ZutaCore, a liquid-cooling vendor based in Foster City, California, used the 2026 OCP Global Summit to name and describe the internal design of its newest cold plates. In an Oct. 13 press release, the company introduced what it calls Two-Phase Matrix Flow Boiling, the boiling architecture inside its OmniTherm cold plate line. The company said deployments of OmniTherm are under way this quarter for servers built on NVIDIA's RTX PRO 6000 Blackwell Server Edition GPU.

The announcement belongs to a larger effort across the industry to pull more heat off accelerator packages as their power draw climbs. ZutaCore's own framing is that chip power has "roughly tripled" over three generations. That figure comes from the company's release, and DataCenterFeed has not independently verified it. One point applies to everything below: nearly all of the performance and design claims come from ZutaCore. DataCenterFeed found no independent benchmark of Matrix Flow Boiling or of the OmniTherm plates.

The questions below set out what was announced, what is new and what remains unverified.

What is Matrix Flow Boiling, according to ZutaCore?

The company describes it as a manifold-based design. Common headers feed fluid straight down onto a grid of many small boiling regions, and the vapor produced in each region escapes straight up. ZutaCore's pitch is that the arrangement combines the stability of pool boiling with the scalability of flow boiling. In pool boiling, liquid sits against a hot surface and boils in place. In flow boiling, liquid is pushed across the surface. Splitting the plate into many small regions, each fed from a shared header, is how the company says it gets the benefits of both. The release adds that every region gets fresh liquid and that the hottest regions automatically boil harder.

The release also says the cold plate works in any orientation. That would matter for server builders who mount boards vertically, inverted, or in other layouts, though no independent testing of the claim has been published.

How big and heavy are the plates?

ZutaCore says the OmniTherm architecture allows cold plates as thin as 11.6 mm, thin enough to fit a single-wide PCIe slot. With a hybrid copper-aluminum construction, the company says weight can come down to as little as 300 g, which it describes as less than 1 lb. Mixing copper and aluminum can raise concerns about galvanic corrosion, the electrochemical reaction between two different metals in contact with a conductive fluid. ZutaCore says its design carries no risk of galvanic corrosion. The company did not cite third-party testing for that statement.

The single-slot form factor is the main selling point for the PCIe-card market. In earlier coverage from May 2026, The Cooling Report described OmniTherm as aimed at the single-slot PCIe version of NVIDIA's RTX PRO 6000 Blackwell Server Edition. That outlet reported that Shahar Belkin, ZutaCore's chief evangelist, framed the single-slot form factor as a way for operators to increase accelerator density in standard server architectures.

Why no water?

ZutaCore's systems run a sealed dielectric fluid through the server, not water. The release says that fluid poses no risk to IT equipment and that low flow rates keep cooling power down. The fluid boils as it takes on heat from the chip. According to The Cooling Report, the heat is then rejected at a remote condenser, with the water loop staying entirely outside the server enclosure, so water never has to be routed into the server itself. That outlet also said OmniTherm reduces fan dependency and the noise that goes with it, and compared the approach with single-phase liquid cooling and immersion cooling.

The Data Center Richness show-floor roundup reported on ZutaCore's booth at the summit. It described a demonstration of an ASRock Rack 4U server with NVIDIA HGX B300 GPUs, cooled by the company's HyperCool system using a dielectric fluid rather than water, with the coolant boiling as it absorbs heat.

Is this a new platform or an add-on to an existing one?

ZutaCore presents it as the next-generation cold plate for its installed HyperCool system. The release says OmniTherm uses the same fluid, connections, racks and coolant distribution units (CDUs) as HyperCool, "already proven across more than 75 deployments," with no re-architecture required. The deployment count comes from the company, and the release does not list the sites or customers.

For operators, the practical claim is that a facility already running HyperCool could add OmniTherm cold plates without changing its fluid or distribution hardware. DataCenterFeed has not confirmed that compatibility with any operator.

What about future, hotter chips?

ZutaCore says the architecture scales beyond the multi-kilowatt packages on industry roadmaps and is designed to move into the chip package and, ultimately, onto the silicon itself. The release says a development program is under way with partners, but it does not name the partners or give a timeline. According to Data Center Richness, ZutaCore also says the approach addresses use cases such as back-side power delivery, inverted orientation and non-standard server configurations. Those are company statements, and no published performance data accompanied them.

Who else was at the summit with ZutaCore technology?

The release lists several partners that show ZutaCore technology at their own booths: MiTAC, with an L11 rack solution using ZutaCore HyperCool for the AMD Instinct MI355X; Chemours, with a pool-boiling demonstration; and Wiwynn, showing the OmniTherm cold plate. The release does not describe the commercial terms of those relationships. It also quotes My D. Truong, ZutaCore's chief product and technology officer, on the announcement.

How does this fit the broader cold-plate market?

ZutaCore was one of several vendors showing cold-plate approaches at the summit. The same Data Center Richness roundup noted that Frore Systems showed its LiquidJet Nexus integrated cold-plate system, designed for half-rack-unit (½U) AI compute trays. Data Center Richness also noted that two-phase showcases from vendors such as Accelsius and ZutaCore now put the boiling inside a cold plate rather than an immersion tank. Vendors are competing to cool denser, hotter accelerator packages in tighter spaces, and two-phase approaches sit alongside single-phase liquid cold plates and immersion.

What should readers treat with caution?

Several points deserve care:

  • The performance claims are self-reported. The thickness, weight, corrosion, orientation and scalability claims all come from ZutaCore. No independent thermal test results were found.
  • One widely circulated version is a reprint. A StorageNewsletter item dated Oct. 20 is labeled as a press release edited by that publication, and it repeats ZutaCore's release. It is not independent confirmation.
  • The deployment count is not itemized. The "more than 75 deployments" figure for HyperCool comes with no list of sites, customers or sizes.

What Happens Next

According to ZutaCore's Oct. 13 release, OmniTherm deployments for NVIDIA RTX PRO 6000 Blackwell Server Edition systems are under way this quarter. The company also says a development program with partners is under way to extend the architecture toward the chip package. The release also lists a ZutaCore two-phase panel on Thursday, Oct. 15, at 13:05 at the summit, with participants from Meta, Wiwynn, ASRock Rack, Corintis and Chemours. The sources give no dates, named partners or milestones for the chip-package program.

Sources