Quick Summary
Covering the latest research from top Wall Street investment banks

AI data center liquid-cooling cold plates: near-term spending opportunity is clear, but commoditization and end-stage phase-out risks cannot be ignored

Institution
Bernstein
Date
2026-07-05
Authors
Varun Govindaraj, Chad Dillard, Alasdair Leslie, Steve Song
Company
-
Ticker
-
Industry
AI data center liquid cooling, electrical equipment, computing hardware
Rating
-
NeutralLow confidenceAI workloads are lifting rack heat density, and rising liquid-cooling penetration is making cold plates a key high-volume component in DTC systems; however, cold plates are hardware-intensive with limited service attach, and as design standardization and supply-chain maturity improve, commoditization and potential substitution risk increase.
AuthorsVarun Govindaraj, Chad Dillard, Alasdair Leslie, Steve Song
CoverageUnited States
Asset classesEquity
Business segmentsCold plate、Direct-to-chip liquid cooling、Data center thermal management、Electrical equipment
Research firm divisions/subsidiariesBernstein(Other)

AI summary card

AI data center liquid-cooling cold plates: near-term spending opportunity is clear, but commoditization and end-stage phase-out risks cannot be ignored

Bernstein views cold plates as a core component of direct-to-chip liquid-cooling architecture, with a 2030 base-case market of roughly $6B to $7B, but long-term earnings defensibility is weaker than that of CDU.

Among covered companies, VRT target price $416, NVT target price $218, TT target price $550, JCI target price $176, ETN target price $534, SU.FP target price €310 are Outperform; CARR target price $75 is Market-Perform.
AI data centerliquid coolingcold plateDTCelectrical equipmentthermal managementcommoditization risk
  • NVIDIA GB200 NVL72 requires about 108 cold plates per rack, showing cold plates are a high-volume deployed component.
  • The report expects the cold-plate market to grow from about $2B-$3B today to about $6B-$7B in the 2030 base case.
  • Near-term demand is supported by rising liquid-cooling penetration, NVIDIA Rubin all-liquid-cooling ecosystem support, and hyperscaler roadmaps.
  • Cold plates are structurally simpler than CDU with lower service attachment, so hardware margins can be more easily compressed as manufacturing scales.
  • Long-term risks come from two-phase DTC maturity, cold-plate design standardization, and potential replacement by direct silicon-etch microchannel cooling or similar technical paths that may reduce or displace cold plates.

Report interpretation

Overview

This report is a Bernstein deep introductory study of cold plates in U.S. multi-industry and electrical equipment liquid-cooling. Starting from rising rack power density in AI data centers, it explains the role of cold plates in direct-to-chip liquid-cooling systems, key performance indicators, ecosystem standards, competitive dynamics, service-attach characteristics, and market size. The core conclusion is that cold plates benefit in the short term from quickly rising liquid-cooling penetration and are a key enabling component of AI infrastructure expansion, but the business model is largely dependent on hardware gross margin, making it more vulnerable to long-term commoditization and technological substitution.

Core views

The report argues that cold plates have a clear demand tailwind in the short term: air cooling can no longer meet thermal demands of high-density AI racks, DTC liquid cooling becomes a key data center architecture, and cold plates are the only component that directly contacts GPUs or CPUs and transfers heat to coolant. From the medium term through 2030, as two-phase DTC matures, supply chains become more complete, and hyperscalers further clarify specifications, the scarcity and innovation premium of cold plates may decline. After 2030, if technologies such as direct silicon-etched microchannel cooling become commercialized, cold plates could face partial substitution risk or even terminal displacement.

Analysis framework

The report uses a combined approach of industry-level teardown and market sizing: it first explains the thermal-circuit relationship among CDU, cold plates, TCS, and FWS in liquid-cooling systems, then decomposes cold-plate thermal capacity, heat flux, pressure drop, thermal resistance, and channel geometry; it next compares OCP, NVIDIA approved-vendor lists, and industrial partnerships and their impact on ecosystem shape, and finally estimates 2030 market size based on assumptions for incremental data-center GW, per-GPU power draw, liquid-cooling penetration, cold-plate pricing, and replacement demand.

Methodology notes

  • Industry supply-demandBottom-up market sizing

    Estimate cold-plate market space using added AI data-center capacity, GPU power draw, liquid-cooling penetration, and per-kW cooling value.

    The report gives a 2030 base-case cold-plate market of about $6B-$7B and emphasizes that the outcome is highly sensitive to the incremental GW and per-kW cooling value assumptions.

  • Technical assessmentLayered liquid-cooling architecture analysis

    Differentiate the functions of CDU, cold plates, TCS, and FWS in the data-center liquid-cooling loop.

    Cold plates directly transfer heat from chips to coolant; CDU manages coolant distribution, flow, and pressure control. Both benefit from demand, but their complexity and service attach differ.

  • Competitive landscapeCommoditization risk assessment

    Judge margin defensibility through design standardization, manufacturing scale, service attach, and substitute technology.

    Cold-plate design value is above system complexity value, and over time with clearer specifications and mature supply chains they are more likely to be driven by contract manufacturing ecosystems, pressuring margins.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

  • VRT
    Cold-cooling and data-center infrastructure beneficiary
    Strengths
    Rated Outperform with a $416 target price; narrative aligns with participation in liquid-cooling ecosystems involving NVIDIA, Vertiv, and related partners.
    Weaknesses
    Cold plates themselves have lower service attach, and hardware profits may be pressured as commoditization increases.
    Comparison
    Compared with pure cold-plate designers or manufacturers, broad infrastructure companies may have a wider mix of products and services.
    Risks
    Liquid-cooling deployment pace, valuation, supply-chain competition, and cold-plate price declines.
  • NVT
    Electrical equipment and data-center infrastructure-linked candidate
    Strengths
    Rated Outperform with a $218 target price; benefits from the AI data-center electrical and thermal-management capex cycle.
    Weaknesses
    The report does not indicate its cold-plate business is a unique or direct core exposure.
    Comparison
    Shares the same AI data-center capex exposure chain as VRT and ETN.
    Risks
    Slower data-center build-out, insufficient product-mix exposure, or valuation compression.
  • TT
    Thermal management and HVAC-related candidate
    Strengths
    Rated Outperform with a $550 target price; thermal-management capabilities are linked to the liquid-cooling infrastructure trend.
    Weaknesses
    The direct participation level in the cold-plate hardware value chain requires further confirmation against company-specific business mix.
    Comparison
    Compared with cold-plate manufacturers, the business is more broadly oriented toward thermal-management and equipment.
    Risks
    Tempo of liquid-cooling replacement versus traditional cooling, project-cycle fluctuations, and margin pressure.
  • JCI
    Industrial and building-equipment company involved in liquid-cooling investment ecosystem
    Strengths
    Rated Outperform with a $176 target price; the report notes JCI’s investment in Accelcius, indicating strategic participation in the liquid-cooling ecosystem.
    Weaknesses
    Investment and ecosystem participation may not immediately translate into material incremental revenue.
    Comparison
    Unlike standalone cold-plate firms, JCI participates through investment and channel-level ecosystem involvement.
    Risks
    Changes in liquid-cooling technology roadmap or slower commercialization of investment targets.
  • ETN
    Electrical equipment and data-center power-infrastructure candidate
    Strengths
    Rated Outperform with a $534 target price; the report cites Eaton’s acquisition of Boyd, increasing liquid-cooling exposure.
    Weaknesses
    Acquisition integration and cold-plate profitability remain exposed to commoditization pressures.
    Comparison
    Through Boyd, ETN has more direct cold-plate and thermal-management supply-chain contact than some peers.
    Risks
    Integration execution risk, price competition, and changes in customer specifications.
  • SU.FP
    European electrical equipment and data-center infrastructure candidate
    Strengths
    Rated Outperform with a €310 target price; benefits from data-center electrification and infrastructure expansion in Europe.
    Weaknesses
    The report does not present cold plates as a single core driver for SU.FP.
    Comparison
    In the same data-center electrical-equipment chain as ETN and VRT, but with different regional and portfolio exposure.
    Risks
    European demand, FX, project execution, and valuation volatility.
  • CARR
    Thermal-management and HVAC-related candidate
    Strengths
    Rated Market-Perform with a $75 target price; has relevance to cooling and temperature-control themes.
    Weaknesses
    More neutral rating, and the report does not highlight a clear direct cold-plate advantage.
    Comparison
    Compared with Outperform names, the report assigns a more cautious investment rating.
    Risks
    Liquid-cooling competitive dynamics, traditional HVAC business cycles, and limited upside to target price.

Key data

  • 2030 base-case cold-plate market sizeabout $6B-$7BThe report says current size is about $2B-$3B, rising to about $6B-$7B in the 2030 base case.
  • New-build liquid-cooling penetration pathcurrently about 30%-40%, moving toward majority adoption by 2030Higher liquid-cooling penetration is the main volume driver of cold-plate demand.
  • NVIDIA GB200 NVL72 cold plate countabout 108 units per rack18 compute trays, 72 Blackwell GPUs, and 36 CPUs, with each GPU/CPU requiring a dedicated cold plate.
  • Current cold-plate heat capacitytypically about 2-5 kW per chip, next-generation designs can reach 5-10 kW+The report notes that CoolIT has announced a 15 kW cold plate.
  • Current heat flux targetaround 100-130+ W/cm², leading designs move toward 150+ W/cm²Heat flux measures how well a cold plate handles local hotspots.
  • Typical pressure droparound 15-25 psi, with high-performance designs reaching 30+ psiLower pressure drop is better for reducing pump power, but there is a tradeoff with high heat-transfer performance.
  • High-performance thermal resistancebelow 0.04°C/WLower thermal resistance means lower chip temperature at the same heat output.
  • VRT rating and target priceOutperform, target price $416Disclosed in both table and text.
  • NVT rating and target priceOutperform, target price $218Disclosed in both table and text.
  • CARR rating and target priceMarket-Perform, target price $75Disclosed in both table and text.

Impact & implications

From an investment perspective, liquid-cooling cold plates should benefit in the near term from rising AI server power consumption and accelerated liquid-cooling deployment by hyperscalers, creating revenue opportunity for related thermal-management, electrical equipment, and data-center infrastructure suppliers. But unlike CDU, cold plates have less service and maintenance upside; value is more concentrated in design and hardware manufacturing. As NVIDIA, OCP, and hyperscalers push toward clearer specifications, supplier pricing power may decline, and the market can shift from an innovation-premium phase to a large-scale manufacturing and cost-competition phase.

Risks

  • As OCP, NVIDIA, and hyperscalers clarify requirements, cold-plate design and specifications become standardized, reducing innovation premium.
  • Cold-plate service attach is significantly lower than CDU, so long-term profitability is more exposed to compression of hardware margins.
  • As contract manufacturing ecosystems expand, cold plates may move from design scarcity to mass-manufactured commodities.
  • After two-phase DTC matures, the existing single-phase cold-plate path may face upgrade and replacement pressure.
  • If direct silicon-etched microchannel cooling commercializes, it could reduce or eliminate traditional cold-plate demand.
  • Immersion cooling is viewed by the report as unlikely to become mainstream, but it remains part of substitute-technology discussion.
  • Market sizing is highly sensitive to assumptions on incremental data-center GW, GPU power consumption, liquid-cooling penetration, and per-kW pricing.

What to watch

  • Whether NVIDIA Rubin and follow-on architectures continue to push 100% liquid cooling and fanless ecosystems.
  • Changes in cold-plate supplier list and mass-production validation progress in the NVIDIA Recommended Vendor List.
  • Whether OCP thermal standards for cold plates move from outcome metrics to more granular design constraints.
  • Commercialization speed and supply-chain maturity of two-phase DTC during the 2028-2030 window.
  • Whether cold-plate value per kW ($/kW) declines as manufacturing scales.
  • Whether hyperscalers further define cold-plate specs, thereby reducing supplier design discretion.
  • Investment levels and commercialization signals from Microsoft, TSMC, and others in direct silicon-etched microchannel cooling.
  • Disclosures from VRT, ETN, JCI on liquid-cooling orders, M&A, partnerships, and margin trends.
Zhejiang ICP No. 2022035445-5
Disclaimer: Market data, charts, indicators, research views, and other information provided on this website are intended solely for information display, research communication, and educational reference. They should not be regarded as personalized investment advice, securities recommendations, trading instructions, solicitations, or guarantees of return. While we strive to improve the reliability of our data and content, such information may still be subject to delays, errors, incompleteness, or untimely updates due to source differences, methodological limitations, system processing, or market volatility. Users should exercise independent judgment based on their own circumstances and bear all risks and responsibilities arising from the use of this website.

Settings

Sign in to view recent logins