Taiwan Cooling Components: JPMorgan sees Nvidia VR Ultra and AI ASIC cooling upgrades driving larger-than-expected content growth

Overview

JPMorgan examines Taiwan liquid-cooling component suppliers through the accelerating cooling-spec upgrades expected in Nvidia GPU systems and cloud-service-provider ASIC platforms. Its central conclusion is that investor expectations underestimate the resulting content increase, particularly for cold plates, quick disconnects and heat spreaders, supporting a favorable sector view and preferences for AVC and Fositek.

Core views

JPMorgan argues that cooling has become the principal bottleneck to further TDP growth in AI chips and to the migration of AI-server systems. It sees broad specification upgrades in Nvidia's Vera Rubin Ultra (VRU), countering concerns that cooling content might be flat or reduced. The report expects double-sided cold plates, diamond-copper materials, higher-spec quick disconnects and a removable three-piece lid design to lift the value of cooling components materially. For VRU compute trays, the report estimates total cooling content of US$3,000-3,500, 30-50% above current Vera Rubin levels. The mechanism is a double-sided cold-plate design required by vertical power delivery, which moves power components, SOCAMM and capacitors to the back of the board and requires incremental backside cooling. JPMorgan expects six cold plates for two Bianca modules on the upper tray section, alongside existing cooling for CX-9 chips and the power-distribution board. Diamond copper around HBM-contact areas adds further value, while additional Mini QDs and migration of selected Bianca-module and plug QDs to ZQD specifications raise content. It models compute-tray QD content at about US$750 for VRU, versus US$400-500 for Vera Rubin. The report also expects the 1.5U VRU switch-tray architecture, with two boards per tray, to lift cooling content to more than US$3,000 per tray. QD units inside the switch tray are modeled to rise to eight plus two external units from four plus two, while the external plug QDs move from UQD08 to ZQD14. This takes estimated QD content per switch tray to about US$260 from roughly US$150. At the rack level, JPMorgan forecasts total cooling content excluding the CDU of about US$98,500 for VRU NVL72, versus approximately US$66,250 for Vera Rubin NVL72 and US$48,250 for GB300 NVL72. It estimates VRU rack QD content of about US$20,500, compared with about US$11,000 for Vera Rubin and US$9,400 for GB300. For heat spreaders, JPMorgan expects Rubin Ultra to use a removable lid, stiffener and multiple-screw, or three-piece, structure rather than the one-piece lid used in Rubin. It estimates a 4-5x lift in heat-spreader/lid content. Small-volume Rubin trial adoption could begin in 4Q26, with Rubin Ultra mass production potentially in 1Q-2Q27. The report also points to AMD MI450's two-piece lid, estimated at roughly 10x the ASP of MI350's design, and prospective CPO/NPO and ASIC upgrades as longer-term structural support for the segment. CSP ASIC designs reinforce the thesis despite sometimes lower chip TDP. AWS Trainium 3 compute-tray cooling content is estimated at US$3,000-3,500, about 30% above Nvidia Vera Rubin, because of eight double-sided cold plates for four ASICs plus a baseboard cold plate. Including 10 switch trays, JPMorgan estimates total Teton 3 Max rack cooling content of about US$86,000, versus US$66,250 for Vera Rubin NVL72. Google TPU v8 compute-tray content is estimated at US$2,000-2,500, similar to Vera Rubin despite lower TDP, due to 12 cold plates and gold plating; rack cooling content is estimated at about US$38,000. TPU v9 could adopt a removable-lid-plus-stiffener solution as chip size and TDP rise, potentially toward 2,000W, although that design remains unconfirmed. JPMorgan expects the sector to outperform and favors AVC for its leading cold-plate position across GPU and ASIC projects, better-than-peer margin and earnings growth potential, and valuation it considers undemanding. It favors Fositek for expected near-term gross-margin upside, ZQD-driven content growth, share gains and total-addressable-market expansion. Looking out one to two years, it identifies broader liquid cooling in CPO/NPO switches and higher-spec optical transceivers or modules, likely from 1.6T/3.2T generations, as the next growth driver.

Analysis framework

JPMorgan compares cooling architectures, component counts, specifications, ASPs and estimated content at compute-tray, switch-tray and rack levels across Nvidia GPU and CSP ASIC systems. It then links the modeled content increases and product positioning to supplier earnings, margin and valuation implications.

Methodology notes

• Industry AnalysisVolume-price decomposition

  Component-content and ASP analysis

  The report estimates cooling value by combining component counts, design specifications and assumed selling prices, then compares content across system generations and platforms.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  AI-chip TDP growth transmitted to cooling-component demand

  JPMorgan links higher-power GPU and ASIC architectures to more complex cooling requirements, then to supplier content, margins and earnings.

Asset mapping & comparison

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

• Asia Vital Components (3017.TW)

  Preferred cold-plate and manifold supplier positioned to benefit from GPU and ASIC cooling-spec upgrades.

  Strengths

  Leading position and market share across projects; potential for better-than-peer margins and earnings growth.

  Weaknesses

  Recent share-price performance has underperformed peers and, according to the report, even the Taiex.

  Comparison

  Trades at about 20x 12-month forward P/E versus its historical 18-22x range; JPMorgan views valuation as undemanding.

• Fositek (6805.TW)

  Preferred quick-disconnect supplier expected to benefit from ZQD adoption, share gains and market expansion.

  Strengths

  Potential near-term gross-margin upside, higher-spec QD content growth and additional customer or project wins.

  Comparison

  Trades at about 20-25x 12-month forward P/E versus a historical 15-30x range.

• Jentech Precision Industrial Co. (3653.TW)

  Comparable heat-spreader/lid supplier.

  Strengths

  Potential beneficiary of removable-lid adoption for Rubin and Rubin Ultra.

  Comparison

  Trades at about 50x 12-month forward P/E, near the upper end of its historical 30-50x range.

  Risks

  Partial Rubin GPU adoption for trial volumes and only 4-5x, rather than higher, content growth when mass production begins.

Key data

• VRU compute-tray cooling contentUS$3,000-3,500Estimated 30-50% increase versus current Vera Rubin cooling content.
• VRU NVL72 total cooling contentUS$98,500Excluding CDU; compared with US$66,250 for Vera Rubin NVL72 and US$48,250 for GB300 NVL72.
• VRU NVL72 QD content~US$20,500About 100% higher than prior generations, driven by ZQD adoption and added units.
• Rubin Ultra heat-spreader/lid content4-5xVersus Rubin's current one-piece lid design.
• AWS Teton 3 Max rack cooling content~US$86,000Excluding CDU; above Nvidia Vera Rubin NVL72's estimated US$66,250.
• Google TPU v8 rack cooling content~US$38,000Driven by a complex 12-cold-plate design and gold plating.

Impact & implications

The report says cooling-content growth should exceed prevailing expectations as AI systems adopt double-sided cold plates, higher-spec QDs and more complex heat-spreader designs. It sees AVC and Fositek as the principal Taiwan beneficiaries, while CPO/NPO switching and higher-spec optical modules could extend the growth runway over the next one to two years.

Risks

• For heat-spreader suppliers, Rubin GPU may adopt the removable lid only in partial trial volumes.
• Rubin Ultra heat-spreader/lid content growth may be limited to 4-5x versus the one-piece design when mass production begins.
• Potential Google TPU v9 removable-lid adoption remains unconfirmed.

What to watch

• Timing and scale of Nvidia Rubin trial production from 4Q26 and Rubin Ultra mass production potentially from 1Q-2Q27.
• Adoption of double-sided cold plates and diamond-copper materials in Nvidia VRU, AWS Trainium 3, AMD MI450 and Google TPU platforms.
• Extent of ZQD adoption and Mini QD additions in VRU compute trays, switch trays and rack manifolds.
• CPO/NPO switch and higher-spec optical-module liquid-cooling adoption in future 1.6T/3.2T generations.