Several factors fueled reports of potential delays:
| Milestone | Date / Window | Source |
|---|---|---|
| Rubin GPU and Vera CPU taped out, in fabrication at TSMC | August 2025 | |
| Full production announced at CES 2026 | January 5–8, 2026 | |
| TSMC wafer starts at 40,000–50,000/month | Q2 2026 | |
| First batch final packaging complete | July–August 2026 | |
| Initial customer shipments | Q3 2026 | |
| Volume ramp | Q4 2026, continuing into H1 2027 | |
| Rubin Ultra (next step) | H2 2027 |
The platform uses approximately 500 billion transistors on TSMC's N2 process, HBM4 memory at 13 TB/s bandwidth, NVLink 6 at 5 TB/s bidirectional, and delivers up to 8 exaFLOPS per rack . Nvidia's official website states the platform is 'ramping into full production, with Taiwan's top server makers and global supply chain leaders manufacturing at scale' .
Vera Rubin delay concerns sit against a pattern of production challenges:
Bottom line: Huang is publicly adamant that Vera Rubin is on track with production already underway. The credible counter-evidence centers on HBM4 memory bottlenecks and packaging constraints that could slow the volume ramp, rather than a full halt. The separate Kyber rack delay (to 2028) and the history of Blackwell's teething problems add to the market's sensitivity to any Nvidia supply-chain signals.