NVIDIA launches Rubin: Six new chips and complete architecture change at TSMC confirmed

NVIDIA is igniting the next stage of its AI offensive program, report our colleagues at Wccftech, and Jensen Huang himself provides confirmation: The new “Ruby” architecture is not only under development, but has already reached the tape-out stage with six different chips at TSMC. According to Huang, this is the company’s most advanced AI platform to date – a statement with strategic weight.

Ruby is more than just a GPU series – it is a complete technological leap: in addition to new GPUs and CPUs, a scalable NVLink switch, a silicon photonics processor and a completely revised chiplet design are also being implemented. NVIDIA is thus increasingly moving away from the classic monolith approach and focusing on a modular, scalable architecture. Production is carried out using TSMC’s N3P process (3nm), including CoWoS-L packaging, and for the first time with a 4x reticle layout – a dramatic improvement over the 3.3x reticles of the previous Blackwell series.

The aim is the seamless integration of high-performance components: Rubin will work with HBM4, the successor to HBM3E, and is designed primarily for data centers with extremely high throughput and energy efficiency. The switch to the chiplet model in particular opens up new possibilities for scaling and yield optimization – a game that Intel and AMD have been playing for years, but in which NVIDIA has not been present until now. Jensen Huang himself emphasized during his trip to Taiwan: “Rubin is very advanced and we have now taped out six brand new chips to TSMC, so all of these chips are now in TSMC’s fabs.” This also makes it clear that NVIDIA is not relying on a gradual introduction of individual products with Rubin, but is modernizing the entire tech stack at the same time.

The timetable is ambitious: Rubin is expected to be ready for the market in 2026 or 2027 – depending on the results of trial production. But the course has been set. The parallels with Hopper’s disruptive launch are obvious. What Hopper was for the first wave of AI, Ruby is set to become for the second: a technological foundation for a new era of data-driven high-performance computing. With Ruby, NVIDIA is underpinning its claim to remain the reference in the AI acceleration market – both technologically and strategically. While competitors such as AMD with MI300X and Intel with Gaudi 3 are trying to catch up, NVIDIA is relying on the proven principle of disruption through integration: new memory, new nodes, new architecture, new infrastructure. The Ruby tape-out is just the beginning.

Source: Youtube