CDU, the Core Component of Liquid Cooling: A High-Barriers Track, Prefer NVIDIA Ecosystem Partners
AI summary card
CDU, the Core Component of Liquid Cooling: A High-Barriers Track, Prefer NVIDIA Ecosystem Partners
Bernstein publishes a deep-dive report on liquid cooling CDUs, arguing that the CDU is the heart of the liquid cooling system with low obsolescence risk and high service stickiness; favors Vertiv, nVent and other manufacturers with both NVIDIA partnerships and OCP certification.
- The CDU is the core of the liquid cooling system, with high complexity and requiring ongoing service, making it difficult to fully commoditize
- The market size is estimated at low single-digit billions of dollars in 2026, growing to mid-to-high single-digit billions by 2030
- Vertiv, nVent, and Boyd simultaneously hold NVIDIA partner status and OCP certification, representing a significant competitive advantage
- Trane Technologies leads in single-unit capacity with standout modular design, exceeding expectations
- Two-phase DTC cooling is the key variable for next-generation technology; currently only Vertiv and Accelsius have made substantive progress
- Carrier and JCI prioritize chillers over CDUs, with product specifications slightly lagging
Report interpretation
Overview
This report focuses on the coolant distribution unit (CDU), a critical component in data center liquid cooling systems. As AI chip power consumption surges, air cooling can no longer meet thermal dissipation requirements, making liquid cooling the mainstream trend. The report thoroughly breaks down the technical principles, evaluation metrics, competitive landscape, and future evolution of CDUs, and provides investment ratings for relevant listed companies. The core conclusion is that the CDU business model is excellent, supply is constrained in the near term, and in the long run, technology iteration and service stickiness will support pricing power for leading manufacturers—especially those deeply tied to the NVIDIA roadmap and participating in the Open Compute Project (OCP).
Core views
CDU possesses an excellent business model and technological moat. As the 'heart' of the liquid cooling system, the CDU controls coolant distribution, flow rate, and pressure, with complexity far exceeding that of cold plates. Because CDU failures can burn out multiple racks, customers have extremely high reliability requirements and will not simply pursue the lowest-cost provider, giving manufacturers strong service revenue stickiness and pricing power. Even as hyperscale cloud providers push design standardization through OCP, the multi-component integration nature of CDUs makes them harder to fully commoditize than cold plates. The report believes that regardless of how the mainstream cooling architecture evolves (single-phase, two-phase, or immersion), CDUs are indispensable and face virtually no technology obsolescence risk. On market size, despite wide divergence in forecasts, the report believes a reasonable assumption is that the global CDU market in 2026 is at low single-digit billions of dollars (LSD $B), with potential to grow at mid-teens+ CAGR over the next 5 years, reaching mid-to-high single-digit billions of dollars (MSD-HSD $B) by 2030. This growth is highly sensitive to variables including data center GW additions, cooling cost per kW, and CDU lifespan. The market remains in severe supply shortage, with order backlogs expected to stay elevated in the medium term. The competitive landscape shows clear differentiation. Within the NVIDIA liquid cooling ecosystem, Vertiv, nVent, Boyd (acquired by Eaton), and Motivair (acquired by Schneider) have the most comprehensive product portfolios. Among these, only Vertiv, nVent, and Boyd have OCP-certified CDUs compliant with Google's Deschutes 5.0 specification, signaling their ability to meet the highest technical standards and engineering relationships with hyperscale customers beyond the project level. Trane Technologies, while not in NVIDIA's core circle, has a Giga-modular CDU with 2.5MW single-unit capacity expandable to 14MW, with performance described as 'exceeding expectations.' In contrast, Carrier and Johnson Controls focus more on traditional chiller businesses, with CDU products slightly inferior in specification details, capacity ceiling (lacking independent >2MW units), and approach temperature difference (ATD). CoolIT is more of a cold plate company, with CDU approach temperature lagging competitors. On technology evolution, the shift from single-phase direct-to-chip cooling (DTC) to two-phase DTC is the key focus over the next 5 years. Two-phase cooling leverages liquid boiling heat absorption, far more efficient than single-phase, but with fundamentally different equipment design requirements. Currently only Vertiv and Accelsius have shown substantive products or published detailed technical white papers, with at least one more year before commercial-scale deployment. This technology leap could reshape the existing market landscape. Additionally, being a NVIDIA partner is viewed as a critical factor for long-term success, as NVIDIA effectively sets industry power consumption and thermal dissipation standards; manufacturers outside this ecosystem (such as JCI, Carrier, CoolIT) may be at a competitive disadvantage in the long run.
Analysis framework
The report employs a 'technology deconstruction + multi-dimensional benchmarking' analytical framework. First, by mapping five liquid cooling technology pathways (single-phase DTC, two-phase DTC, immersion, etched cold plates, silicon-based etching), it establishes the universality and irreplaceability of CDUs across all architectures, thereby ruling out technology obsolescence risk. Second, it constructs a quantitative evaluation system with four core metrics for horizontal comparison across manufacturers: cooling capacity (Nominal Capacity), approach temperature difference (ATD), flow ratio (Flow Ratio), and pressure head. The report specifically notes that manufacturers often inflate nominal capacity through specific test conditions (such as using pure water instead of PG25 coolant, high ATD operating points), so analysis must be normalized to hyperscale customers' actual operating environments (e.g., PG25 coolant, low ATD, 80psi pressure head) for 'apples-to-apples' comparison. Finally, combining technical capability with strategic positioning, it introduces 'NVIDIA partner' and 'OCP certification' as two qualitative dimensions serving as proxy variables for long-term competitiveness, thereby translating pure technical parameters into investment judgment criteria.
Methodology notes
Four-dimensional evaluation index system for CDU product performance
The report proposes evaluating CDU performance through four metrics: cooling capacity, approach temperature difference (ATD), flow ratio, and pressure head. ATD measures heat exchange efficiency (lower is better), flow ratio reflects fluid consumption per unit of heat dissipation (1.5 lpm/kW is the gold standard), and pressure head determines ability to drive complex flow paths. This methodology helps investors see through manufacturer marketing claims to identify real performance differences.
Technology moat based on mission-criticality and service stickiness
The report argues that the CDU moat stems not only from technical complexity but also from the customer psychology that 'failure cost far exceeds service price differentials.' Because CDU failures can destroy expensive racks, customers will not switch providers for lower prices; this risk-aversion-driven stickiness is more durable than pure product differentiation.
Liquid cooling technology maturity and commercialization stage positioning
The report arranges different liquid cooling technologies by maturity: single-phase DTC in the maturity phase, two-phase DTC on the eve of early commercialization, and silicon-based etching still in the conceptual phase. This stratification helps judge investment windows for each technology—current CDU investment should focus on single-phase DTC volume, while tracking two-phase DTC breakthroughs as the next catalyst.
Ecosystem positioning analysis method (NVIDIA partnership + OCP certification)
The report uses 'whether a NVIDIA partner' and 'whether having OCP-certified CDU' as binary filters for screening long-term winners. The former represents foresight on future technology roadmaps, the latter represents ability to meet top-tier customers' customized demands. Those with both are viewed as having the 'right to win'—a practical method of translating industry standard-setting power into stock selection criteria.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- Vertiv (VRT.US)NVIDIA core partner + OCP certification + two-phase cooling leader, most comprehensive layout across all technology pathways
- Strengths
- Only manufacturer covering all technology pathways including single-phase, two-phase, and immersion; has multiple >2MW CDUs and OCP-compliant products; two-phase DTC already has physical demonstration
- Comparison
- Broader technical coverage than nVent and Boyd; deeper ecosystem binding than Trane
- Risks
- High valuation (2026E P/E 46.4x); delay in two-phase technology route may affect premium
- nVent (NVT.US)NVIDIA partner + OCP certification, balanced product portfolio
- Strengths
- Has both rack-level and row-level CDUs; OCP-compliant; expected to launch new-generation CZ/CX series in 2H26
- Weaknesses
- Currently no public two-phase DTC product
- Comparison
- Together with Vertiv as one of the OCP top three; more focused on electrical connections than Boyd
- Risks
- Low visibility on two-phase technology reserves
- Trane Technologies (TT.US)Facility-side cooling partner, CDU product performance exceeded expectations
- Strengths
- Giga-modular CDU at 2.5MW single-unit and expandable to 14MW, with unique modular design; ATD as low as 3°C
- Weaknesses
- Not a NVIDIA core cooling partner (only facility-side); no OCP-certified CDU; no public progress on two-phase DTC
- Comparison
- Single-unit capacity and modularity superior to most competitors; but ecosystem position weaker than Vertiv/nVent
- Risks
- If NVIDIA-dominated two-phase standards are incompatible with existing products, may face transformation pressure
- Carrier (CARR.US)Indirectly participates in two-phase cooling through equity investment in ZutaCore; CDU not core business
- Strengths
- Claims ability to achieve 2°C ATD (under specific conditions); has 5MW skid-mounted solution
- Weaknesses
- No independent >2MW CDU; insufficient product specification disclosure; not a NVIDIA/OCP partner
- Comparison
- CDU product strength and strategic priority clearly lower than Vertiv, nVent, and Trane
- Risks
- May continue to lag in liquid cooling CDU segment; rated Market-Perform
- Johnson Controls (JCI.US)Participates in two-phase cooling through equity investment in Accelsius; CDU product competitiveness average
- Strengths
- Silent-Aire CDU supports 10MW skid-mounted; flow ratio reaches 1.5 lpm/kW
- Weaknesses
- ATD as high as 5.2°C (under PG25 conditions); no >2MW independent unit; not a NVIDIA/OCP partner
- Comparison
- ATD metric significantly inferior to industry benchmark (3-4°C); overall specifications below peers
- Risks
- CDU business may be marginalized; relies on external investment for technology
- Eaton (ETN.US)Acquired CDU capabilities and NVIDIA/OCP dual qualifications through Boyd acquisition
- Strengths
- Inherited Boyd's OCP-compliant CDU and NVIDIA partnership; has historical two-phase cooling experience
- Weaknesses
- Has not yet clearly launched own-brand two-phase CDU product
- Comparison
- Post-integration ecosystem position in the same first tier as Vertiv and nVent
- Risks
- M&A integration effects to be validated
- Schneider Electric (SU.FP)Entered CDU field through Motivair acquisition; part of NVIDIA ecosystem
- Strengths
- Motivair MCDU70 reaches 2.5MW and supports 10MW expansion; flow ratio >1.5 lpm/kW
- Weaknesses
- Two-phase cooling only has blog post, no substantive product demonstration
- Comparison
- Excellent product parameters, but two-phase progress slower than Vertiv
- Risks
- Uncertain timing of two-phase technology landing
Key data
- 2026 CDU Market Size EstimateLow single-digit billions of dollars (LSD $B)Report's base case assumption; market has wide divergence on specific scale
- 2030 CDU Market Size EstimateMid-to-high single-digit billions of dollars (MSD-HSD $B)Expected CAGR over next 5 years at mid-teens+
- Flagship CDU Capacity Benchmark>2MWIndependent units below this capacity seen as damaging manufacturer credibility
- OCP-Compliant CDU Manufacturers3 (Vertiv, nVent, Boyd)Based on Google Deschutes 5.0 specification, showing high technical barrier
- Flagship CDU Approach Temperature Difference (ATD)3-4°CTop products reliably achieve 3°C; Carrier claims 2°C but requires additional high-efficiency heat exchanger
- Gold Flow Ratio Standard≥1.5 lpm/kWKey metric measuring CDU pumping efficiency; too low causes overheating, too high wastes energy
Impact & implications
For the liquid cooling industry chain, the certainty of the CDU segment is higher than that of cold plates. Cold plates are closer to consumables; once designs are finalized, they are easily replaced by contract manufacturers; whereas CDUs, due to high system integration and strong service dependency, are better able to withstand price wars. For investors, priority should be given to manufacturers that can keep pace with NVIDIA's rapid iteration rhythm while locking in hyperscale customers' long-term demand through open standards like OCP. The commercialization progress of two-phase cooling technology will be the next important watershed, with enterprises that first launch mature products potentially gaining阶段性超额收益. Meanwhile, traditional HVAC giants that cannot effectively elevate the strategic priority of CDUs in their product portfolios may gradually become marginalized in the liquid cooling era.
Risks
- CDU market size calculations are highly sensitive to assumptions on added GW, liquid cooling penetration rate, and unit cost; actual scale may deviate significantly from expectations
- Hyperscale customers pushing design standardization through projects like OCP may bring some commoditization pressure in the medium to long term
- Two-phase DTC cooling technology commercialization timeline is uncertain; delays may affect related manufacturers' valuation premiums
- Some manufacturers (e.g., Carrier) have incomplete product specification disclosure; actual performance may be below stated values
What to watch
- Two-phase DTC cooling product commercialization progress and measured performance data
- NVIDIA next-generation chip platform's new cooling requirements
- OCP newly added CDU certified manufacturer list and technical specification updates
- Each manufacturer's CDU order backlog and capacity expansion pace
- Hyperscale customers' actual preferences for parameters such as ATD and flow ratio in real-world deployments