Morgan Stanley Heavyweight Report: GPU vs XPU, Who Will Win?

Future competition in AI infrastructure will no longer be limited to GPUs, but will gradually evolve into a landscape where GPUs and various AI-specific processors (XPUs) develop together.

According to the Chase Trading Desk, Morgan Stanley's latest semiconductor report points out that as cloud computing providers continue to expand capital expenditures, AI inference demand grows rapidly, and custom chips accelerate adoption, the value chain of the AI semiconductor industry is undergoing new changes.

The global AI semiconductor market is expected to reach approximately $485 billion in 2026, and is projected to further grow to about $753 billion by 2030, accounting for roughly half of the global semiconductor market's approximately $1.5 trillion size.

In its supply chain data-driven bull case, cloud capex is expected to be $796 billion in 2026, with AI server capex at about $600 billion and cloud AI ASIC and non-NVIDIA GPU spending at approximately $90 billion.

The focus of AI industry development is gradually shifting from model training to inference applications, making compute demand more diverse. GPUs will still maintain a core position in training and high-performance computing, but AI ASICs, NPUs, and other XPUs designed for specific scenarios are rapidly rising, becoming important tools for cloud service providers to optimize costs and improve efficiency.

For the entire semiconductor industry, this means the winners of the AI era will no longer be just GPU manufacturers, but will cover multiple segments including chip design, advanced manufacturing, advanced packaging, testing, and AI-specific chips, as the industry value chain enters a new phase of distribution.

GPUs are no longer the single dominant force; AI computing enters a diversified era

In the past few years, AI computing power has been almost entirely dominated by GPUs, but this landscape is changing.

As AI applications continue to enrich, major cloud service providers have begun to develop custom chips based on their own models and business needs. Even though GPU performance continues to improve, cloud providers still need to deploy a large number of AI ASICs to improve inference efficiency, reduce total cost of ownership, and optimize for different workloads.

Future AI infrastructure will show a trend of GPU and XPU collaborative development.

Among them, XPU is not a single product, but covers various dedicated processors for AI computing scenarios, such as AI ASICs. As different tasks like training, inference, and agentic AI continue to refine their computing demands, chips with different architectures will play roles in their respective areas of expertise.

Cloud providers continue to increase capital expenditures, and the AI value chain is extending to advanced manufacturing and packaging

AI infrastructure investment is still in an expansion phase. The four major cloud service providers—Amazon, Google, Microsoft, and Meta—saw year-over-year capital expenditure growth of 95% in Q1 2026, with capex as a percentage of EBITDA expected to remain at about 50%. Global major listed cloud service providers' cloud capex in 2026 will approach $811 billion.

The sustained increase in capital investment not only drives demand for GPUs and AI ASICs but will also drive simultaneous expansion in the supply chain segments such as advanced process nodes, advanced packaging, and test equipment.

TSMC's CoWoS advanced packaging capacity will continue to expand in 2027, and advanced packaging technologies like SoIC will also become key development directions in the coming years. At the same time, the growing demand for AI computing wafers will further enhance the importance of advanced process nodes and packaging.

The focus of future AI industry competition is no longer just the chip itself, but the entire AI infrastructure system, including multiple segments such as wafer manufacturing, advanced packaging, testing, and system integration.

It is worth noting that rising costs of wafers, OSAT, and memory, as well as the resource squeeze from AI on non-AI chips, may increase profit margin pressure on chip design companies in 2026.

The rise of inference demand opens a development window for Chinese AI chips

The focus of AI industry development is shifting from training to inference, and this change is driving the development of China's AI chip supply chain.

DeepSeek has verified the feasibility of low-cost AI inference, driving rapid growth in inference demand, while also enhancing development opportunities for the domestic AI GPU supply chain. The report estimates that by 2030, China's AI GPU market size is expected to reach approximately $91 billion, and the self-sufficiency rate of domestic AI chips is expected to increase to about 70%.

As China's advanced process capacity gradually expands, domestic AI chips will continue to enhance their competitiveness in inference scenarios, and AI infrastructure construction will increasingly rely on local supply chains.

Competition in the AI era shifts from "who owns GPUs" to "who owns a complete computing ecosystem"

The competitive logic of the future AI industry will shift from single-chip performance competition to competition over the entire computing system.

The future AI industry needs to focus on structural changes between training and inference, cloud and edge, GPUs and custom ASICs. At the same time, budgets, energy, chip capacity, and regulation will remain major constraints on AI development.

For the market, this means the AI investment theme is further expanding. GPUs remain an important part of AI infrastructure, but as XPU diversity increases, cloud providers continue to advance their in-house chips, and AI inference demand grows rapidly, the winners of the future AI era are more likely to come from the entire AI computing ecosystem, rather than a single technology path.

The above exciting content comes from the Chase Trading Desk.

For more detailed interpretations, including real-time analysis and frontline research, please join [**Chase Trading Desk · Annual Membership**]

Risk Warning and Disclaimer

          

            Market risk exists, and investment should be cautious. This article does not constitute personal investment advice, nor does it consider the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investments made based on this article are at your own risk.