The AI factory stack takes shape: power, efficiency, and trust
DSX targets the power bottleneck behind $300B of backlogged equipment; cloud partners squeeze more GPUs per watt; factory blueprints and confidential computing handle design and trust.
- Power is now the biggest infrastructure bottleneck, with more than $300 billion in equipment backlogged and over 200 gigawatts waiting in US interconnection queues, per NVIDIA. Self-reported m:…4843
- Emerald AI's Conductor integration with DSX Flex is modeled to unlock up to 100 gigawatts on the existing US grid, per NVIDIA. Self-reported m:…4843
- Nscale demonstrated DSX MaxLPS fitting 37% more GPUs and nearly 50% higher aggregate inference throughput in the same provisioned power budget, per NVIDIA. Self-reported m:…5725
- NVIDIA released the Factory Operations Blueprint (FOX) in early access on GitHub; Audi is testing production concepts with a battery-production digital twin built on Omniverse libraries, per NVIDIA. Self-reported m:…4906
- NVIDIA introduced Confidential Bare Metal; Cohere plans to integrate it into its Model Vault, and Google Cloud is extending support to A4X Max instances on GB300 NVL72. Self-reported m:…7242
Power is the bottleneck
The constraint on AI infrastructure is no longer chips alone. Power is now the biggest bottleneck in infrastructure development, NVIDIA said, with more than $300 billion in equipment backlogged and over 200 gigawatts of projects waiting in US interconnection queues. The company's DSX AI Factory Platform is aimed squarely at that: DSX Sim lets engineers simulate the grid, substation, and AI factory together, and DSX Flex syncs factory demand with grid conditions and onsite generation — turning AI factories from static loads into flexible grid assets. m:…4843
The partner roster reads like an energy-industry roll call. Schneider Electric, AVEVA, and ETAP are advancing DSX with validated Vera Rubin power and cooling reference architectures; GE Vernova is deploying grid-to-facility and rack power expertise; Hitachi Energy's battery energy storage qualified as DSX Ready; Siemens Energy is developing grid-aware reference designs and evaluating NVIDIA technologies for its Noedra digital twin grid-monitoring platform. On the software side, Phaidra integrated DSX components into self-learning agents for about 10 percent more compute within the same power envelope, and Emerald AI's Conductor integration with DSX Flex is modeled to unlock up to 100 gigawatts on the existing US grid. m:…4843
Efficiency and factory blueprints
Inside the factory, efficiency is the lever. Nscale demonstrated that DSX MaxLPS enabled 37 percent more GPUs and nearly 50 percent higher aggregate inference throughput within the same provisioned power budget, NVIDIA said. Nebius adopted Groq 3 LPX for production inference integrated with Vera Rubin NVL72 on its Token Factory. m:…5725
For the factory floor itself, NVIDIA released the Factory Operations Blueprint (FOX) in early access on GitHub — a framework for deploying autonomous factory agents. Siemens is integrating FOX-built agents into Intelligence Center X, its agentic enterprise system, for an estimated 20 percent cost savings, and Audi is testing production concepts with a battery-production digital twin built on Omniverse libraries before physical equipment exists. m:…4906
The trust layer
Confidential computing rounds out the stack. NVIDIA introduced Confidential Bare Metal, giving frontier model builders hardware-rooted security for sensitive workloads on dedicated third-party infrastructure while keeping control of bespoke software stacks. The platform now offers three deployment options — bare metal, containers, and virtual machines — with support extending to Grace Blackwell NVL72 and x86-based HGX and DGX Rubin NVL8 systems. Cohere plans to integrate Confidential Bare Metal into its Model Vault; Google Cloud is extending support to A4X Max instances on GB300 NVL72. Ten cloud providers and six systems vendors — Cisco, Dell, HPE, Lenovo, Supermicro, and Bull — support the deployments, and DSX OS will support all confidential computing modes. m:…7242
What it means
This is my read, not a reported fact: NVIDIA is selling the factory, not the chip. The stack now spans grid simulation to confidential execution, which reframes capacity planning as a power-grid conversation rather than a GPU-procurement one. For AI executives, the vendors worth watching are those fluent in both languages — electrons and tokens. The open question is whether modeled flexibility numbers like Emerald AI's 100 gigawatts materialize in real utility interconnections, or stay in the demo layer. Either way, the bottleneck has moved upstream of the chip. m:…4843 m:…5725
What is still unresolved
- Whether modeled grid-flexibility gains survive real utility interconnection processes.
- Whether DSX MaxLPS efficiency gains hold outside Nscale's test configuration.
- Whether confidential computing overhead is acceptable for large-scale agent inference.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.