Rubin GPU could face delay
A report says NVIDIA's Rubin GPU could see delays because HBM4 supply is falling short, potentially complicating launch timelines as cloud and OEM demand climbs reported. That supply squeeze intersects with TPU demand and partner capacity plays.
Samsung Electronics and SK Hynix have reportedly been designated as Vera Rubin’s HBM4 suppliers, with industry sources saying Micron was left out of the design win. (kedglobal.com) Market reports put the supplier split heavily in SK Hynix’s favor — roughly 70% SK Hynix and 30% Samsung in some accounts — even as TrendForce and others expect HBM4 validation and initial shipments to ramp around 2Q 2026. (dataconomy.com) Taiwan’s Commercial Times, cited in coverage of the chain, says memory makers are redesigning base-die components for HBM4, a technical tweak that could push Rubin shipments back about one quarter and has prompted reported reductions in Rubin wafer starts. (benzinga.com) Separately, analysts flag TSMC’s CoWoS/advanced-packaging bottleneck as a parallel constraint: several forecasts trimmed Google’s 2026 TPU target from ~4.0M to roughly 3.0–3.2M because CoWoS slots are tight. (moomoo.com) NVIDIA teams have reportedly been auditing Samsung’s packaging campus in Cheonan as part of HBM4 verification and qualification work, signaling last-stage OEM testing rather than finished mass supply. (digitimes.com) Analysts tracking fabs say relief depends on TSMC and packaging capacity expansion, with major CoWoS capacity increases expected to phase in around 2027–2028 — a timeline that would affect both Rubin volume and third‑party TPU availability. (creating-nanotech.com)
