NVIDIA BESS Qualification Program Narrows AI Data Center Batteries to Three Suppliers

NVIDIA launched its first BESS qualification program with only three approved suppliers, turning data center batteries into another tightly specified part of its AI infrastructure stack.

The initial NVIDIA BESS qualification program includes Tesla, LG Energy Solution, and Hitachi Energy. Their listed systems vary considerably in output and capacity. However, they passed a common review focused on how battery systems respond to demanding electrical conditions around AI computing campuses.

That short list creates the real tension. NVIDIA is giving builders a clearer procurement path, but qualification is not certification of an entire data center. The company explicitly warns that passing its process does not establish site-level stability.

The program also extends NVIDIA’s influence beyond GPUs, networking, and rack systems. Its reference designs now reach into power conversion, grid interaction, cooling, facility controls, and operational software.

For data center developers, this offers a more coordinated design framework. For battery vendors outside the initial group, it creates pressure to prove that their controls can satisfy NVIDIA’s requirements.

NVIDIA BESS Qualification Program Starts With a Short List

NVIDIA has moved battery storage from optional facility equipment into its formal AI factory qualification system.

The company introduced DSX Ready on September 21, 2026. DSX Ready evaluates partner products against applicable requirements in NVIDIA’s AI factory reference designs.

The program begins with battery energy storage systems, commonly called BESS, and cooling distribution units. NVIDIA says more infrastructure and software categories will follow.

A BESS combines batteries, bidirectional power conversion equipment, controls, and protection functions. It can store electricity and rapidly adjust active or reactive power when grid or facility conditions change.

NVIDIA’s initial qualified battery list contains three products:

• Hitachi Energy’s system combines its WD4 power conversion system with battery storage. The listed configuration provides 5 MW of AC output and 10 MWh of capacity.

• LG Energy Solution’s Vertech JF2 AC Link supports two-hour or four-hour configurations. The listed system provides 5 MW of AC output and 10 MWh of capacity.

• Tesla’s Megapack 2XL provides 1.93 MW of AC output and 3.85 MWh of capacity in the listed configuration.

The list appeared through NVIDIA’s launch materials and was detailed in the initial product list. NVIDIA also qualified cooling systems from LG Electronics, LiquidStack, and Vertiv under the program’s separate cooling category.

The battery choices do not represent a ranking of cell chemistry, energy density, or project economics. Instead, NVIDIA evaluates the power conversion system at its AC connection point.

That boundary matters. It places the focus on whether a storage system can behave predictably as a grid-connected power asset under difficult operating conditions.

The qualified system must respond to active and reactive power commands. It must handle current limits, voltage disturbances, islanded operation, black starts, and transitions between operating modes.

Partners perform the prescribed tests and submit the results to NVIDIA. Required evidence includes raw time-series data, test configurations, summary plots, model files, and explicit pass-or-fail statements.

NVIDIA then reviews the submission against its requirements. This differs from a simple compatibility badge based primarily on vendor declarations or component specifications.

The approach gives developers a comparable technical evidence package across participating suppliers. It also gives NVIDIA a mechanism for shaping how batteries interact with its broader data center architecture.

The first three suppliers already operate in large-scale energy storage. Their presence is less surprising than the fact that NVIDIA now defines a common entry gate.

That gate can become more consequential as AI campuses require faster grid interconnections, steadier load profiles, and closer coordination among batteries, generators, utilities, and computing systems.

Why AI Data Centers Need Batteries That Behave Like Control Systems

The central problem is not simply finding enough electricity. AI campuses must also control sharp and unpredictable changes in demand.

Traditional data centers operate variable workloads, but large AI clusters introduce another level of electrical concentration. Thousands of accelerators can change operating states in coordinated patterns.

Those changes create rapid power ramps. A facility’s computing load can rise or fall more quickly than utility equipment, generators, and mechanical systems prefer.

NVIDIA describes battery storage as a grid-interactive control resource for this environment. The battery is not limited to providing energy during a prolonged outage.

Instead, it can buffer short load changes, support demand response, maintain operation during voltage disturbances, and assist transitions to onsite generation.

NVIDIA’s BESS design guidance frames the challenge as one of control, quality, and interconnection. That is a wider role than conventional emergency backup.

Load buffering is one example. A battery can respond when AI equipment suddenly requests more power, reducing the ramp seen by the utility connection.

The same battery can absorb energy when computing demand falls. This reduces abrupt swings at the facility’s point of interconnection.

Demand response adds a longer operating window. A data center can temporarily draw from stored energy when the grid requests a reduction in demand.

Low-voltage ride-through serves a different purpose. It requires equipment to remain connected and behave safely when grid voltage drops within specified limits.

Without suitable controls, a disturbance can cause power equipment to disconnect at the worst moment. Coordinated disconnection across a large campus could amplify the original event.

Islanded operation presents another test. A site may need to operate independently from the grid while batteries coordinate with generators and facility controls.

Black-start capability addresses the restoration process. A qualified system must show that it can energize a de-energized electrical bus and establish stable voltage and frequency.

These use cases make the power conversion system as important as the stored energy. The inverter and its control software determine how quickly and predictably the battery responds.

NVIDIA’s process therefore treats the BESS as an active grid asset. It does not judge qualification primarily through battery capacity or cell chemistry.

That distinction explains why products with different capacities can appear on the same list. Their configurations need not be identical to satisfy a behavior-focused framework.

The timing also reflects NVIDIA’s expanding facility ambitions. Its DSX platform coordinates compute, networking, storage, power, cooling, controls, and operational software.

NVIDIA’s current reference design describes a 250 MW-class example site with 1,536 GPU racks. It includes a 34.5 kV backbone connecting batteries and optional onsite generation.

That facility reference design is a sizing example, not a universal prescription. Still, it shows the electrical scale NVIDIA expects infrastructure partners to address.

At that scale, a battery cannot remain an isolated box selected late in construction. Its controls, models, telemetry, and protection behavior must fit the campus before equipment arrives.

NVIDIA Is Setting the Gate, While Suppliers Run the Tests

The primary conflict is between NVIDIA’s push for standardized components and the site-specific engineering that still determines whether a project works.

NVIDIA describes DSX Ready as a qualification program, not an independent safety certification. The difference becomes clearer inside its battery guidelines.

Suppliers run the required tests themselves. They prepare the evidence package and submit it for NVIDIA’s review and approval.

This structure can accelerate participation because vendors do not need NVIDIA engineers to conduct every physical test. It also places substantial responsibility on the submitting company.

The required package is detailed. Every test must include raw time-series data, summary plots, operating configurations, controller settings, and a pass-or-fail conclusion.

Suppliers must identify known issues and deviations. They must also submit models that reproduce hardware behavior within the stated tolerances.

NVIDIA requests electromagnetic transient models, control blocks, impedance plots across multiple operating points, and version-controlled parameter exports.

These materials help engineering teams simulate how equipment will behave inside a wider electrical network. They also provide more information than a conventional product data sheet.

The submission requirements additionally call for relevant industrial standards and regulatory certificates. Examples include grid, fire-safety, quality, and cybersecurity frameworks.

However, NVIDIA does not replace the organizations responsible for those standards. Its program does not make the company a fire-safety authority, utility regulator, or electrical inspector.

The qualification boundary ends at the BESS AC terminals, including the power conversion system. Site transformers, relays, switchgear, generators, and campus controls remain outside that boundary.

Battery sizing is also excluded from the qualification decision. Developers must determine the required duration and capacity for each site’s objectives.

Cell chemistry is another excluded area. Chemistry influences cost, lifespan, thermal behavior, energy density, and supply strategy, but it does not decide this qualification.

NVIDIA instead tests whether the complete system responds correctly at the connection point. That makes the process chemistry-neutral by design.

For buyers, this creates a useful starting point. A DSX Ready listing indicates that a vendor submitted evidence addressing defined AI factory operating conditions.

It does not mean the product can be inserted into any project without additional studies. The data center’s electrical topology can materially change system behavior.

A transformer changes impedance. Protection settings determine when equipment disconnects. Generators introduce their own control characteristics and response times.

Utility requirements vary across markets. Local authorities can impose different fire, permitting, seismic, environmental, and interconnection rules.

Even the same battery may require different settings at two sites. A stable configuration on a strong grid can behave differently on a weak grid.

This is why NVIDIA states that qualification does not imply site-level stability. It also says a failure within its defined operating envelope still results in non-qualification.

The program therefore creates a floor rather than a complete deployment guarantee. It filters products before project-specific engineering begins.

That balance makes the NVIDIA BESS qualification program useful, but narrower than the word “qualified” might initially suggest.

Tesla, LG, and Hitachi Gain an Early Procurement Advantage

The first three suppliers receive more than recognition because their products now appear inside NVIDIA’s preferred infrastructure path.

Tesla brings a standardized, factory-produced storage platform through Megapack. The company has positioned the product for utility, commercial, and large infrastructure projects.

Its listed Megapack 2XL configuration is smaller than the two other launch configurations. Multiple units can be combined when a project requires greater output or duration.

Tesla made no separate public statement in the initial report. That silence does not diminish its inclusion, but it leaves fewer details about its DSX strategy.

LG Energy Solution enters through its Vertech integration business and JF2 AC Link product. The company says the system uses high-capacity lithium iron phosphate technology.

LG’s launch announcement says the product satisfied NVIDIA’s requirements for AC energy storage solutions. Its listed configuration provides 5 MW and 10 MWh.

Chang Beom Kang, head of LG Energy Solution’s ESS battery division, called the selection validation of the company’s storage products. That remains a supplier statement, not independent proof of project performance.

The company’s qualification announcement also emphasizes local manufacturing. Domestic production can matter for delivery schedules, policy eligibility, procurement rules, and supply-chain planning.

Hitachi Energy combines a battery system with its WD4 power conversion technology. It also lists a 5 MW, 10 MWh configuration.

The company has extensive experience in grid integration, substations, power quality, and high-voltage equipment. Those capabilities align closely with the problems NVIDIA’s requirements address.

Marco Berardi, Hitachi Energy’s head of grid and power quality solutions and service, emphasized deployment speed, integration, reliability, and expansion.

Hitachi Energy’s DSX Ready statement presents the qualification as part of a broader AI factory electrification offering.

The advantage for all three suppliers is early visibility. Infrastructure teams following NVIDIA’s reference designs now have an initial set of reviewed battery options.

That can shorten preliminary vendor screening. It can also reduce perceived integration risk during early project planning.

The qualification does not guarantee a contract. Developers will still compare delivery capacity, warranties, service coverage, manufacturing locations, financing, safety records, and project economics.

Yet appearing on the initial list can influence which vendors enter those comparisons. Project teams frequently begin with equipment already documented inside a preferred architecture.

Vendors absent from the launch list face a different challenge. They must decide whether to complete NVIDIA’s testing process or compete outside the DSX pathway.

Major storage suppliers such as Fluence, Wärtsilä, Sungrow, BYD, CATL, Samsung SDI, and others are not named in the initial BESS list.

Their absence does not establish that their products failed. It can reflect timing, participation decisions, incomplete submissions, or other factors that NVIDIA has not publicly detailed.

That distinction is essential. The launch list identifies successful participants, not a comprehensive judgment on every competing product.

Still, NVIDIA has created an incentive to participate. A supplier seeking AI data center projects now has a visible qualification target tied to a major computing platform.

The competitive question is no longer limited to who can supply battery capacity. Vendors must show that their power electronics and controls fit AI-oriented electrical operating envelopes.

What NVIDIA’s Qualification Does Not Prove

A three-vendor list can simplify procurement, but it can also create more confidence than the underlying process supports.

The clearest limitation comes from NVIDIA itself. Passing qualification does not establish stability for a complete data center site.

That caveat covers several major engineering risks. The reviewed battery ends at its AC terminals, while real projects contain many interacting systems beyond that point.

A developer must integrate transformers, switchgear, protective relays, generators, utility controls, and building management systems. Each component can change the overall response.

The test data also comes from the supplier. NVIDIA reviews it, but the public materials do not describe routine independent laboratory testing for every listed configuration.

That does not make the evidence invalid. It does mean buyers should understand the program’s partner-run structure before treating the badge as independent certification.

Transparency presents another open question. NVIDIA specifies extensive data for its review, yet most raw submissions are not available publicly.

External engineers therefore cannot compare the response curves, model quality, deviations, or margins behind each product’s pass result.

The three systems also have different listed power and energy capacities. Qualification confirms behavior within defined conditions, not direct interchangeability.

A campus needing hours of peak management has different sizing needs from one focused on short transients and generator bridging.

Operating strategy matters as much as nameplate capacity. A battery reserved for ride-through cannot simultaneously offer its full capacity for market participation or demand response.

State of charge must remain within a range that preserves the required reserve. Poor scheduling could leave insufficient energy during a grid event.

Battery degradation adds another variable. Available capacity, internal resistance, and thermal behavior change as systems accumulate cycles and age.

Qualification at the product level does not remove the need for long-term performance guarantees. Buyers must examine maintenance, augmentation, and replacement plans.

Cybersecurity also remains a shared responsibility. A grid-interactive BESS connects controls, telemetry, remote services, and facility networks.

NVIDIA points suppliers toward established cybersecurity frameworks. However, each operator must still assess access controls, software updates, incident response, and network segmentation.

Safety requires similar caution. Compliance documents and test evidence support a project, but local authorities retain jurisdiction over installation and operation.

Fire separation, ventilation, suppression, emergency procedures, and first-responder access depend on the site and its applicable codes.

There is also a strategic risk for customers. NVIDIA’s qualification can concentrate design decisions around one company’s evolving reference architecture.

Standardization can lower integration work. It can also make builders more dependent on NVIDIA’s preferred interfaces, requirements, and update cycles.

The program currently offers a short list, which limits procurement diversity. Supply constraints or project-specific rules could still force developers toward other products.

NVIDIA says more categories and products will follow. Until that expansion occurs, the launch list should be read as an initial ecosystem milestone.

The correct buyer response is neither dismissal nor unquestioning acceptance. DSX Ready provides meaningful technical screening, while leaving full project accountability with the developer.

The Next Three Signals Will Show Whether DSX Ready Matters

The program’s significance will depend on adoption, list expansion, and evidence from operating data centers.

The first signal is the number and diversity of additional qualified systems. NVIDIA has launched with three BESS suppliers, but a durable procurement standard needs broader participation.

More vendors would improve regional availability and give developers alternatives across different system designs, service models, and supply chains.

If the list grows quickly, NVIDIA’s requirements are becoming an industry entry point. If it remains narrow, DSX Ready may function mainly as a preferred-partner directory.

The details of future additions will also matter. New listings should clarify exact configurations, tested software versions, qualification dates, and update requirements.

Battery products change over time. Suppliers revise firmware, controls, cells, cooling systems, and power electronics as they improve manufacturing or address shortages.

NVIDIA will need a method for determining when those changes require renewed testing. Otherwise, a badge could outlive the configuration that originally earned it.

The second signal is whether major AI campuses specify DSX Ready systems in actual procurement documents. Reference designs gain authority through deployment, not publication alone.

A public project naming Tesla, LG Energy Solution, or Hitachi Energy under DSX requirements would show that the program influences purchasing decisions.

The strongest evidence would connect qualification with faster engineering approval, shorter interconnection work, or fewer integration problems.

Those results will take time. Large data centers face long equipment lead times, complex utility studies, and construction schedules measured in years.

Near-term announcements can still reveal direction. Developers may disclose preferred suppliers, framework agreements, or completed integration studies before sites begin operating.

The third signal is operating evidence from completed facilities. That includes response to power ramps, grid disturbances, generator transitions, and demand-response events.

Real systems will show whether submitted models accurately reproduce field behavior. They will also reveal how batteries interact with site equipment outside NVIDIA’s qualification boundary.

Transparent incident reporting will be particularly valuable. Failures, unexpected trips, controller conflicts, and firmware corrections often expose limitations that laboratory tests miss.

NVIDIA’s broader power strategy also deserves attention. The company is promoting 800 VDC rack architectures while its facility-level batteries connect through AC power systems.

These layers address different parts of the electrical path. Their coordination will affect conversion losses, protection design, controls, and operational flexibility.

Cooling qualification could provide an early comparison. DSX Ready launched cooling distribution units alongside BESS, showing that NVIDIA wants common rules across several facility layers.

If developers adopt both categories, NVIDIA will influence more of the infrastructure surrounding its computing systems. That expands DSX from a design guide toward a procurement framework.

For enterprises buying AI capacity, this matters even without direct involvement in electrical engineering. Power constraints affect deployment schedules, service availability, and future computing costs.

For infrastructure teams, the NVIDIA BESS qualification program offers a practical starting point, not a final engineering answer. Ask suppliers for the tested configuration, model package, deviations, and recertification policy.

Then ask the harder question: does the complete site remain stable when that qualified battery meets the utility, generators, protection system, and real AI workload?

The next stage will determine whether DSX Ready becomes a widely accepted infrastructure gate or stays an NVIDIA-centered partner program. Watch the supplier list, named deployments, and field data in that order.