NVIDIA's DSX Ready Certifies AI Factory Power and Cooling

NVIDIA has introduced DSX Ready, a qualification program that certifies third-party power and cooling hardware against the requirements of its DSX reference design for AI factories. The program launched with two categories: battery energy storage systems (BESS) and cooling distribution units (CDUs).

NVIDIA frames the need for the program bluntly: power, cooling, water, site, and grid constraints are shaping what builders can actually deploy, not compute availability. That is a striking reframing from the company whose GPUs are the reason hyperscale AI buildouts exist in the first place, and it signals that the industry's next bottleneck sits in the electrical room and the mechanical plant, not the server rack.

Why power and cooling became the real constraint

For most of the last two years, the AI infrastructure conversation centered on GPU supply: who could get allocation, and how many racks a given data center could fill with accelerators. DSX Ready quietly acknowledges that this framing is incomplete.

A rack of next-generation AI accelerators draws far more power and rejects far more heat than a rack of general-purpose servers, and a facility's ability to deliver stable power and remove that heat now gates how many racks can actually go live, independent of chip supply. Grid interconnection queues, water rights, and site selection have become the practical limits on AI factory scale-out. By naming BESS and CDUs as its first two qualification categories, NVIDIA is pointing directly at the two components that sit closest to that bottleneck: energy storage that buffers grid variability and demand spikes, and cooling distribution that moves heat away from increasingly dense compute.

An ecosystem play, not a hardware business

NVIDIA is not building its own batteries or cooling distribution units. Instead, it is publishing reference design requirements and running a qualification process that lets established power and cooling vendors prove compliance.

That is a deliberate strategic choice: it lets NVIDIA extend its influence over how AI factories are built without taking on the manufacturing, warranty, and field-service burden of BESS and CDU hardware itself. It also gives NVIDIA a mechanism to make its own DSX reference design the de facto specification that power and cooling vendors build toward, the same way its GPU and networking reference designs became the template that data center builders followed. Builders get a shorter list of products they can trust to work with a DSX-based deployment, which reduces integration risk and shortens the path from reference design to a working AI factory.

Who is qualified, and what qualification requires

At launch, three vendors are DSX Ready certified for BESS: Hitachi Energy, LG Energy Solution, and Tesla. Three more are certified for CDUs: LG Electronics, LiquidStack, and Vertiv. The two categories use different qualification paths.

BESS providers submit test data to NVIDIA for review, meaning NVIDIA itself checks the results against its reference requirements. CDU providers instead use a self-qualification test suite to demonstrate compliance with functional requirements, a lighter-touch process that still ties their products to the same DSX specification. For both groups, DSX Ready status becomes a marketing and sales asset: a way to signal to hyperscalers and enterprise AI-factory builders that a given battery or cooling unit was built to specifications NVIDIA itself defined, rather than requiring a builder's engineering team to independently validate compatibility from scratch.

The disclaimer is the most important sentence in the announcement

NVIDIA states explicitly that DSX Ready qualification does not replace site-level engineering or imply site-level stability. That caveat matters more than it might first appear. A CDU or BESS unit can meet every DSX reference requirement in isolation and still behave differently once it is integrated into a specific facility's power topology, water loop, redundancy scheme, or climate.

NVIDIA is drawing a clear line between component-level qualification, which it is now willing to certify, and system-level performance, which remains the responsibility of the builder's own site engineers. That distinction protects NVIDIA from being on the hook for outcomes it cannot control, while still letting it claim credit for reducing the integration risk that qualification does address. For anyone evaluating DSX Ready as a signal of reliability, the honest reading is that it narrows the search for compatible components without replacing the engineering work of putting them together.