We believe understanding the roadmaps—and roadblocks—of the globe’s largest AI ecosystems is essential to seeking alpha within broader global technology portfolios. Here we contrast the development of AI in the U.S. and China, and highlight some potential investment implications arising from similar challenges across both markets.
Different Approaches
Broadly speaking, the U.S. AI industry has sought optimal model performance and end-to-end control of its ecosystem, while China has focused on spreading its technology quickly, at lower cost, and threading it into myriad applications.
Using smartphones as a comparison, U.S. AI development has tracked the iOS model, while China’s has mirrored Android. Apple’s iOS is a closed, tightly managed ecosystem where hardware and software are deeply integrated; the experience is polished and smooth; and prices reflect premium design and performance. By contrast, Android offers a more open ecosystem with many device makers, more affordability and broader reach and customization.
In the U.S., generative AI models like Gemini, ChatGPT and Claude command premium pricing to compensate for superior computational horsepower, accuracy and security features. In China, open-source and lower-cost models—from the likes of DeepSeek, Alibaba and Tencent—aim for “good enough” performance, fast customization and on-device or hybrid deployment. Higher-cost U.S. players have been aiming to accrue significant value over time as their competitive moats theoretically deepen, while lower-cost Chinese competitors have focused on monetizing their technologies more quickly through faster adoption.
Common Choke Points
While the U.S. and China continue to take different approaches to AI development, we find common constraints within both countries’ AI ecosystems. In our view, these choke points are creating potentially attractive opportunities for active investors.
Memory chips
As demand for advanced AI chips continues to skyrocket, supply of legacy DRAM and flash memory chips has tightened sharply: Heavyweights, including SK Hynix and Micron, have shifted production to more advanced, high-bandwidth chips (HBM) to meet the AI boom. As a result, prices for legacy chips may rise as much as 20% in 2026, estimates Counterpoint, a research firm.1
Now, several global chipmakers are jumping into the breach. Last November, Micron Technology announced a $9.6B investment to build a new HBM4 manufacturing facility in Hiroshima, Japan.2 Samsung said it aims to build a new semiconductor research-and-development complex at its Gi Heung campus in South Korea, with the goal of accelerating mass production of memory chips.3 And China’s ChangXin Memory Technologies plans to begin mass-producing HBM3 chips by the end of 2026.4
Network Connectivity
Another primary AI bottleneck, in both the U.S. and China, is network connectivity. In the U.S., mounting pressure on the electrical grid is capping the size of new data centers, forcing hyper-scalers to build smaller facilities in different locations and connecting them with high-speed optical networking to support further frontier-model development.
Ultimately, we believe the challenge comes down to near-term supply and demand: Adding enough global optical-networking manufacturing capacity will take time. And with the U.S. looking to reduce its reliance on Chinese suppliers (who now dominate the optical-component market) and expand its supply chain, we expect network connectivity will remain a common AI choke point for the next few years.
Potential Winners: Semiconductor Equipment Makers
In light of these critical bottlenecks—memory, networking—in both the U.S. and China, we believe makers of advanced semiconductor equipment for churning out next-generation memory chips and optical-networking chips (which manipulate light and transmit data, offering both high performance and energy efficiency) may be poised to benefit in the near-to-medium term. Global semiconductor equipment sales are expected to hit $133B in 2025, up 13.7% from the previous year, then rise to $145B in 2026 and peak at $156B in 2027, according to SEMI.5
In the U.S., we believe select suppliers of memory chips and network-testing equipment appear well positioned to meet rising demand, and in China, we suggest keeping an eye on semiconductor-equipment manufacturers.
