. The funding depends on post-twenty-twenty-eight EU budget negotiations, and member-state commitments remain unconfirmed.
One number worth tracking into twenty-twenty-seven: MediaTek has doubled its twenty-twenty-six data center revenue forecast to two billion dollars, and is targeting ten to fifteen percent of a seventy to eighty billion dollar custom AI ASIC market. That's an aggressive claim.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.