NVIDIA starts certifying the power and cooling side of AI factories: DSX Ready launches with batteries and cooling units only, and passing is not site stability

2026-10-26·12 min read

On the surface this is supply chain housekeeping. In reality it marks a turn: the AI compute bottleneck has moved from can you buy the GPUs to is there enough power and can you carry the heat away. NVIDIA is no longer only selling chips and racks; it is issuing qualification labels for the least glamorous, most engineering-heavy links, because a failure there costs the output of an entire factory.

Pin the source first. The primary material is an NVIDIA blog post published October 21, 2026 titled NVIDIA Launches DSX Ready to Qualify Power and Cooling Products for AI Factories, part of the company's infrastructure announcements around GTC Berlin. Every definition, process step and caveat below comes from that post, not from aggregator rewrites.

What DSX Ready actually is. It is not a product and not a standards-body certification but a qualification program: partner products that meet applicable NVIDIA DSX AI factory reference design requirements can go through it and earn a qualification, giving builders a clear label to look for when selecting equipment. NVIDIA frames the value as confidence when evaluating offerings, reduced integration risk and a faster path to deployment.

Why now. NVIDIA's reasoning is plain but heavy: as AI infrastructure expands, constraints on power, cooling, water, site and grid are shaping what builders can deploy, and choosing products that fit the whole factory design is what turns computing capacity into useful AI output. Buying GPUs is not the same as producing tokens; power and heat stand in between.

That only two categories open at launch matters. One is battery energy storage systems, the other cooling distribution units. NVIDIA says these initial categories bring power systems and liquid-cooling infrastructure into one program with requirements tailored to each technology, and that additional categories across infrastructure and software will roll out over time. No GPUs, no switches, no racks: NVIDIA is tightening the two items most likely to fail at site level first.

How the BESS path works. Providers run the required qualification tests themselves, then submit supporting data for NVIDIA review and approval within a defined qualification boundary. That phrase deserves a pause, because it means the qualification is scoped: what passed is this product's behaviour inside that boundary, not a claim that it holds in any factory you drop it into.

The CDU path is different. Cooling distribution units go through the CDU self-qualification suite, which determines whether a specific offering meets applicable NVIDIA functional requirements. Self-qualification pushes execution and declaration onto the provider, with NVIDIA defining requirements and supplying the suite rather than retesting each unit, and that difference between the two paths reflects how differently mature the storage and liquid-cooling segments are.

The caveat is the thing to remember. NVIDIA writes that passing qualification does not replace site-level engineering and does not imply site-level stability, and adds in the same breath that a qualified CDU may meet the relevant cooling criteria while the builder still has to evaluate how it will fit the planned facility. It is the most honest passage in the post and the one procurement teams skim past: the qualification covers a product, not your site.

Put it back in the DSX picture. The DSX AI factory platform unifies compute, networking, power, cooling, facilities and software so partners can design and operate the factory as one system, producing more useful AI output within available power, cooling, water and grid constraints. NVIDIA also makes a key point: optimising one part of a factory shifts the bottleneck elsewhere, which is why qualification has to be systematic rather than a series of local wins.

What it changes for buyers and builders. Selecting BESS and CDU used to rest on vendor datasheets and project experience, with no common comparable dimension across projects; now there is at least one more filter, whether an offering has gone through DSX reference design qualification. Keep in mind it is a starting condition, not a finish line: site engineering, grid interconnection approvals, water consumption assessment and maintenance windows all remain. Anyone treating the label as acceptance criteria will pay tuition during construction.

How it connects to NVIDIA's other moves. Within days the company also released Minerva for end-to-end observability of AI factories, collapsing alert storms into a single root cause, while DSX Ready governs the supply chain entrance. Together they express NVIDIA's definition of an AI factory: not just selling compute but putting power, cooling, monitoring and operations into one engineering language. For data center integrators that is both an opening and a barrier.

Where the boundaries are. First, DSX Ready covers only BESS and CDU today, with additional categories promised over time and no schedule. Second, NVIDIA published no list of qualified products, no test specifications and no pass thresholds, so an outsider cannot verify whether a specific product qualifies. Third, this article records only what was officially published through October 26, 2026, and forecasts neither which vendors will qualify nor whether the label becomes a de facto requirement.

🤔 Frequently Asked Questions

Is DSX Ready a certificate for power and cooling products?

Roughly, but scope matters. It is a qualification program that checks whether a product meets applicable NVIDIA DSX AI factory reference design requirements, starting with battery energy storage systems and cooling distribution units. NVIDIA states explicitly that passing does not replace site-level engineering and does not imply site-level stability.

Why do BESS and CDU use different processes?

NVIDIA set two paths. BESS providers run the required qualification tests and submit supporting data for review and approval inside a defined qualification boundary, while CDU providers use the self-qualification suite to determine whether an offering meets applicable functional requirements. The first leans on NVIDIA review, the second on provider execution and declaration, which reflects how differently standardised the two device classes are.

Does qualification mean the AI factory will run reliably?

No, and NVIDIA says so directly. Passing qualification does not replace site-level engineering and does not imply site-level stability, and a qualified CDU may meet cooling criteria while the builder still evaluates how it fits the planned facility. Qualification covers the product, site engineering covers deployment, and neither substitutes for the other.

What does this mean for data center integrators?

One more selection dimension, and a new bar. BESS and CDU move from datasheet comparison to whether an offering has gone through DSX reference design qualification, while qualification boundaries, site engineering and grid approvals all remain mandatory. Since NVIDIA published neither a qualified product list nor test specifications, the practical move for now is to use DSX reference design requirements as a checklist rather than treating the label as acceptance criteria.

🛠️ Recommended Tools

  • AI ROI Calculator because the stated goal is turning computing capacity into useful AI output, and that conversion belongs in the business case before procurement.
  • YAML Formatter since building a selection checklist against DSX reference requirements works better as structured config than as scattered spreadsheets.
  • Docker Compose Generator and standing up monitoring and ops locally before touching racks is the cheapest way to validate an observability plan.
  • GraphRAG Builder because once compute stops being the limit, retrieval quality on the data side becomes the bottleneck for useful output.

Summary

NVIDIA published NVIDIA DSX Ready on October 21, 2026, a qualification program for partner products that meet applicable NVIDIA DSX AI factory reference design requirements. Two categories opened at launch, battery energy storage systems and cooling distribution units, with additional infrastructure and software categories promised over time and no schedule given. The BESS path has providers run the required qualification tests and submit supporting data for NVIDIA review and approval within a defined qualification boundary; the CDU path uses the CDU self-qualification suite to determine whether a specific offering meets applicable functional requirements. NVIDIA states explicitly that passing qualification does not replace site-level engineering and does not imply site-level stability, and that a qualified CDU still requires the builder to evaluate how it fits the planned facility. The stated context is that as AI infrastructure expands, power, cooling, water, site and grid constraints shape what builders can deploy, and that optimising one part of a factory shifts the bottleneck elsewhere, which is why the DSX platform treats compute, networking, power, cooling, facilities and software as one system. No qualified product list, test specification or pricing was published.

Sources: NVIDIA 官方博客:NVIDIA Launches DSX Ready to Qualify Power and Cooling Products for AI Factories(2026-10-21)
NVIDIA 官方:GTC Berlin 2026 发布汇总