Morgan Stanley predicts that 2026 will mark a pivotal year for explosive growth in AI tech hardware, primarily driven by surging demand for AI server hardware.
According to the latest research report, AI server hardware is undergoing a major design upgrade fueled by GPUs and ASICs. NVIDIA's upcoming GB300, Vera Rubin platform, and Kyber architecture, along with AMD's Helios server rack project, are expected to deliver higher computing power and rack density.
The report highlights 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. Additionally, power solutions supporting higher GPU power consumption could see their value grow over 10x by 2027, while liquid cooling solutions—now a standard—will also see rising per-rack value. Further upgrades are expected in PCB/substrates and high-speed interconnects.
This signals a revaluation of the entire AI server hardware supply chain. Morgan Stanley maintains a positive outlook on multiple AI hardware supply chain companies, anticipating substantial gains from next-gen AI server design upgrades.
**Explosive Growth in AI Server Rack Demand** The report notes that AI hardware growth is transitioning from the H100/H200 era to a new cycle driven by NVIDIA's GB200/300 (Blackwell platform) and the upcoming Vera Rubin (VR series) platform.
NVIDIA's product roadmap shows a leap in GPU power and performance—from the H100's 700W TDP to the B200's 1,000W, GB200's 1,200W, and the Vera Rubin (VR200) platform's 2,300W TDP in late 2026, with VR300 (Kyber architecture) reaching 3,600W by 2027.
Morgan Stanley forecasts that AI server rack demand for NVIDIA platforms alone will more than double from ~28,000 units in 2025 to at least 60,000 units in 2026. Meanwhile, AMD's Helios server rack project (based on MI400 series) is progressing well, further boosting demand for advanced AI hardware.
The report emphasizes that as rack designs shift from single-GPU upgrades to integrated rack systems, ODM manufacturers like Quanta, Foxconn, Wistron, and Wiwynn will dominate GB200/300 rack supply due to their strong integration capabilities and stable delivery records.
**Power & Cooling Bottlenecks Drive Value Surge** One of the report's key takeaways is that AI hardware upgrades are creating significant opportunities for power and cooling suppliers.
With rising rack power consumption, traditional power architectures are becoming inadequate. The report predicts a shift to 800V high-voltage DC (HVDC) solutions, substantially increasing power solution value. By 2027, power solutions for Rubin Ultra racks (Kyber architecture) could be worth 10x more than current GB200 server racks.
Liquid cooling is now a necessity, especially as GB300 platform TDP exceeds 1,400W. Cooling component value is rising—a GB300 (NVL72) rack's cooling components total ~$49,860, while Vera Rubin (NVL144) racks will see a 17% increase to $55,710, with switch tray cooling modules surging 67%.
Suppliers of cold plates, cooling fans, and quick connectors (NVQD) stand to benefit significantly.
**Full Value Chain Upgrade: PCBs & High-Speed Interconnects** The report also underscores the impact of AI platform upgrades on PCBs and interconnects. Each GPU iteration demands higher PCB layers, material grades, and sizes.
Key upgrades include: - **ABF substrates**: Layer count increases from 12L (H100) to 14L (Blackwell/B200) and 18L (Vera Rubin/VR200). - **OAM motherboards**: PCB specs upgrade from 18L HDI (H100) to 22L HDI (GB200/300), potentially reaching 26L HDI (VR200). - **CCL materials**: Transitioning from ultra-low-loss (M7) to extreme low-loss (M8) for higher data rates.
These advancements mean more complex, higher-value PCB manufacturing, creating growth opportunities for suppliers with technical expertise.
The report concludes that AI GPU and ASIC server upgrades will drive expansion in PCB/substrate production, while data and power interconnects evolve to meet higher-speed demands.