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

```

The competition for future AI infrastructure will no longer be limited to GPUs, but will gradually evolve into a trend of joint development of GPUs and various AI-specific processors (XPU).

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

The global AI semiconductor market size will reach about $485 billion in 2026, and is expected to further grow to about $753 billion in 2030, accounting for about half of the $1.5 trillion global semiconductor industry market size.

In its supply chain data-driven bull scenario, it is estimated that cloud capital expenditure in 2026 will be $796 billion, with AI server capital expenditure about $600 billion, and cloud AI ASIC and non-NVIDIA GPU market size around $90 billion.

The development focus of the AI industry is gradually shifting from model training to inference application, making computing power demands increasingly diverse. GPUs will still maintain a core position in training and high-performance computing, but AI ASICs, NPUs and other scenario-specific XPUs are rising rapidly, becoming important tools for cloud providers to optimize costs and improve efficiency.

For the semiconductor industry as a whole, this means that the winners of the AI era will no longer be just GPU companies, but will cover multiple segments including chip design, advanced manufacturing, advanced packaging, testing, and AI-specific chips. The value distribution of the industry chain is entering a new stage.

GPUs No Longer "Dominating Alone", AI Computing Power Enters a Diversified Era

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

As AI applications continue to diversify, major cloud providers have begun developing custom chips around their own models and business needs. Even with continuous improvement in GPU performance, cloud providers still need to deploy a large number of AI ASICs to improve inference efficiency, reduce total ownership costs, and optimize for different workloads.

In the future, AI infrastructure will show a trend of coordinated development between GPUs and XPUs.

Here, XPU is not a single product, but encompasses AI ASICs and various dedicated processors for AI computing scenarios. As the demands for computing power of training, inference and agentic AI tasks are increasingly subdivided, chips of different architectures will play their roles in their respective specialized scenarios.

Cloud Providers Continue to Increase Capital Expenditure, AI Value Chain Extends to Advanced Manufacturing and Packaging

Investment in AI infrastructure is still in the expansion phase. The capital expenditure of Amazon, Google, Microsoft, and Meta — the four major cloud providers — is expected to increase by 95% year-on-year in the first quarter of 2026, with the capital expenditure as a proportion of EBITDA expected to remain at around 50%. Global major listed cloud providers' capital expenditure on cloud computing in 2026 will approach $811 billion.

The ongoing increase in capital investment not only drives demand for GPUs and AI ASICs, but also brings synchronized expansion in advanced process technology, advanced packaging, and test equipment segments of the industry chain.

TSMC's CoWoS advanced packaging capacity will continue expanding in 2027, and advanced packaging technologies like SoIC will also become key development directions in the next few years. Meanwhile, the demand for AI computing wafers continues to grow, further raising the importance of advanced process and packaging segments.

The future competition focus in the AI industry chain is not just about the chip itself, but about the entire AI infrastructure system, including wafer manufacturing, advanced packaging, testing and system integration, among multiple segments.

It is worth noting that rising costs in wafers, OSAT, and memory, and AI's crowding out resources for non-AI chips, may further pressure chip design companies' profit margins in 2026.

Rising Inference Demand, China’s AI Chips Welcome Development Window

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

DeepSeek has validated the feasibility of low-cost AI inference, driving rapid growth in inference demand, and also boosting opportunities for the domestic AI GPU industry chain. According to the report, by 2030, China's AI GPU market size is expected to reach about $91 billion, and the self-sufficiency rate of domestic AI chips is expected to increase to about 70%.

With the gradual expansion of China's advanced process capacity, domestic AI chips will continue to strengthen their competitiveness in inference scenarios, and AI infrastructure construction will increasingly rely on the local supply chain.

Competition in the AI Era: From "Who Owns GPUs" to "Who Owns a Complete Computing Power Ecosystem"

The future competition logic in the AI industry will shift from single chip performance competition to competition over the entire computing power system.

Future AI industry will need to pay close attention to structural changes between training and inference, cloud and edge, GPU and custom ASICs, while budget, energy, chip production capacity and regulation will remain as key constraints on AI development.

For the market, this means that the main line of AI investment is further expanding. GPUs remain an important component of AI infrastructure, but as XPUs grow richer, cloud providers accelerate the development of their own chips, and AI inference demand expands rapidly, winners in the future AI era are more likely to emerge from the entire AI computing power ecosystem, rather than a single technology route.

~~~~~~~~~~~~~~~~~~~~~~~~

The above content is from Chase Trading Desk.

For more detailed interpretation, including real-time commentary and frontline research, please join [Chase Trading Desk▪Annual Member]

Risk Reminder and DisclaimerThe market has risks, investment needs to be cautious. This article does not constitute personal investment advice, nor does it take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, viewpoints, or conclusions in this article fit their particular circumstances. Investments made accordingly are at their own risk. ```