Nvidia GTC Taipei 2026: Vera Rubin NVL72 Specs, N1X Laptop Chip, and What to Expect

Jensen Huang is returning to the Taipei Music Center on Monday, June 1, at 11:00 a.m. Taipei time to kick off GTC Taipei 2026, which runs concurrently with Computex . Nvidia has described the Vera Rubin platform as "probably the largest product launch in the history of Taiwan," involving nearly two million parts and roughly 150 ecosystem partners in the country . Beyond the much-anticipated data center silicon, Huang is expected to formally reveal the N1X—Nvidia's first laptop SoC designed specifically for Windows on ARM—and detail a $150 billion annual investment in Taiwan along with the location of a new Taipei headquarters .

Vera Rubin Data Center Platform: Breaking Down the Rubin R200 GPU

The Rubin R200 GPU is the computational core of the Vera Rubin platform. Built on TSMC's N3P 3nm process, the chip is a dual-die behemoth containing approximately 336 billion transistors—a density increase of roughly 1.6x over the Blackwell B200 . Nvidia has positioned the GPU for the era of agentic AI, equipping it with a new Transformer Engine and support for the NVFP4 4-bit floating point format to dramatically boost inference throughput .

Each R200 GPU is paired with 288 GB of HBM4 memory delivering up to 22 TB/s of bandwidth . Nvidia's own product pages caution that initial production bandwidth may be lower than the 22 TB/s target due to memory supplier ramp challenges, but the long-term specification remains unchanged . The GPU achieves 50 PFLOPS of NVFP4 inference performance and 35 PFLOPS of NVFP4 training performance, representing a 5x and 3.5x improvement over the previous Blackwell generation, respectively . For FP8/FP6 training workloads, each GPU delivers 17.5 PFLOPS .

Interconnect bandwidth also sees a generational leap. NVLink 6 delivers 3.6 TB/s of GPU-to-GPU bandwidth, double that of NVLink 5 in the GB200 NVL72 .

Vera Rubin NVL72: Full Rack Specifications

The Vera Rubin NVL72 is Nvidia's third-generation rack-scale AI supercomputer. A single rack houses 72 Rubin GPUs and 36 Vera CPUs, paired as 36 Vera Rubin superchips . The Vera CPU features 88 custom Armv9.2 "Olympus" cores, delivering 2.4x the memory bandwidth and 3x the memory capacity of the previous Grace CPU .

Key NVL72 rack specifications include:

• Total GPU memory: 20.7 TB HBM4, with an aggregate bandwidth of up to 1,580 TB/s .
• Total fast memory: 75 TB combining HBM4 and LPDDR5X .
• NVFP4 inference: 3,600 PFLOPS (3.6 ExaFLOPS) .
• NVFP4 training: 2,520 PFLOPS .
• FP8/FP6 training: 1,260 PFLOPS .
• Scale-up fabric bandwidth: 260 TB/s via nine NVLink 6 switch trays .
• Networking: 144+ ConnectX-9 800G ports supporting Spectrum-6 Ethernet or Quantum-CX9 InfiniBand .
• Cooling: 100% liquid cooling with a 45°C inlet water temperature, enabling free cooling in most climates without additional chillers .
• System power: Approximately 190 kW (Max Q) to 230 kW (Max P) .

Vera Rubin Production Timeline

Nvidia officially launched the Vera Rubin platform at GTC 2026 in March, with customer sampling already underway . Huang is expected to use the Taipei keynote to provide updated production timing and volume commitments. The current public roadmap points to a production ramp through the second half of 2026, with the first DGX Vera Rubin NVL72 systems shipping during that window . Nvidia has also teased its next-generation "Feynman" architecture, though details remain limited .

Nvidia N1X: An ARM-Based Windows Laptop Processor

Beyond the data center, Nvidia is making its most aggressive consumer PC push in years with the N1X, an ARM-based system-on-chip co-developed with MediaTek for Windows on ARM laptops. Nvidia, Microsoft, and Arm have all been teasing the reveal ahead of GTC Taipei .

CPU and GPU Configuration

The N1X uses a 20-core hybrid CPU arrangement split evenly between 10 Arm Cortex-X925 performance cores and 10 Cortex-A725 efficiency cores, all built on the Armv9.2 architecture . A leaked Geekbench entry showed a base frequency of 2.81 GHz, though boost clocks are expected to be considerably higher .

The integrated GPU is the standout specification. It packs 6,144 CUDA cores on Nvidia's Blackwell architecture—matching the raw core count of a desktop GeForce RTX 5070 . Early performance estimates from supply-chain sources suggest the iGPU could land in the performance class between a desktop RTX 4070 and RTX 5070, though real-world benchmarks remain unconfirmed .

Memory, Manufacturing, and Thermal Design

The N1X uses a unified memory architecture with support for up to 128 GB of LPDDR5X memory, sharing a single pool between the CPU and GPU . The chip is fabricated on TSMC's 3nm (N3) node in a 2.5D multi-die package: the CPU chiplet is designed by MediaTek, while the GPU chiplet is designed by Nvidia, connected by a 300 GB/s bidirectional NVLink C2C interconnect .

Nvidia has not officially published a TDP, but the N1X is designed for actively cooled gaming laptops. Industry reports peg the target SoC power envelope at roughly 25–45W, with headroom depending on chassis design .

Expected Availability

The N1X is expected to debut in gaming laptops from Dell Alienware and Lenovo Legion during the first half of 2026 . A lower-tier N1 variant with fewer cores is also planned . Nvidia is already roadmapping a next-generation N2 series for a 2027 launch .

AMD, Intel, and Qualcomm at Computex 2026

Information on the specific Computex plans for AMD, Intel, and Qualcomm was unavailable within the scope of the reporting gathered for this preview. All three companies have historically used Computex for major product keynotes and reveals, and interested readers should search directly for the latest on their 2026 announcements.