Oracle Ramps Up NVIDIA Compute and Adds Google TPU, Liquid-Cooled Server Demand Heats Up

Oracle ramps up NVIDIA compute and introduces Google TPU for the first time, pushing AI data center power density even higher. Liquid cooling has shifted from an optional configuration to a mandatory element of compute infrastructure, drawing attention to multiple stocks in the A-share supply chain.

Oracle ramps up NVIDIA to multi-gigawatt scale

Oracle announced on Tuesday local time that it is expanding its cloud AI chip partnership with NVIDIA, using NVIDIA's latest-generation Vera Rubin AI servers to scale its cloud deployment to "multi-gigawatt" levels, while also planning to bring chips from NVIDIA competitors such as Google and Cerebras Systems onto its cloud platform.

Specifically, Oracle will for the first time offer customers compute based on Alphabet's Google TPU chips on its cloud platform, and plans to launch chips from AI inference chip makers SambaNova Systems and Cerebras Systems. Even for NVIDIA-based systems, Oracle will drop some NVIDIA networking chips in favor of its own Acceleron networking chips. The company maintained its capital expenditure target of roughly $90 billion to $95 billion for fiscal 2027.

As compute scale expands, heat dissipation is upgrading from a supporting link to a hard constraint. According to an announcement by the State Administration for Market Regulation and the Standardization Administration of China, China's first national standard for data center liquid cooling—GB/T 48023-2026 "Technical Specification for Cold Plate Liquid Cooling Systems in Data Centers," drafted under the leadership of Sugon DataEnergy, a subsidiary of Dawning Information Industry—has been officially approved and released, and will take effect on February 1, 2027; another standard, GB/T 48097-2026 "General Specification for Cold Plate Liquid Cooling Cabinets in Data Centers," will also take effect on March 1, 2027.

Market space on the industry side is opening up in tandem. According to estimates cited by the Ningbo Municipal Bureau of Economy and Information Technology, the global data center liquid cooling market is expected to reach $16.5 billion in 2026, up 59% year on year, with an average annual compound growth rate expected to exceed 30% over the next five years; China's liquid cooling data center market is projected to reach 47 billion yuan by 2028.

Liquid cooling moves from optional configuration to mandatory standard

Liquid cooling servers refer to data center cooling solutions that use liquid as the heat transfer medium, removing heat from chips and servers through cold plates, immersion and other methods, supported by CDU, quick connectors, piping and cold source systems. As AI chip power consumption rises rapidly, traditional air cooling is approaching its heat dissipation limits in high power density single-rack scenarios, and liquid cooling is shifting from an option in data center construction to a mandatory element of compute infrastructure.

"The rigid demand for AI compute heat dissipation is established, and the hundred-billion-yuan liquid cooling track is timely." Yuanda Information noted in an industry report that with the rollout of GB200, GB300 and the next-generation Rubin compute platforms, AI server rack heat density will climb further, significantly raising the comprehensive performance thresholds of liquid cooling systems for CDUs, manifold assemblies, cooling media, pipeline connectors and overall control systems.

Project demand is gradually transmitting from planning to equipment orders. Changjiang Securities believes that domestic and international liquid cooling standards are being continuously refined, and as energy efficiency constraints are gradually incorporated into project procurement evaluations, the energy efficiency of cooling equipment will directly affect bidding competitiveness, and high-efficiency liquid cooling systems and supporting cold source products are expected to win more orders.

This judgment echoes the view of Huaxi Securities, which believes that the release of the first national liquid cooling standard marks the industry's entry into a new stage of standardization, and that unified standards are expected to alleviate pain points such as inconsistent interfaces and long delivery cycles, accelerating the large-scale deployment of liquid cooling solutions.

Earnings of multiple concept stocks expected to multiply

According to East Money's concept sector data, 193 A-share stocks currently involve the liquid cooling server concept, with a combined total market value of about 5.87 trillion yuan.

Foxconn Industrial Internet leads with a scale of 911.241 billion yuan, Midea Group ranks second with 648.881 billion yuan, Luxshare Precision follows with 377.869 billion yuan, and Gree Electric, Sanhua Intelligent Controls, Lens Technology and ZTE all have market values exceeding 130 billion yuan.

Over the past month, the liquid cooling server concept sector has trended weakly, with 12 stocks rising and 181 falling. Bohui Shares surged 53% to lead, Dayang Motor rose 19%, Zhongjie Shares gained 17%, and Longtech Intelligent, Moons' Electric and Jiangsu Leili all rose more than 4%. On the decline side, Sinoshield Technology, Oulutong, Dingtong Technology and other stocks fell more than 40%.

In terms of future growth potential, according to consensus institutional forecasts, a total of 51 liquid cooling server concept stocks are expected to see their earnings double this year. Among them, institutions expect Far East Smarter Energy's net profit to surge 12.6 times year on year, Jiechuang Intelligent's net profit to grow more than 9.6 times, and Lopal Tech, China Greatwall, Cambridge Technology and Yidong Electronics to all see forecast net profit growth of more than 5 times, while the remaining 47 stocks are also forecast to see net profit growth of more than 1 times.

Envicool has launched the Coolinside full-chain liquid cooling solution, with products covering CDUs, dry coolers, fluid distribution manifolds, liquid cooling quick connectors, cooling media and cold source systems, providing an integrated heat dissipation solution from primary side to secondary side. The company said in an investor relations activity that in addition to adapting to CPUs, GPUs and computing and switching ASIC chips, its cold plate products have begun to match other heat sources such as memory, SSDs and optical modules to support 100% full liquid cooling demand for compute equipment, and that overseas incremental demand currently being tracked is expected to gradually start from the third quarter.

Inspur's liquid cooling server production line remains under high-load operation, providing full-chain support from product assembly and liquid path testing to complete machine verification. The company has achieved 120-day commissioning of a 10MW intelligent computing center with a prefabricated AIDC model, shortening the construction cycle of a project of the same scale from 18 months to 4 months, with PUE reduced to 1.1; it has also recently taken the lead in breaking through micron-level surface waterproof coating technology for liquid cooling server motherboards.

(Source: East Money Research Center) Mid-article embedded advertisement