2026 Will Be the Year of AI Technology Hardware

Source: Hard AI
Author: Dong Jing

Morgan Stanley predicts that 2026 will be the pivotal year for explosive growth in AI technology hardware, primarily driven by robust demand for AI server hardware.

On November 3, according to HardAI News, Morgan Stanley noted in its latest research report that AI server hardware is undergoing a significant design upgrade driven by GPUs and ASICs. NVIDIA’s upcoming GB300, Vera Rubin platform, and Kyber architecture, as well as AMD’s Helios server rack project, are expected to deliver higher computing power and rack density.

Morgan Stanley emphasized in the report that with the launch of NVIDIA’s Vera Rubin platform in the second half of 2026, rack demand is projected to surge from approximately 28,000 units in 2025 to at least 60,000 units in 2026. More importantly, to support GPUs with higher power consumption, the value of power solutions could grow more than tenfold by 2027. Liquid cooling solutions, which are becoming standard, will also see increasing per-rack value. Additionally, components such as PCB/substrates and high-speed interconnects will undergo significant specification upgrades.

This indicates that the value of the entire AI server hardware supply chain is being reassessed. The firm maintains a positive outlook on several companies within the AI hardware supply chain, believing that the new generation of AI server design upgrades will bring substantial benefits to related manufacturers.

Explosive Growth in AI Server Rack Demand

Morgan Stanley explicitly pointed out in the report that the growth driver for AI hardware is transitioning from the H100/H200 era to a new cycle powered by NVIDIA’s GB200/300 (Blackwell platform) and the subsequent Vera Rubin (VR series) platform.

According to NVIDIA’s product roadmap, its GPUs are experiencing a leapfrog upgrade in both power consumption and performance.

From the H100’s 700W TDP, to the B200’s 1,000W, then to the GB200’s 1,200W, and finally to the Vera Rubin (VR200) platform set to debut in the second half of 2026, the maximum TDP for its GPU will soar to 2,300W. By 2027, the VR300 (Kyber architecture) will reach an even higher level at 3,600W.

Morgan Stanley believes that this increase in computational density is directly driving a complete overhaul in server rack design, from conceptualization to component selection.

Morgan Stanley forecasts that demand for AI server racks based solely on NVIDIA’s platforms will surge from approximately 28,000 units in 2025 to at least 60,000 units in 2026, representing more than a twofold increase. Meanwhile, AMD’s Helios server rack project (based on the MI400 series) has also made significant progress, further intensifying market demand for advanced AI hardware.

The report emphasizes that as rack designs shift from single-GPU upgrades to integrated designs encompassing entire rack systems (Rack), ODM manufacturers with strong integration capabilities and a track record of stable deliveries, such as Quanta, Foxconn, Wistron, and Wiwynn, will dominate the supply chain for GB200/300 racks.

Power consumption and thermal bottlenecks are driving up the value of power supply and liquid cooling solutions.

One of the most striking points in the report is that the challenges posed by power consumption and heat dissipation in AI hardware upgrades are transforming into significant business opportunities for power supply and cooling solution providers. This is regarded as one of the fastest-growing value segments in this round of upgrades.

On the power front, the sharp rise in total power consumption per rack has rendered traditional power architectures unsustainable.

The report forecasts that power architecture will transition to an 800V high-voltage direct current (HVDC) solution. This shift will significantly enhance the value content of power solutions.

Morgan Stanley expects that by 2027, the value of power solutions designed for the Rubin Ultra cabinet (using Kyber architecture) will be more than 10 times that of the current GB200 server cabinet. Additionally, by 2027, the value of power solutions per watt in AI server cabinets will double compared to the present stage.

In terms of cooling, liquid cooling has evolved from an optional solution to a necessity. The report points out that as the TDP of the GB300 platform exceeds 1,400W, liquid cooling has become a standard feature. This directly increases the value of cooling components. Specific data shows:

The total value of cooling components for a GB300 (NVL72) cabinet is approximately USD 49,860. In the next-generation Vera Rubin (NVL144) platform, due to increased cooling demands for both computing trays and switch trays, the total value of cooling components per cabinet will grow by 17%, reaching USD 55,710. Notably, the value of cooling modules designed for switch trays will surge by 67%.

Morgan Stanley stated that this clearly indicates that key component suppliers in the liquid cooling supply chain, such as cold plate modules, cooling fans, and quick connectors (NVQD), will benefit directly.

Comprehensive upgrade across the value chain, with PCBs/substrates and high-speed interconnects experiencing proportional growth.

In addition to power and cooling, the report also highlights the profound impact of AI platform upgrades on printed circuit boards (PCBs) and various interconnect components. Each GPU iteration brings higher requirements for the number of PCB layers, material grade, and size.

According to NVIDIA’s GPU evolution roadmap compiled by Morgan Stanley:

ABF substrate: The number of layers has increased from 12 layers (12L) for H100 to 14 layers for Blackwell (B200), and further to 18 layers for Vera Rubin (VR200), with corresponding increases in size.

OAM (OCP Accelerator Module) motherboard: PCB specifications have been upgraded from 18-layer HDI (High-Density Interconnect) for H100 to 22-layer HDI for GB200/300, and may reach 26-layer HDI for VR200.

CCL (Copper Clad Laminate) materials: Transitioning from ultra-low loss (Ultra low-loss, M7) to extreme low-loss (Extreme low-loss, M8) grades to meet the requirements of higher data transmission rates.

Morgan Stanley believes that these technical upgrades signify more complex manufacturing processes and higher value for PCB boards. The increasing demand for high-layer-count HDI boards and high-grade CCL materials will bring structural growth opportunities to PCB and upstream material suppliers with corresponding technological capabilities.

The report explicitly states that design upgrades in AI GPU and ASIC servers will strongly support the capacity expansion wave in the PCB/substrate industry.

Meanwhile, Morgan Stanley notes that data and power interconnection solutions are also being upgraded to match the demands for higher data transmission speeds and capacities.

Editor/jayden