英伟达发布DSX Pod集装箱模块，称14天可部署AIDC

3 Min Read

Nvidia DSX Pod product photo

The Nvidia DSX Pod.Nvidia

Nvidia today introduced DSX Pod, a containerized AI data center module that comes with seven Vera Rubin NVL72 racks, along with networking, cabling, and cooling already installed. The company says the design can cut deployment time at a prepared site from roughly six months to as little as 14 days.

Announced at GTC Berlin, the modular offering is aimed at organizations that need to bring large amounts of AI compute online more quickly without going through the full planning, construction, and labor demands of a traditional on-site build. Nvidia said it is working with 17 partners, including Bull, Dell Technologies, HPE, Lenovo, Eaton, Schneider Electric, and Vertiv, to make DSX Pod available in early 2027.

“This is a fully assembled, fully tested, fully pre-plumbed, prebuilt, pre-validated container that is the building block for AI factories,” said Ian Buck, Nvidia’s vice president of hyperscale and high-performance computing, in a media briefing. “It’s designed to accelerate Europe’s infrastructure deployment and dramatically shorten the time to first token.”

Although Nvidia unveiled the reference design in Berlin, Buck said the container is intended for worldwide deployment. He said the company chose GTC Berlin for the announcement because it sees modular, faster-to-deploy infrastructure playing an important role in Europe’s expansion of AI capacity.

Buck said the product is designed for organizations that want to bring compute online at the hundreds-of-megawatts scale more quickly while reducing the complexity of a conventional site build.

Ashish Nadkarni, an IDC analyst, said the DSX Pod could be appealing for three use cases: sovereign AI deployments that require physical isolation, smaller-footprint AI deployments built for specific needs, and edge AI environments.

“It provides accelerated compute capacity in a turnkey fashion wherever there is demand,” Nadkarni said.

Manufacturing Expansion Targets AI Supply Bottlenecks

The DSX Pod announcement came alongside a broader Nvidia update on manufacturing expansion. To support the growing demand for AI infrastructure, Nvidia and its supply chain partners said they plan to increase substrate capacity by 2.5 times over the next three years, expand advanced packaging capacity by more than 300% by 2028, and increase optical engine capacity for co-packaged optics by 15 times by mid-2027, with a goal of doubling that again by the end of 2028.

OpenShell Adds Governance for AI at Scale

Nvidia also used the event to highlight software and security controls for AI deployments at scale. The company introduced the OpenShell Reference Architecture, which it said is designed to help organizations govern AI agents by defining what they can access and do, monitoring their behavior, and enforcing policies if unsafe activity is detected.

Justin Boitano, Nvidia’s vice president of enterprise, said the architecture combines Vera Rubin NVL72 infrastructure for training and inference with Nvidia Vera systems that run and govern the “agent harness.” BlueField DPUs act as an isolated security monitor between the two systems, inspecting traffic and enabling policy updates that can disconnect an agent if unsafe behavior is detected.

Nvidia said companies using the OpenShell Reference Architecture to strengthen agent security include Anthropic, AWS, Google Cloud, Meta, Microsoft, OpenAI, and Perplexity.

About the Author

Chips and Hardware Writer

Wylie Wong is a journalist and freelance writer specializing in technology, business and sports. He previously worked at CNET, Computerworld and CRN and loves covering and learning about the advances and ever-changing dynamics of the technology industry. On the sports front, Wylie is co-author of Giants: Where Have You Gone, a where-are-they-now book on former San Francisco Giants. He previously launched and wrote a Giants blog for the San Jose Mercury News, and in recent years, has enjoyed writing about the intersection of technology and sports.