Two-phase direct-to-chip cooling has carried a consistent caveat in every analyst report and conference panel for the past three years: strong technical case, limited deployment base, and insufficient institutional capital to build the supply chain that would let operators treat it as a default option rather than a managed experiment. The funding gap was real. A startup with promising refrigerant-based cooling hardware cannot build manufacturing at the scale that hyperscale deployment requires on venture capital alone.
Accelsius closed a $65 million Series B in January 2026. Johnson Controls International led the round. Legrand joined as a strategic investor. These are not generalist technology investors placing bets on a cooling startup. They are infrastructure companies with existing product lines in the data center thermal and IT enclosure markets, writing checks that signal where they think the technology is heading.
Johnson Controls — NYSE: JCI, roughly 140 years old — is not a newcomer to data center cooling. The company's Silent-Aire division builds coolant distribution units at 500 kW to 10 MW capacity and supplies CDUs to hyperscale operators. Its YORK YVAM magnetic bearing chiller, which consumes 40 percent less power than conventional chillers, is a standard spec in high-efficiency data center mechanical plants. Johnson Controls had announced its initial investment in Accelsius in October 2025, before the formal Series B close.
Austin Domenici, Johnson Controls VP and GM of Global Data Center Solutions, framed the strategic logic plainly: "AI-scale computing rewrites thermal requirements. Our investment aligns with delivering advanced cooling solutions." What that means in practice is that Johnson Controls is holding a position in two-phase DTC while also selling the single-phase CDUs and chilled water infrastructure that current deployments run on. They are not betting that two-phase wins outright. They are ensuring they are not caught flat-footed if it does.
Legrand's position is different. The company — represented by President and CEO Brian DiBella — builds the IT infrastructure that surrounds the servers: cabinets, power distribution units, cable management, rack-level monitoring. DiBella's stated rationale was direct: "Thermal management is now strategic for IT performance." Legrand's cabinet and PDU business intersects directly with how liquid cooling hardware gets housed, powered, and monitored at the rack level. An investment in a rack-integrated two-phase CDU company is an investment in the product category their enclosure business will need to support.
Accelsius is an Austin, Texas company founded through Innventure, a firm that creates businesses by commercializing technology from large corporate R&D portfolios. The NeuCool platform uses non-conductive dielectric refrigerants in a two-phase direct-to-chip loop — the fluid absorbs heat at the cold plate through liquid-to-vapor phase change, then rejects that heat at the facility-side heat exchanger before returning as liquid. The non-conductive characteristic means the refrigerant can coexist with server hardware without the catastrophic failure mode that water contamination would create.
CEO Josh Claman described the company's position in terms that tracked the scale of the problem the funding is meant to address: "AI-scale computing is rewriting the thermal playbook. Building responsibly at gigawatt scale." The platform claims 35 to 44 percent annual OpEx savings versus single-phase direct-to-chip cooling and up to 90 percent cooling energy reduction versus air cooling. Accelsius is a member of NVIDIA's Inception program, which provides access to technical resources and go-to-market support from NVIDIA for companies building infrastructure in the AI ecosystem.
The $65 million goes to three things: expanding the manufacturing footprint in Austin, accelerating global expansion, and speeding deployment of the NeuCool platform. Manufacturing capacity is the binding constraint for any liquid cooling hardware company trying to serve hyperscale procurement timelines. A hyperscale operator that wants 500 or 5,000 units on a 12-month schedule cannot be accommodated by a facility running at startup production volumes. The funding addresses that ceiling directly.
The DarkNX deployment — a 300-megawatt campus in Ontario, Canada — is the commercial reference point anchoring the round. A 300 MW campus is not a pilot. It is the kind of scale that requires a vendor to have solved manufacturing, logistics, installation support, and ongoing maintenance before the first rack ships. The DarkNX agreement is evidence that Accelsius has cleared enough of those hurdles to be trusted with that scale.
When a company like Johnson Controls invests in a two-phase cooling startup, the natural interpretation is that they are hedging against their own product line becoming obsolete. That reading is incomplete. Johnson Controls is not hedging against CDUs disappearing. They are building toward a world where data center thermal management requires multiple modalities — traditional chilled water plants, row-level CDUs, and rack-integrated two-phase systems — and they want to have a product position at each layer.
The cooling infrastructure market does not work on winner-take-all dynamics. Different workloads, different rack densities, and different facility configurations will use different thermal architectures for years. What the JCI investment signals is that two-phase DTC has cleared the threshold from "interesting emerging technology" to "technology we need to be commercially positioned in." That threshold matters more than the dollar amount of the round.
Accelsius announced general availability of the NeuCool IR150 in April 2026, roughly three months after this round closed. The IR150 — a rack-integrated system rated at 150 kW with zero water in the IT enclosure — is the product that the Series B funded into production. The capital and the product are now in the market at the same time.