As AI data centres scale from thousands to hundreds of thousands of GPUs, a new constraint has emerged that has little to do with chip performance: power delivery and cooling capacity. NVIDIA has responded with DSX Ready, a qualification programme designed to certify that partner power and cooling products meet the requirements of its DSX AI factory reference architecture — giving builders a standardised way to choose compatible infrastructure at speed.
The Problem: Bottlenecks Move, They Don’t Disappear
NVIDIA’s own framing of the launch is instructive: “optimising one part of an AI factory can shift the bottleneck elsewhere.” As GPU racks have grown denser and more power-hungry, the systems around them — battery storage, coolant distribution, electrical infrastructure — have increasingly determined how much compute an operator can actually deploy, regardless of how many GPUs they can source. Builders have lacked a fast way to identify which third-party power and cooling products will reliably integrate with NVIDIA’s reference designs, while suppliers have lacked a clear, structured path to prove their products qualify.
DSX Ready addresses both sides of that gap at once: it gives infrastructure buyers a qualified shortlist, and gives component makers a defined certification route tied directly to NVIDIA’s own architecture standards.
Two Categories, Six Launch Partners
The programme launched with two initial product categories, chosen because they represent the most immediate constraints operators are running into today.
Battery Energy Storage Systems (BESS) smooth out power delivery to GPU clusters, which draw electricity in highly variable bursts depending on workload. NVIDIA’s first qualified BESS partners are Hitachi Energy, LG Energy Solution, and Tesla — an initial line-up that spans grid-scale energy specialists and one of the world’s largest battery manufacturers.
Cooling Distribution Units (CDUs) manage the liquid cooling loops that keep dense GPU racks within safe operating temperatures, a requirement that has become non-negotiable at the power densities modern AI training clusters now run at. NVIDIA’s qualified CDU partners at launch are LG Electronics, LiquidStack, and Vertiv — the latter a long-established name in data-centre thermal management.
Vertiv confirmed its own qualification separately, noting that its Coolant Distribution Unit has been certified as DSX Ready for AI factory infrastructure, underscoring how partners are treating the qualification as a market signal in its own right.
How Qualification Works
The two product categories are certified through different processes, reflecting how mature each testing discipline already is. BESS providers go through a formal qualification track: manufacturers run NVIDIA’s required tests and submit supporting data for NVIDIA’s review and approval, within a defined qualification boundary that maps to the DSX reference design.
CDU providers, by contrast, use a self-qualification suite that lets manufacturers determine for themselves whether a specific product meets NVIDIA’s functional requirements — a lighter-weight process that reflects the more standardised nature of liquid cooling hardware compared to the newer, more variable field of grid-scale battery storage for data centres.
Why This Matters Beyond the Data Centre Industry
For hyperscalers, cloud providers, and sovereign AI infrastructure projects, DSX Ready effectively shortens procurement cycles. Instead of independently testing whether a given cooling or battery product will work reliably alongside NVIDIA’s GPU platforms, buyers can filter directly for qualified partners — a meaningful advantage when AI factory build-outs are being measured in months, not years, and delays in permitting a single subsystem can hold back an entire facility’s go-live date.
It also signals where NVIDIA sees the next real constraint on AI’s growth curve. Chip supply has dominated headlines for the past several years, but as the industry pushes into gigawatt-scale AI factories, the ability to actually power and cool that compute reliably is becoming the more binding limit. By formalising a qualification standard, NVIDIA is positioning itself not just as a GPU supplier but as the reference point for how an entire AI factory — power, cooling, and compute together — should be engineered.
Expect the qualified partner list to expand well beyond the current six companies as more power and cooling vendors seek DSX Ready certification, particularly as AI infrastructure investment continues to accelerate across enterprise, hyperscale, and government-backed projects globally.
