Supermicro Claims 1,000x Impedance Coolant for AI Racks at Computex: Independent Testing Pending

Super Micro Computer took the stage at Computex 2026 in Taipei on Tuesday to unveil what its CEO called a potential breakthrough in liquid cooling for AI infrastructure: a proprietary dielectric fluid called SMC PG25-A that the company claims achieves electrical impedance 1,000 times higher than standard coolants — a figure that, if independently confirmed, could change the risk calculus for full-rack immersion cooling in next-generation AI data centers. The announcement, made during a CEO keynote at the show's second day, paired the coolant claim with a fully configured NVIDIA Vera Rubin NVL72 rack and a new end-to-end data center blueprint designed to scale from 5 megawatts all the way to 1 gigawatt.

Why Electrical Impedance Determines AI Rack Uptime

The core technical question at the heart of Tuesday's announcement is whether a coolant's electrical properties can make the difference between a rack worth roughly $8 million continuing to operate through a minor leak versus shutting down immediately.

"Our new coolant formula reaches up to 1,000 times higher electrical impedance than a standard cooler," CEO Charles Liang told the Computex audience. "In case there are small leaks, when you have a high electrical impedance, the system will not shut down and will keep running."

The practical implication is significant. Standard water-glycol coolants used in direct liquid cooling systems carry some electrical conductivity. When coolant contacts a motherboard, GPU, power delivery circuitry, or connectors, even minor leakage currents can trigger a system shutdown or, in worse cases, cause component damage. A coolant with 1,000-fold higher electrical impedance would be far more resistant to current flow, substantially reducing the likelihood that a minor leak forces an immediate shutdown. For the Vera Rubin NVL72 — a fully liquid-cooled rack containing 72 NVIDIA Rubin GPUs, 36 NVIDIA Vera CPUs, and 331 terabytes of HBM4 memory — that uptime protection has direct financial value. Each rack draws roughly 440 kilowatts from four 110-kilowatt power shelves, each with redundant power supply units, and racks are reported to cost upward of $8 million apiece.

Supermicro is branding the underlying cooling architecture as DLC-2 (Direct Liquid Cooling, second generation), a full-stack system pairing the SMC PG25-A fluid with cold plates, coolant distribution units, manifolds, rear-door heat exchangers, and cooling towers — all engineered for what the company describes as near-total heat capture.

How Dielectric Cooling Fluid Works in AI Data Centers

Standard direct-to-chip cooling routes chilled water through cold plates mounted on the hottest components. Full immersion cooling — where entire rack assemblies are submerged in a dielectric fluid — offers a step-change improvement in heat capture because every component, including HBM memory stacks and networking hardware, is continuously bathed in coolant. The tradeoff has historically been the risk that any conductive residue in the fluid could short electronics. As IEEE Spectrum reported in November 2025, dielectric immersion fluids "must be nonconductive and have strong thermal transfer properties," along with requiring "long-term stability and low environmental and fire risk."

A fluid with 1,000-fold higher electrical impedance would address the conductivity risk directly, removing the primary safety barrier to whole-rack immersion. That is the claim Supermicro is making. Whether it holds up at production scale — across thousands of racks, operating temperatures, and maintenance cycles — remains to be established.

Tom's Hardware, which attended the Computex keynote and published an independent technical review, noted that the claim is difficult to evaluate because Supermicro has not disclosed the fluid's key specifications: electrical conductivity in microsiemens per centimeter, resistivity in megaohms per centimeter, or dielectric strength in kilovolts per millimeter. The company also has not identified the baseline coolant used for comparison. Until those figures are published and tested by independent labs, the 1,000× figure represents a marketing claim rather than a verified engineering advance.

DCBBS Blueprint AI Infrastructure: From 5 MW to 1 Gigawatt

The coolant announcement is one component of a broader product offensive Supermicro unveiled at Computex. The company simultaneously introduced its Data Center Building Block Solutions (DCBBS) Blueprints — end-to-end data center designs based on the Vera Rubin NVL72 and the NVIDIA HGX Rubin NVL8 platforms, engineered to scale from a 5-megawatt starting point to a full gigawatt.

Each base scalable unit within the blueprint supports 1,152 NVIDIA Rubin GPUs delivering a combined 331 terabytes of HBM4 memory, along with compute, storage, networking, liquid cooling infrastructure, power distribution, battery backup, and Supermicro's SuperCloud management software — packaged under what the company is calling "single-vendor accountability."

The full cooling infrastructure in the blueprint's reference configuration includes 5-megawatt cooling towers, four in-row coolant distribution units rated at up to 1.8 megawatts each, 16 vertically mounted cooling distribution manifolds, and 576 direct-to-chip copper cold plates. For data centers that lack liquid cooling infrastructure, Supermicro is offering liquid-to-air sidecar options: a 200-kilowatt unit for a single rack, and a 500-kilowatt unit covering two racks.

The pitch addresses a documented pain point for large AI infrastructure buyers. A typical hyperscale build-out involves more than a dozen separate vendor relationships spanning compute, storage, networking, racks, cooling, power, cabling, and services. Every handoff introduces schedule risk and accountability gaps. Supermicro is positioning DCBBS as a way to collapse that complexity into a single contract — a strategy Dell, HPE, and Lenovo are also pursuing with their own AI factory architectures, though none has yet published a competing coolant specification claim comparable to SMC PG25-A.

Deliveries Scheduled for H2 2026

The DCBBS Blueprints for both the Vera Rubin NVL72 and the NVIDIA HGX Rubin NVL8 are now open for customer engagements, with full deployments scheduled to align with NVIDIA Vera Rubin's general availability, expected in the second half of 2026. Major hyperscalers are expected to be among the earliest recipients, with Supermicro demonstrating both platforms live at its Computex booth (N0824) in Taipei through June 6 and at NVIDIA GTC Taipei.

The company's AMD roadmap is also on track. Supermicro confirmed it is on track to release AMD MI455X-based Helios rack-scale solutions in the second half of this year, though it did not specify whether those systems use UALink or UALink-over-Ethernet interconnects. The company also expects to bring 1-way and 2-way servers based on AMD's 6th-generation EPYC "Venice" processors — manufactured on TSMC's 3nm process — to market before the end of 2026.

Supermicro's Legal Shadow: What Buyers Must Know

The Computex showcase arrives in the wake of the most serious legal crisis in Supermicro's history. On March 19, 2026, the U.S. Department of Justice unsealed an indictment in Manhattan federal court charging three individuals associated with the company — including co-founder Yih-Shyan "Wally" Liaw, who served on Supermicro's board of directors — with conspiring to smuggle approximately $2.5 billion worth of Nvidia-equipped AI servers to China through shell companies in Southeast Asia, in violation of U.S. export control laws. Liaw has pleaded not guilty. Supermicro itself was not named as a defendant, placed both employees on administrative leave, terminated the contractor relationship, and stated it is cooperating fully with the government's investigation.

Taiwan's first formal criminal enforcement against illegal AI chip exports followed on May 21, 2026, when the Keelung District Prosecutors' Office raided 12 locations and sought detention for three individuals in a related case involving Supermicro servers shipped to China, Hong Kong, and Macau. Nvidia CEO Jensen Huang, in Taipei two days later ahead of his Computex keynote, told reporters publicly that Supermicro needs to improve its regulatory compliance.

Supermicro has launched an independent internal investigation led by its audit committee chair and an outside director, supported by law firm Munger, Tolles & Olson and forensic accounting firm AlixPartners. Securities class action lawsuits have been filed and an ongoing SEC investigation is active. None of these proceedings change the technical specifications of Tuesday's product announcement — but they are directly relevant to any procurement decision in which Supermicro is the designated single vendor of critical AI infrastructure.

Frequently Asked Questions

What is Supermicro's SMC PG25-A coolant and what does it claim?

Supermicro's SMC PG25-A is a proprietary dielectric coolant used in its DLC-2 direct liquid cooling system. The company claims it achieves electrical impedance 1,000 times higher than standard coolants, which would reduce the risk that a minor coolant leak causes a server shutdown in high-density AI racks like the NVIDIA Vera Rubin NVL72.

Has Supermicro's 1,000x impedance claim been independently verified?

Not yet. As of Computex 2026, Supermicro has not published the fluid's key specifications — including electrical conductivity, resistivity, or dielectric strength — nor identified the baseline coolant it used for comparison. Without those figures, independent laboratories and buyers cannot evaluate whether the 1,000× figure represents a meaningful advance over competing dielectric fluids already on the market.

What is the DCBBS Blueprint and when will it ship?

Supermicro's Data Center Building Block Solutions (DCBBS) Blueprints are end-to-end data center designs that scale from 5 megawatts to 1 gigawatt, covering compute, cooling, networking, storage, power, and management software in a single vendor engagement. Blueprints based on the NVIDIA Vera Rubin NVL72 and HGX Rubin NVL8 are open for customer engagements now, with full deployments scheduled for the second half of 2026, aligned with NVIDIA Vera Rubin's general availability.

What legal issues is Supermicro facing as of mid-2026?

A DOJ indictment unsealed March 19, 2026 charges three individuals associated with Supermicro — including co-founder Wally Liaw — with conspiring to route approximately $2.5 billion in Nvidia-equipped AI servers to China through shell companies. Supermicro is not a named defendant and says it is cooperating fully. The company has launched a second independent internal investigation, appointed a new Chief Compliance Officer, and faces ongoing SEC scrutiny and securities class action lawsuits.