Nvidia's AI Rack Is Almost Doubling In Value. Which Suppliers Win?

Nvidia's next AI rack cycle may be bigger than just another GPU upgrade. According to the Morgan Stanley BOM estimate, the value of a single VR200 NVL72 rack could reach about $7.80 million, up 95% from about $3.99 million for GB300.

That matters because the value increase is spread across the whole AI rack. GPU value rises, but the bigger investment question is which non-GPU hardware layers get repriced as AI systems become denser, more power hungry and more memory intensive.

Memory is the biggest value jump

The standout line item is memory. Morgan Stanley estimates memory content rising from about $374,000 in GB300 to about $2.00 million in VR200, up 435%. That is the clearest bottleneck signal in the rack.

The most relevant U.S.-listed name is $Micron Technology (MU.US)$ , given its DRAM and HBM exposure. Micron has also said it began volume shipments of HBM4 36GB 12H in Q1 calendar 2026, designed for $NVIDIA (NVDA.US)$ Vera Rubin.

Storage-adjacent names also matter as AI workloads move from training into inference and agentic use cases. $Seagate Technology (STX.US)$ and $Western Digital (WDC.US)$ are the most obvious HDD and cloud storage beneficiaries, while $SanDisk (SNDK.US)$ gives investors public NAND and flash exposure after its 2025 spin-off from Western Digital.

PCB, substrate and MLCC become hidden winners

The next big value shift is in the "small but critical" hardware layers. Morgan Stanley estimates PCB value rising 233%, MLCC rising 182%, and ABF substrate rising 82%.

For U.S.-listed exposure, $TTM Technologies (TTMI.US)$ is the cleaner PCB-related name. $Jabil (JBL.US)$ is broader electronics manufacturing and PCB-adjacent proxy. For ABF substrate, direct U.S.-listed exposure is limited because key suppliers are mostly Asia-listed. $Amkor Technology (AMKR.US)$ is better viewed as an advanced packaging and test proxy, rather than a direct ABF substrate pure play.

Japan-listed suppliers are also important here. $Ibiden (4062.JP)$ is the standout ABF substrate name, with a roughly ¥500 billion FY2026-FY2028 investment plan to expand high-performance IC package substrate capacity for AI servers and high-performance servers. $Meiko Electronics (6787.JP)$ is a key PCB name, helped by fast-growing AI server PCB and IC substrate demand. For MLCC and passive components, investors usually watch $Murata Manufacturing (6981.JP)$ , $TDK (6762.JP)$ and $Taiyo Yuden (6976.JP)$ .

Power and cooling are no longer side stories

Power supply value rises 32%, while cooling rises 12%. These numbers may look smaller than memory or PCB, but they sit in a structural bottleneck category. More powerful AI racks need more stable power delivery and better heat management.

Relevant U.S.-listed beneficiaries include $Vertiv Holdings (VRT.US)$ for data center power and thermal infrastructure, $Trane Technologies (TT.US)$ as a broader cooling proxy, and $Flex Ltd (FLEX.US)$ for power and electronics manufacturing exposure.

These are not pure Nvidia rack suppliers. They are infrastructure proxies for the same problem: AI compute is becoming power dense, and power density creates spending opportunities outside the chip itself.

Networking and interconnect still matter

Morgan Stanley estimates NVLink switch chip value rising 122% and other networking chips rising 121%. GPU, CPU and NVLink are largely Nvidia-controlled value pools, but the broader data movement theme still matters for public-market investors.

The U.S.-listed proxies are $Broadcom (AVGO.US)$ , $Marvell Technology (MRVL.US)$ and $Astera Labs (ALAB.US)$ . These companies sit around AI connectivity, switching, optical DSP, retimers and high-speed data movement. Nvidia's own networking business also shows how large this layer can become. In Q1 FY2027, Nvidia Data Center networking revenue reached $14.8 billion, up 199% year over year and 35% sequentially.

Rack assembly turns complexity into value

Morgan Stanley estimates rack assembly value add rising 29%, while the "others" bucket rises 55%. This reflects a simple reality: Nvidia racks are becoming full systems, not loose collections of chips.

Relevant U.S.-listed system integration names include $Super Micro Computer (SMCI.US)$ , $Dell Technologies (DELL.US)$ , $Hewlett Packard Enterprise (HPE.US)$ , $Flex Ltd (FLEX.US)$ and $Jabil (JBL.US)$ . The opportunity is not just building servers. It is full rack integration, validation, cooling, power delivery, supply-chain execution and faster deployment.

Summary

Morgan Stanley's VR200 rack BOM suggests the AI hardware trade is moving beyond the headline GPU. The biggest incremental value pools are memory, PCB, MLCC, networking, power, cooling, storage and rack integration.

Nvidia remains the center of the AI infrastructure cycle, but the next investment question is more specific: who owns the bottleneck inside the rack?

123

1.5M Views