NVIDIA Launches DSX Ready: A Seal of Approval for the Batteries and Cooling Gear That Keep AI Factories Alive

For two years the AI infrastructure conversation has been dominated by one question: which GPU? But anyone actually building an AI factory knows the hard questions have quietly moved somewhere else. How many megawatts can the grid deliver to this site, and when? Where does the heat go when every rack draws as much power as a small neighborhood? What happens when a pump fails at 3 a.m. in a facility that cost eleven figures to build?

Today NVIDIA took a concrete step toward standardizing answers to those questions. The company launched NVIDIA DSX Ready, a qualification program that identifies third-party power and cooling products that meet the applicable requirements of its DSX AI factory reference designs. The program debuts with two categories: battery energy storage systems (BESS) and coolant distribution units (CDUs) — the two pieces of physical infrastructure that most directly determine whether a gigawatt-class AI campus actually functions.

What DSX Ready actually is

DSX is NVIDIA’s umbrella platform for designing and operating “AI factories” as a single system, spanning compute, networking, power, cooling, facilities and software. A reference architecture only gets you so far, though. When a builder following a DSX design goes shopping for batteries or cooling hardware, they face a market full of products with wildly different specs, control behaviors and integration profiles. Reference designs don’t pick vendors; somebody still has to decide which 2-megawatt cooling unit actually plays well with the rest of the factory.

DSX Ready is NVIDIA’s answer to that gap. Products that pass the program’s category-specific review get designated as qualified offerings within the DSX ecosystem, giving builders a filter for procurement and giving suppliers a defined path to demonstrate compatibility. NVIDIA frames the benefit in three parts: evaluate offerings with greater confidence, reduce integration risk, and move faster toward deployment.

It is worth being precise about what the qualification is not. NVIDIA’s own materials stress that passing DSX Ready does not replace site-level engineering or imply site-level stability. A qualified CDU meets functional requirements; the builder still has to work out whether it fits their facility’s hydraulics, layout and operating profile. The program standardizes the building blocks, not the building.

The first names on the list

At launch, DSX Ready includes qualified BESS solutions from Hitachi Energy, LG Energy Solution and Tesla, and qualified CDU solutions from LG Electronics, LiquidStack and Vertiv.

The presence of Tesla on a qualified-storage list is notable — the same company building AI factories for its own workloads is now formally in the supply chain for everyone else’s. Vertiv, meanwhile, announced that its 2.3 MW coolant distribution unit is the first CDU to qualify specifically as DSX Ready, positioning it among the initial providers of qualified cooling infrastructure. LiquidStack, a Trane Technologies company, revealed that its GigaModular CDU platform earned the designation and, importantly, that it collaborated with NVIDIA during the development of the CDU requirements themselves — meaning at least some of the qualification criteria were shaped by vendors with products already in the pipeline.

The two categories were not chosen at random. They map exactly onto the two bottlenecks the industry keeps running into: power availability and heat rejection. Battery storage lets an AI factory smooth its draw against grid constraints, participate in demand-response programs, and bridge the gap between utility interconnects that arrive years late. Liquid cooling is no longer optional at modern rack densities — direct-to-chip cooling with centralized CDUs is the default architecture for NVIDIA’s own reference designs, and the CDU has become the component whose failure modes matter most.

What a CDU has to prove

The LiquidStack announcement includes the most detailed public description so far of what the CDU qualification actually tests. The requirements cover hydraulic, thermal and control performance, including:

• Constant differential-pressure and constant-flow performance — the unit must hold steady hydraulic behavior as racks come and go from the loop
• Thermal performance under specified operating conditions
• Flow-sensor accuracy, so the factory’s telemetry can be trusted
• Cold-start behavior — how the unit behaves when brought up from a cold state
• Pump failover — graceful degradation when a pump dies
• Group control — multiple CDUs coordinating as one cooling plant rather than a shelf of independent boxes

That last item gets at the philosophy behind the whole program. LiquidStack’s own framing is that “AI factory performance does not begin and end with the GPU” — a CDU cannot be judged by its headline cooling capacity alone, because its controls, materials and response to changing conditions determine whether the wider system stays stable. Group control requirements push vendors toward the coordinated, multi-megawatt deployments that actual AI campuses are becoming.

On the BESS side, the qualification process works differently: partners run the required tests themselves and submit supporting data for NVIDIA review within a defined qualification boundary. More categories across infrastructure and software will roll out over time.

The bigger picture: NVIDIA keeps annexing the stack

Seen from a distance, DSX Ready is one more move in a strategy NVIDIA has been executing all year. First the company defined the AI factory concept. Then it published reference designs. Then it showed production results — earlier this month at the AI Infra Summit, NVIDIA reported that DSX power-allocation software let cloud provider Lambda squeeze 24% more token throughput from a fixed power budget. Now it is extending its authority outward from compute into the physical plant: batteries, pumps, heat exchangers.

Each step converts what used to be a builder’s judgment call into an NVIDIA-blessed decision. For data center operators, that cuts both ways. A qualification program genuinely reduces integration risk in a market where every vendor claims compatibility with everything. It also deepens NVIDIA’s gravitational pull over a supply chain it doesn’t manufacture — a seal of approval from the dominant AI compute vendor is close to a procurement requirement for anyone building DSX-aligned factories, whether or not NVIDIA ever touches the hardware.

The infrastructure incumbents seem to have decided that joining is better than fighting. Vertiv, Trane (via LiquidStack), Eaton and their peers are lining up to qualify products against requirements their own engineers helped shape. That pattern — competitors collaborating on the dominant platform’s terms — is how ecosystems get locked in.

For an industry racing to build factories that consume as much power as small cities, some standardization is overdue. The question worth watching is whether DSX Ready categories keep expanding until “qualified” becomes effectively mandatory for anyone selling into AI factories. Based on the trajectory of everything else NVIDIA has touched this year, the safe bet is yes.

Sources

• NVIDIA Blog — NVIDIA Launches DSX Ready to Qualify Power and Cooling Products for AI Factories
• LiquidStack — Cooling the AI Factory as One System
• Vertiv — Coolant Distribution Unit qualified as NVIDIA DSX Ready
• NVIDIA — DSX Ready program page