Overview of IDC 2026 and the AI Thermal Challenge

The IDC 2026 8th Shanghai International Data Center Liquid Cooling Exhibition, scheduled for December 9-11 at the Shanghai New International Expo Centre, highlights a critical juncture in AI infrastructure. Expected to host over 300 exhibitors and 18,000 professionals across 18,000 square meters, the event underscores the urgent need for advanced thermal management. The exhibition will prominently feature PUE ≤ 1.1 practices, addressing the physical limits of traditional power architectures in high-density AI racks. As AI high-computing chips push data centers to their limits, the synergy between 800V high-voltage direct current (HVDC) and liquid cooling has emerged as the definitive pathway to sustainable AI scaling.

Technical Details: 800V HVDC and Advanced Liquid Cooling

Traditional 50V power architectures are buckling under the immense current and copper costs associated with modern AI racks. The industry's pivot to 800VDC significantly reduces current, thereby minimizing wire loss. Leading manufacturers are deploying 660kW 800VDC in-row power racks equipped with battery backup units to maximize GPU density in IT cabinets.

Concurrently, thermal management is transitioning to mandatory liquid cooling solutions, including:

  • Cold plate liquid cooling for targeted chip extraction
  • Immersion cooling for complete server submersion
  • Phase-change liquid cooling for maximum thermal transfer

High-capacity in-row cooling distribution units and micro-channel cold plates are essential. Looking ahead, Solid State Transformers utilizing silicon carbide devices will convert 13.5kV AC directly to 800VDC, achieving 98.5% efficiency.

Market Context: Strategic Positioning and Revenue Shifts

The financial landscape reflects this technological paradigm shift. Industry leaders are experiencing record-breaking revenues driven by AI data center demands. Recent financial disclosures indicate that power and component segments have surged to 57% of total revenue, overtaking traditional infrastructure. In the first half of 2026, cumulative revenue growth reached 41%, proving that AI server power supplies have become the most certain growth engine. The transition from traditional power suppliers to co-designers of comprehensive rack-level power and cooling architectures demonstrates a deep integration with major AI chip manufacturers, aligning production capabilities with next-generation specifications.

Industry Impact: Overcoming Physical Limits in Power Density

The physical limits of power and cooling are starkly evident in the exponential leap of rack power density. While traditional data centers operated at approximately 10kW per cabinet, current mainstream AI racks demand 125-140kW, with next-generation platforms approaching 250kW. Single-chip power consumption has escalated dramatically, with next-generation GPU architectures requiring up to 2300W per chip, and entire rack-scale systems demanding 110kW per three-rack-unit shelf. Consequently, the global AI data center liquid cooling market is projected to reach approximately $15 billion in 2026, with penetration rates expected to exceed 40% globally and surpass 50% in China by 2027.

Implications for AI Data Centers: Capacity Expansion and Ecosystem

To meet surging orders, aggressive capacity expansion is underway. Major investments are being directed toward manufacturing hubs in mainland China, including a 269 million RMB new facility in Dongguan designed to produce 1.4 million switching power supplies annually, alongside expansions in Chenzhou. The Dongguan base has already achieved 100% green power. Overseas, new facilities in Thailand and the US are being accelerated to ensure just-in-time delivery for North American cloud providers. Ecosystem synergies are also strengthening, with strategic partnerships securing massive computing power orders and ensuring seamless supply chain integration for AI clients.

Future Outlook: From Current Cash Flow to Next-Gen Architecture

Financially, the immediate revenue drivers remain AI server AC-DC power upgrades and continuous liquid cooling penetration. While ±400VDC is already shipping in volume, 800VDC is slated for small-batch deliveries to North American cloud providers in the fourth quarter of 2026. The true inflection point for 800V HVDC architecture will arrive in 2027, coinciding with the mass production of rack-level systems. Following the 2026-2027 white space renovation phase, the industry will move toward hybrid power distribution, ultimately reaching the SST final state by 2029-2030. This synergy will fundamentally redefine the operational models of AI data centers.