Chips & Architecture
5 articles covering this beat
NVIDIA GTC 2026: Vera Rubin in Production, Feynman on the Horizon
NVIDIA's Vera Rubin platform — 336 billion transistors, 288GB of HBM4, 50 petaflops per GPU — is already in production and shipping to hyperscalers. But the real reveal at GTC 2026 may be Feynman: the first 1.6nm AI chip with silicon photonics, targeting 2028.
The HBM4 Race: SK Hynix, Samsung, and Micron Battle for NVIDIA's Memory Orders
With NVIDIA's Vera Rubin demanding 288GB of HBM4 per GPU — 576 stacks per rack — the three memory giants are in an all-out production war. SK Hynix leads with ~70% of NVIDIA's allocation, Samsung is shipping its turnkey solution, and Micron is ramping 15,000 wafers per month.
Alphabet Loses TPU Production Slots as NVIDIA Locks Up CoWoS Capacity
Google has cut its 2026 TPU production target from four million to three million units after NVIDIA secured over 50% of TSMC's CoWoS advanced packaging capacity through 2027.
The Weekly Briefing
Every Sunday. The most important supply chain developments of the week, with analysis on what they mean for the AI ecosystem. Free.