Nvidia Highlights DSX Pods, Robotics, and Supply Chain Expansion at GTC Europe

Nvidia outlines faster AI datacenter deployment

Jensen Huang opened GTC Europe by saying Nvidia aims to reduce the time required to bring new datacenters online from several months to a few weeks. The company presented a Vera Rubin-powered AI factory packaged in a transportable system as part of that effort.

The main hardware announcement was the DSX Pod, a containerized implementation of Nvidia’s DSX reference architecture. Huang’s keynote focused on Europe’s role in the AI era and highlighted partners including Schneider Electric and Siemens Electric, alongside updates to Nvidia’s existing platforms.

Huang described the pod as “bringing this full-stack platform into a factory-built integrated system.” Nvidia said the design starts with Vera Rubin NVL72. Seven liquid-cooled compute racks are connected by 17 kilometers of Spectrum X cables to form a Vera Rubin DSX Pod. The pods are integrated, fully cabled, factory-tested, and delivered in transportable containers.

The system uses dry cooling rather than traditional evaporative cooling, along with 800V DC power delivery. These features are part of the broader DSX architecture, but Nvidia is positioning the factory-built format as a way to address datacenter construction and supply chain constraints.

Huang said the scale of current computing deployments is unprecedented. Even after the systems are delivered to a site, he noted, months of assembly may still be required, including the installation of thousands of kilometers of cabling for large-scale computing systems.

Datacenter projects can face labor shortages before they encounter shortages of GPUs or DRAM. At TeraWulf’s Lake Mariner site outside Buffalo, New York, nearly half of the 1,800 tradespeople working on the project are electricians.

Nvidia’s approach is to complete more of the assembly within the supply chain. Systems could be built, tested, qualified, and burned in before shipping, reducing both onsite cabling and installation work. Huang also said this could reduce the amount of computing capacity sitting idle in datacenters while construction and commissioning are completed.

Huang estimated that a 1GW datacenter costs approximately $50 billion and could generate about $40 billion in revenue. A three-month delay, he said, could mean $10 billion in missed revenue. Nvidia’s goal is to reduce the time from the arrival and installation of computing equipment to token production from roughly four or five months to about one month.

The DSX Pod presentation showed Nvidia’s green branding alongside partner logos, including Schneider Electric and Vertiv, as well as server manufacturers HPE, Lenovo, and Dell. Nvidia said it is working with these and other companies to deliver DSX Pod AI factories worldwide in early 2027.

Isaac robotics platform receives updates

Nvidia also presented updates to its Isaac robotics platform, including the Agile One SN reference humanoid robot. The robot is 1.4 meters tall and has five-finger hands designed for tactile sensing and dexterous manipulation. It includes multimodal sensing, 2,070 FP4 teraflops of onboard compute, and 128GB of unified memory.

Huang said the reference robot is expected to become available in the first quarter of the following year. It is planned to ship pre-assembled, with its computer installed and software integrated, so it can perform with minimal setup.

Nvidia expands manufacturing partnerships

Nvidia announced plans to work with a group of supply chain partners to expand global manufacturing capacity for its accelerated computing products and full-stack AI factory platform.

The companies named were:

• Amkor Technology
• IBIDEN
• King Yuan Electronics (KYEC)
• Samsung Electro-Mechanics
• Siliconware Precision Industries (SPIL)
• Unimicron Technology Corporation

The stated objective is to expand manufacturing capabilities across substrates, advanced packaging, testing, and optical engines while increasing the resilience and redundancy of Nvidia’s global supply chain.

Samsung Electro-Mechanics is joining Nvidia’s substrate supply chain, using capacity from its existing operations in South Korea. Nvidia also expects its total semiconductor testing capacity to double between calendar years 2026 and 2027, supported by facilities in the United States, Taiwan, and Singapore.

The agreement with SPIL will expand manufacturing capacity in Taiwan for optical engines used in co-packaged optics. Nvidia said the expansion targets a 15-fold increase by the middle of 2027, with another doubling planned by the end of 2028.

Huang revisits Moore’s law

Huang again argued that Moore’s law no longer describes the pace of AI performance improvements. He said AI performance had improved by 1 million times over the previous 10 years. By comparison, he estimated that Moore’s law would have produced a 100-fold improvement over the same period, with a tenfold increase in each five-year interval.

He also cited European companies using Nvidia’s DSX reference designs, which rely on liquid cooling and dry coolers, to reduce water and power consumption and help stabilize electrical grids.

The datacenter industry is facing opposition across Europe because of environmental and sovereignty concerns. Activists disrupted the early part of the keynote, reportedly including people associated with Greenpeace. They shouted about “environmental destruction” and reportedly referred to reports concerning Nvidia technology being used in Russian drones in Ukraine.

Huang remained composed during the interruption. He asked security to “go easy” and give one protester a chance, then said he would discuss the issues later while asking to continue with the keynote.