NVIDIA quickly denies! Fubon indicates that the next-generation GPU Rubin may be delayed due to redesign

According to the Zhitong Finance APP, Fubon Financial Holding, a financial group in Taiwan, stated in its latest research report that the production process of NVIDIA's (NVDA.US) next-generation graphics processor Rubin at TSMC (TSM.US) may be delayed due to chip redesign. Fubon analyst Sherman Shang mentioned in a research report: "We believe that the Rubin chip is very likely to face delays. Although the first version of the design was completed in late June, NVIDIA is currently redesigning the chip to better match AMD's (AMD.US) upcoming MI450 series products." Shang further analyzed: "According to the latest progress, the redesigned chip is expected to complete tape-out in late September or October. Based on this timeline, the shipment volume of the Rubin chip in 2026 will be relatively limited." In the semiconductor manufacturing field, tape-out refers to the final stage where the integrated circuit design is delivered to the wafer fab for production after verification. In response, NVIDIA officially denied this. A company spokesperson stated: "The content of the report is inaccurate, and the Rubin project is still progressing as planned." Previously, multiple sources indicated that the Rubin chip was originally scheduled to enter mass production by the end of 2025 and officially launch in early 2026. Nomad Semi analyst Moore Morris revealed on social platform X that the Rubin GPU will succeed the currently ramping-up Blackwell architecture. Data shows that the shipment volume of Blackwell chips reached 750,000 units in the first quarter of 2025, increasing to 1.2 million units in the second quarter, and is expected to reach 1.5 million and 1.6 million units in the third and fourth quarters, respectively. Morris also pointed out that currently, AMD and Broadcom (AVGO.US) are the fastest-growing customers of TSMC's CoWoS (Chip-on-Wafer-on-Substrate) packaging technology. In 2025, NVIDIA will still hold a dominant capacity share of 51.4%, while Broadcom and AMD will account for 16.2% and 7.7%, respectively. However, by 2026, the capacity shares of Broadcom and AMD are expected to increase to 17.4% and 9.2%, respectively, while NVIDIA's share will slightly decrease to 50.1%