At GTC Berlin on October 21, 2026, Google Cloud and NVIDIA announced that general availability for A5X instances powered by NVIDIA Vera Rubin infrastructure is scheduled for next month. The companies also detailed new local AI hardware, software updates, and the general availability of the NVIDIA cuDSS simulation library.
Vera Rubin cloud infrastructure
Google Cloud announced plans to scale its Vera Rubin-based infrastructure to as many as 960,000 Rubin GPUs across multiple sites, with planned expansion of availability to Europe. The upcoming A5X instances integrate NVIDIA Vera Rubin NVL72, NVIDIA ConnectX-9 SuperNICs, Google Cloud Virgo Network, and Google Cloud's fourth-generation liquid-cooling technology.
Ineffable Intelligence was named the first customer to access A5X to support large-scale reinforcement learning, while WPP demonstrated a creative agent reference architecture on the platform using NVIDIA Nemotron models and NVIDIA OpenShell. In addition, Google Cloud announced it will offer NVIDIA Confidential Computing on bare-metal A4X Max instances powered by NVIDIA GB300 NVL72, alongside general availability on G4 VMs powered by NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs.
Local AI systems and developer hardware
NVIDIA detailed updates across its local hardware lineup, announcing that new DGX Spark 64GB systems will begin rolling out from hardware partners on Friday, October 23. DGX Spark now supports manual clustering of up to eight units to deliver 1TB of unified memory, and NVIDIA Sync will add Model Launcher later this month alongside an October software update that adds a Wi-Fi Assistant.
For enterprise workstations, DGX Station systems delivering 748GB of coherent memory and up to 20 petaflops of AI performance are shipping from partners, with DGX Station for Windows expected in December. Frontier AI lab Recursive Superintelligence also launched Recursive Flint, an open-weight 40-billion-parameter mixture-of-experts coding model with 2.5 billion active parameters optimized for RTX Spark and DGX Spark hardware.
cuDSS simulation library
NVIDIA also announced the general availability of cuDSS, a CUDA-X library designed to accelerate solving large sparse systems on GPUs during engineering and scientific simulations. The library scales from individual workstations to clusters and provides programming interfaces for C, C++, Python, and Fortran.
According to NVIDIA, software vendors are adopting the library: Applied Materials reported accelerating sparse solves in its Ginestra platform by up to 10x compared with CPU-only approaches, and Dassault Systèmes cited an 8-10x speedup in Abaqus solvers. Additional partners integrating cuDSS include Cadence, Siemens, Synopsys, and COMSOL with ZEISS SMT.
Key facts and where they come from
- Google Cloud plans to scale its Vera Rubin-based cloud infrastructure up to 960,000 Rubin GPUs across multiple sites.
Google Cloud plans to scale its Vera Rubin-based infrastructure to as many as 960,000 Rubin GPUs across multiple sites.
- General availability for Google Cloud A5X instances powered by Vera Rubin is scheduled for next month.
General availability of the instances is coming next month, with planned expansion of availability to Europe
- NVIDIA cuDSS launched in general availability to accelerate large sparse systems on GPUs for simulations.
NVIDIA cuDSS, launched today in general availability, is an NVIDIA CUDA-X library that shifts a critical part of engineering and scientific simulations onto GPUs.
- DGX Spark 64GB systems begin rolling out from partners on Friday, October 23.
New DGX Spark 64GB systems — that can run models including Nemotron 3.5 Bolt 74B 24GB, Qwen3.8 27B and Gemma 4 26B-A4B — start rolling out from partners on Friday, Oct. 23.
- DGX Station for Windows is scheduled to launch from system manufacturer partners in December.
DGX Station for Windows, expected from system manufacturer partners in December, extends local AI into the applications and operating environment enterprises already use.
- Recursive Superintelligence launched Recursive Flint, a 40-billion-parameter local mixture-of-experts coding model.
Today, frontier AI lab Recursive Superintelligence released Recursive Flint, a local LLM built for fast, accurate coding. A 40-billion-parameter mixture-of-experts model with 2.5 billion active parameters
