Filtered to AI Hardware & Semiconductors · show all articles
Coverage of China's chip industry usually frames the gap with the West as a performance-gap story. The sharper, less-told version is a yield story — and the number behind it is startling.
Nvidia designs the GPU. SK hynix supplies the memory stacked directly on top of it, at a market share high enough that a single Korean company's production schedule now sets a real ceiling on AI accelerator supply.
While most of this cohort tracks gigawatt-scale datacenter deals, SiMa.ai is building AI chips meant to run inside a camera, a robot, or a car — nowhere near a datacenter at all.
While Samsung Foundry chases TSMC's yield, Samsung's separate memory division has been raising DRAM contract prices by double digits per quarter — a distinct business having a distinctly better year.
Samsung competes with TSMC for advanced-logic customers and with SK hynix for memory customers, using entirely different technology and entirely different scoreboards. Coverage that treats "Samsung" as one competition misses both.
Most accelerators run a fixed circuit and make the software adapt to it. SambaNova's reconfigurable dataflow architecture does the opposite — reshaping the hardware itself around whatever model it's running.
A multi-billion-dollar manufacturing partnership with GlobalFoundries, next-generation R-Car silicon, and a legacy-fab closure squeezing older analog parts all at once — Renesas's 2026 is a study in managed transition.
Korea's chip industry is world-famous for memory. Rebellions is trying to build the country's own answer to Nvidia — a very different bet than anything SK hynix or Samsung's memory business is making.
"$73 billion backlog." "$95.2 billion in supply commitments." This cohort has cited figures like these across a dozen companies. Here's what they actually mean, and the specific questions worth asking before treating one as a guarantee.
Most countries rebuilding a domestic chip industry start with a mature node and work up. Japan's state-backed Rapidus is attempting to go straight to 2-nanometer production by 2027 — with IBM's help and more than 60 companies watching.