The four giants are "burning money fiercely," non-U.S. and second-tier cloud manufacturers are undervalued, GB200 yield improvement! Morgan Stanley is very optimistic about AI servers
The arms race among tech giants in the AI field is driving capital expenditures to new highs, but the market's focus should not be limited to this. Morgan Stanley believes that the enormous potential demand from non-U.S. regions and second-tier cloud vendors, along with the improvement in NVIDIA's GB200 yield, together constitute the underestimated growth momentum of the AI server market
A global cloud infrastructure competition driven by AI, costing hundreds of billions of dollars, is heating up at an unprecedented pace.
According to news from the Chase Wind Trading Desk, Morgan Stanley's latest research report has significantly raised its capital expenditure forecasts for the four major U.S. cloud service providers—Amazon, Google, Meta, and Microsoft. The combined capital expenditure of these four companies is expected to reach $359 billion in 2025, a year-on-year increase of 57%; it will further rise to $454 billion in 2026, a year-on-year increase of 26%, both significantly higher than previous forecasts.
This round of "burning money" competition extends far beyond the leading players. The report emphasizes that the market may have seriously underestimated the demand from non-U.S. regions and second-tier cloud service providers. These challengers are actively positioning themselves, and the potential order scale for their AI servers should not be underestimated.
Meanwhile, supply-side bottlenecks are gradually easing, with the much-watched NVIDIA GB200 chip assembly yield continuing to improve. The boom cycle of AI infrastructure may not only be deepening but also spreading to a broader market.
Giants Accelerate Spending, Capital Expenditure Exceeds Expectations
According to Morgan Stanley's report, the capital expenditure growth of the four major U.S. cloud giants is accelerating again. Its U.S. analyst team predicts that the total cloud capital expenditure of these four companies will be $100 billion in the fourth quarter of 2025, a year-on-year increase of 39%.
The team also raised its year-on-year growth forecast for the capital expenditure of these four companies in 2025 from 38% to 57%, reaching $359 billion, and the growth forecast for 2026 was raised from 17% to 26%, reaching $454 billion. Among them, Microsoft and Meta are the two companies with the largest upward adjustments in capital expenditure forecasts for 2026.
When expanding the view to the top 11 global cloud service providers, this investment boom becomes even more apparent. Report data shows that the total capital expenditure of these companies is expected to reach $445 billion in 2025, far exceeding the $400 billion forecast before the earnings season, and roughly equivalent to the total cloud capital expenditure of 2023 and 2024.
Morgan Stanley expects that by 2026, the capital expenditure of these companies will account for more than 20% of their revenue, continuing to set a historical high. This ratio is expected to reach 18% in 2025, hitting a historical peak.
Undervalued Corners of the Market: Second-Tier and Non-U.S. Cloud Vendors are Catching Up
Although the market's attention is mostly focused on U.S. giants, Morgan Stanley believes that another major engine of growth comes from overlooked corners. The report's Asia-Pacific team pointed out that the market may have underestimated the demand for PCIe/HGX servers from non-U.S. countries. Data shows that the demand for B200 servers has shown a strong rebound in the second quarter, and the demand for B300 is expected to continue to grow in the third quarter More importantly, Tier 2 Cloud Service Providers (CSPs) are catching up. Reports indicate that these companies' AI server reserves may even surpass those of leading cloud providers and are expected to significantly increase capital expenditures in the second half of 2026. This suggests that the trend of popularizing AI infrastructure is accelerating, opening up a broader market.
GB200 Yield Improvement, New Projects Gearing Up
The booming demand must be supported by the supply side. The good news is that the supply chain issues previously concerning the market are easing. Morgan Stanley's supply chain survey in the Asia-Pacific region found that the assembly yield of NVIDIA's next-generation flagship product, GB200, is continuously improving.
The report also mentioned that sample testing for GB300 is expected to begin in the third quarter, with no significant issues currently identified. Improvements in the supply chain manufacturing process have cleared key obstacles to meet the surging global AI server orders.
Even more noteworthy is that large-scale projects like "Stargate" are no longer just theoretical. This collaborative project announced by OpenAI, SoftBank, and Oracle in 2025 has begun substantial engagement with the Asian supply chain, particularly in the procurement of server cabinets.
Based on supply chain feedback estimates, a 4.5 gigawatt (GW) data center will require at least 28,000 GB200 NVL72 cabinets to meet its computing power needs. This indicates a significant leap from "order-based" to "project-based" future demand.
In summary, Morgan Stanley maintains a positive outlook on the cloud semiconductor industry, believing that strong global demand, undervalued growth areas, and gradually improving supply chains together form a solid foundation for sustained industry growth in the coming years.
Cloud capital expenditures have a significant impact on global semiconductor revenue, with a correlation of 0.69; Source: Morgan Stanley
The above exciting content comes from [Zhui Feng Trading Platform](https://mp.weixin.qq.com/s/uua05g5qk-N2J7h91pyqxQ).
For more detailed interpretations, including real-time analysis and frontline research, please join the【 [Zhui Feng Trading Platform ▪ Annual Membership](https://wallstreetcn.com/shop/item/1000309)】
[](https://wallstreetcn.com/shop/item/1000309)