NVIDIA’s Dominance and Semiconductor Supply Chain Limitations: Insights from Morgan Stanley’s Asia Value Chain Analysis on the New Phase of the AI Semiconductor Market
Editor
A New Paradigm in the Semiconductor Ecosystem Driven by Explosive AI Demand
As of the end of 2025, the AI semiconductor market stands at the center of structural change, surpassing mere growth. According to Morgan Stanley’s recent Asia Value Chain meeting, AI demand has created an unprecedented cycle, pushing the entire semiconductor supply chain to its limits. This phenomenon is seen as a restructuring of the industry’s supply structure, beyond the improved performance of specific companies.
Notably, despite the emergence of alternative solutions such as TPUs (Tensor Processing Units) and ASICs (Application-Specific Integrated Circuits), NVIDIA’s market dominance is becoming even more solidified. The common key issue mentioned by all Asia Value Chain companies was “how many NVIDIA chips can be secured,” with intense competition for the next-generation architecture, Vera Rubin, being particularly highlighted. This indicates that the AI semiconductor market is expanding on a much larger scale than anticipated, and the competitive landscape is unfolding in a direction different from previous predictions.
The reality of the data center AI accelerator market is clearly reflected in the numbers. NVIDIA recorded quarterly revenue of $51 billion in the data center segment, approximately 14 times the total revenue of Google’s TPU. More intriguingly, even in workloads with a high TPU share, many cases have been confirmed where NVIDIA’s actual revenue contribution is greater. Although TPUs appear numerous in volume, the revenue contribution structure is overwhelmingly dominated by NVIDIA.
Based on this market reality, Morgan Stanley significantly raised its outlook for NVIDIA. With increased confidence in the growth of the NVIDIA platform compared to last July, they raised the CY27 (2027) revenue forecast and the target price from $235 to $250. This is based on a conservative verification of the $50 billion forecast over five quarters mentioned at GTC, with the actual market demand likely to exceed this.
Broadcom’s AI Leverage Expansion and Structural Performance Improvement
Following NVIDIA, Broadcom, which ranks second in AI exposure in absolute terms, is also gaining attention for its growth story. Morgan Stanley adjusted Broadcom’s CY27 performance estimates and raised the target price from $409 to $443. This is due to confirmed AI leverage expansion across all product lines, including AI ASIC, networking, and virtualization (EM/gateway). It is particularly recognized as the company benefiting the most in the TPU, ASIC, and networking fields.
Broadcom’s strength lies in its comprehensive portfolio across various AI infrastructure components. With the expansion of CapEx by hyperscale companies, it is analyzed that maintaining a premium valuation is possible in the medium term. Although the inventory adjustment of the networking and storage cycle has been prolonged beyond expectations, the non-AI segment is expected to rebound from the end of 2026. The integration of VMware is acting as a factor strengthening the stability of the entire portfolio through cost reduction and cash flow stabilization.
Notably, Broadcom is not merely riding on NVIDIA’s growth but is creating independent value across various layers of AI infrastructure. By comprehensively providing key components necessary for AI data center construction, from custom ASIC design to high-speed networking solutions and virtualization platforms, it is showing growth directly linked to the expansion of AI infrastructure investments by client companies.
Unprecedented Supply Shortage and Market Restructuring in Memory Semiconductors
One of the most shocking findings from the Asia Value Chain meeting was the supply shortage in the memory semiconductor market. Morgan Stanley announced that an unprecedented level of supply shortage was confirmed in all areas of DRAM, HBM, and NAND during their 30-year coverage. This is more serious as it is based on actual demand rather than simple double ordering.
In the case of DDR5 memory, once inventory is depleted, it has entered a phase where replacement is impossible, with data center OEMs’ purchase volumes significantly exceeding expectations. The shortage is spreading to non-AI demand sources, such as automotive OEMs, even affecting DDR4. This shows that AI demand is not only creating new markets but also restructuring the entire supply structure of the existing memory market.
The NAND flash memory market is similarly facing a severe supply shortage. QLC enterprise NAND is in the most critical situation, and the shortage is in its early stages of spreading to the consumer sector. Although there is a slight price adjustment risk for HBM (High Bandwidth Memory) in the first quarter, the strength of DDR5 is functioning to protect HBM prices, with overall memory price pressure expected to continue.
For Micron Technology, while the initial market share in HBM4 is expected to be low, the high profitability of DDR5 offsets this, so it is not expected to be a major variable in overall performance. This indicates that memory companies are forming a structure where they can secure profitability across their portfolio without relying on a single product line.
TSMC’s situation is also noteworthy. Wafer supply has reached full capacity at all nodes of 3nm, 4nm, and 5nm, and with AI demand accelerating, they are considering additional expansion of 3nm. CoWoS (Chip-on-Wafer-on-Substrate) packaging has become one of the most severe bottleneck processes due to AI acceleration. This suggests that the entire ecosystem, from equipment companies like Applied Materials to memory, is simultaneously entering a supply constraint cycle.
In the CPU market, AMD’s Turin processor is dominating. While the server CPU market is not at a ‘crisis’ level, it is clearly in a supply shortage phase, with AMD Turin’s performance and power efficiency superiority firmly maintained. Intel’s 18A process is judged to have failed to close the performance gap due to the limitations of the BSP (Booster Separation Program), and AMD is expected to maintain its lead in the server CPU market through 2026.
For Astera Labs, with the confirmed adoption in AWS Trainium 3, a large-scale ramp is expected in 2026. The adoption of Astera Scorpio X solutions in the PCIe scale-up configuration of Trainium 3 has been confirmed, and rumors of NVLink Fusion transition or Broadcom Ethernet scale-up transition have been clarified as false. The 2026 revenue is very strongly confirmed, with the acquisition of customers beyond AWS remaining a key variable for TAM (Total Addressable Market) expansion.
China’s efforts in semiconductor localization continue but still face structural constraints. Attempts to advance using DUV-based multi-patterning are underway, but scale-up is limited in key processes such as Epi (epitaxial growth), RTP (Rapid Thermal Processing), laser annealing, and inspection equipment. Chinese AI chips have limitations in cluster size, software ecosystem, and scale-out capabilities, and while there is an expansion of DRAM and NAND production in China, it is not analyzed to disrupt the global balance.
From this comprehensive analysis, it is evident that the current AI semiconductor market is in a phase of structural change beyond a simple growth cycle. NVIDIA, Broadcom, AMD, and Micron are all expected to maintain continuous performance upward pressure over multiple quarters, and AI acceleration is simultaneously causing bottlenecks across the entire supply chain, from wafers (3nm) to packaging (CoWoS) and memory (HBM/DDR5). While TPUs and ASICs are expanding, this is leading to a market expansion of simultaneous surges in GPUs, ASICs, and TPUs, with memory showing an unprecedented supply shortage in 30-year coverage, indicating a new paradigm of AI-centric demand realignment.
For investors, this suggests a focus on mid- to long-term structural growth rather than short-term volatility. As constraints across the supply chain occur simultaneously, the execution capability of individual companies and the expandability of the entire ecosystem are expected to become the core investment themes in the semiconductor industry for the coming years.
—
**Disclaimer**: This content is for informational purposes only and is not intended as investment solicitation or advice. Investment decisions should be made at one’s own discretion and responsibility, and no liability is accepted for investment losses based on this information.
