AI Server Boom: Huawei Atlas 950, Dell PowerEdge, HPE ProLiant, NVIDIA Vera Rubin Lead the Liquid-Cooling Revolution

The global AI computing infrastructure has entered a period of explosive growth in 2026. At the recent Mobile World Congress (MWC) 2026 in Barcelona, flagship AI server platforms including Huawei Atlas 950 SuperPoD, Dell PowerEdge, HPE ProLiant, and NVIDIA Vera Rubin made high-profile appearances. These next-generation systems, powered by full liquid-cooling technology, have greatly improved computing density, energy efficiency, and cluster scalability, reshaping the landscape of data centers and intelligent computing infrastructure worldwide.

Driven by trillion-parameter large language models, multi-modal AI agents, autonomous driving training, industrial digital twins, and cloud-edge intelligence, traditional server architectures can no longer meet surging computing demands. The AI server industry is shifting from simple “GPU-integrated servers” to full-rack, AI-native system design. Liquid cooling has become a standard configuration, with rack power exceeding 1MW, and overall computing capacity rising by an order of magnitude. Huawei, Dell, HPE, and NVIDIA have emerged as the four core leaders in this AI server boom, driving the industry from general-purpose computing to true AI-native computing.

At MWC 2026, Huawei officially unveiled its Atlas 950 SuperPoD for the global market, attracting widespread attention. Based on the self-developed Lingqu Unified Interconnect architecture, the system supports up to 8,192 Ascend AI accelerators working in parallel, delivering up to 8 EFLOPS of FP8 computing power. Its internal interconnection bandwidth reaches 16PB/s, enabling unified memory addressing and near real-time latency. This design solves the pain points of low cluster utilization and unstable training in traditional large-scale systems. Huawei also launched the TaiShan 950 SuperPoD for general computing, forming a full-stack product matrix covering AI training, inference, and general computing to provide secure, stable, and high-performance solutions for global customers.

Dell Technologies has become one of the fastest-growing vendors in AI servers with its PowerEdge series. In its Q4 fiscal 2026 earnings report, Dell’s AI server revenue reached nearly $9 billion in a single quarter, representing a year-on-year increase of more than 340%. Backlog orders have exceeded $430 billion. PowerEdge AI servers support the latest NVIDIA Blackwell and Rubin architectures, equipped with HBM3e high-bandwidth memory and high-speed interconnection modules, and are widely deployed by cloud providers, intelligent computing centers, and enterprise AI platforms.

Due to the global shortage of DRAM, NAND flash, and AI computing components, Dell’s AI server lead times have been extended to 20–28 weeks. Starting in March 2026, prices will increase by 10%–25%, and quotation validity will be shortened to 7–14 days, directly reflecting the tight supply of global AI computing power.

HPE is building an integrated “compute + network + security” AI infrastructure based on its ProLiant series and the recently acquired Juniper Networks. Flagship models such as ProLiant EL9000 and EL140 Gen12 support NVLink 4, high-speed InfiniBand switching, and full liquid cooling. These servers work seamlessly with Juniper’s PTX series 800G routers to meet ultra-low-latency and high-bandwidth requirements in AI data centers.

HPE also faces supply constraints, with AI server lead times stretching to 20–27 weeks and price increases of 11%–28% starting in March. Thanks to its global delivery capability and hybrid cloud adaptability, HPE remains a top choice for large enterprises and telecom operators.

NVIDIA’s Vera Rubin platform marks a new era of full-rack supercomputing for AI. As a fully liquid-cooled flagship solution, it integrates 72 Rubin GPUs and 36 Vera CPUs, paired with NVLink 6 and Spectrum-X high-speed networking, achieving an internal bandwidth of up to 260TB/s in a single cabinet. Compared with the previous-generation Blackwell platform, Vera Rubin improves training efficiency by 4x, reduces inference costs by 90%, and boosts performance-per-watt by 10x, optimized for next-generation agentic AI and large-scale inference. Mass deployment is expected in the second half of 2026, supporting the world’s top intelligent computing centers.

Full liquid cooling is a key enabler of this AI Server Upgrade. As power consumption per accelerator exceeds 2,000W, traditional air cooling is no longer sufficient. Cold-plate and immersion liquid cooling have been rapidly adopted. Huawei Atlas 950, Dell PowerEdge, HPE ProLiant, and NVIDIA Vera Rubin all featurefull liquid-cooling design, reducing PUE below 1.1, increasing computing density by 3–5 times, and lowering long-term operating costs, aligning with global green data center trends.

The AI server boom has triggered a global shortage of upstream components. HBM3e memory prices have risen 30%–50% in six months, DDR5 server memory has increased by more than 90%, and 800G/1.6T optical modules, high-speed switching chips, and high-efficiency power supplies are in short supply. Major manufacturers have raised prices and extended delivery times to 16–28 weeks, yet demand from cloud giants, financial institutions, energy companies, and research institutes remains strong.

Looking ahead, 2026 is widely regarded as the first year of large-scale commercialization for AI servers. Supported by trillion-parameter models, industrial intelligence, and cloud computing upgrades, AI servers will maintain high growth in the next 1–2 years. Full liquid cooling, large-scale clustering, ultra-low latency, and high energy efficiency will become core technical directions.

Huawei Atlas 950, Dell PowerEdge, HPE ProLiant, and NVIDIA Vera Rubin represent the cutting edge of the industry and define its future path. As technology matures and the supply chain stabilizes, AI servers will lower the threshold of intelligent computing and accelerate the digital transformation of industries worldwide, becoming the core infrastructure of the digital economy era.