Investors should build exposure to Power Grid & Utility Infrastructure and independent energy producers to capitalize on critical power supply bottlenecks limiting US AI data center expansion. Consider a long position in Alibaba Group Holding Limited (BABA) as it monetizes its leading Qwen open-weight models through rising Alibaba Cloud enterprise pricing. Gain long-term thematic exposure to Chinese semiconductor plays, including Cambricon and the Huawei Ascend ecosystem, which are supported by government mandates targeting a cost-competitive domestic AI hardware stack by 2028. Closely monitor and manage long-term exposure to NVIDIA Corporation (NVDA), as low-level software workarounds and domestic chip mandates in Asia threaten its CUDA software moat. In private markets and future public offerings, focus on frontier US AI labs like Anthropic and OpenAI that leverage enterprise workflows and proprietary data lock-in to protect margins.

NVIDIA Corporation (NVDA)

• Chinese AI labs are increasingly coding beneath NVIDIA's proprietary CUDA software layer, writing machine-level assembly instructions and custom kernels to maximize performance on constrained hardware.
• NVIDIA released its open-source Nemotron model, directly competing with software customers by leveraging unique internal insights to maximize throughput on its own hardware architecture.
• Export restrictions on high-end chips like the A100, H100, and H800—combined with conditioned imports on H20 and H200 chips—have accelerated China's government push to mandate domestic chip alternatives in state-funded data centers.

Takeaways

• Monitor NVDA's long-term data center revenue exposure in Asia as Chinese domestic chip substitution and low-level software workarounds reduce dependence on the CUDA software moat.
• NVIDIA's move into open-source model releases (Nemotron) highlights its strategy to offer hardware-optimized AI solutions, protecting margins against commoditized compute.

Alibaba Group Holding Limited (BABA)

• Alibaba is a primary developer of the Qwen open-weight model family, which represents one of the leading open architectures globally.
• The company is shifting its monetization strategy by introducing commercial revenue thresholds on its open-weight models to capture enterprise value.
• Alibaba Cloud, alongside domestic competitors, raised AI cloud pricing during early 2025, signaling tightening compute capacity and efforts to improve cloud infrastructure margins.

Takeaways

• Alibaba is well-positioned to benefit from enterprise adoption of cost-effective open models, especially as it converts broad open-source adoption into billable Alibaba Cloud consumption.
• Keep an eye on Chinese AI cloud pricing normalization as a key metric for domestic chip supply and infrastructure profitability.

Chinese Semiconductor & AI Hardware Sector (Cambricon / Huawei Ascend Ecosystem)

• Chinese AI developers (such as Z.ai / GLM and Moonshot) are engaged in deep hardware-software co-design with domestic hardware providers like Cambricon and Huawei (Ascend).
• Domestic superclusters are being deployed with over 500,000 Ascend chips, delivering roughly 1.3x the total raw compute capacity of large Western clusters like xAI's Colossus.
• Research estimates project a mass-produced, cost-competitive domestic AI hardware stack in China by 2028, even if total output remains around 10% to 15% of aggregate US compute capacity.
• Domestic hardware faces near-term supply constraints with estimated demand-to-supply imbalances around 5:1, making multi-trillion parameter training on purely domestic chips the critical technical hurdle to track.

Takeaways

• Consider long-term thematic exposure to Chinese domestic chipmakers and semiconductor equipment providers that benefit from state subsidies and mandatory domestic replacement policies.
• Track announcements of multi-trillion parameter frontier models successfully trained entirely on domestic chips (Ascend / Cambricon) as the primary catalyst for semiconductor self-sufficiency in the region.

US Frontier AI Labs (OpenAI & Anthropic / Private Market AI)

• Leading US private labs are seeing massive revenue expansion, with specific private players ramping annual recurring revenue (ARR) from $100 million to over $1 billion within months.
• US frontier labs currently trade at relatively attractive revenue multiples compared to Chinese private AI startups, which trade at stretched multiples between 40x to 120x+ price-to-sales.
• High token price compression driven by low-cost Chinese open-weight models is pressuring API margins, forcing US labs to release cheaper "Flash" models and shift focus toward post-training agentic workflows and intellectual property generation.

Takeaways

• For private market investors or those tracking future public offerings (e.g., potential Anthropic IPO), focus on monetization durability, proprietary enterprise data lock-in, and agentic workflows rather than raw base-model pre-training capabilities.
• Watch for policy developments around US cloud infrastructure protections or restrictions on foreign model distillation, which could restore mid-market pricing power to Western API providers.

Power Grid & Utility Infrastructure (AI Energy Sector)

• As AI model weights commoditize and software techniques lower compute requirements, token pricing increasingly reflects the marginal cost of power—behaving like an electricity derivative.
• China currently generates twice the electricity of the US and expands power grid capacity approximately six times faster, with AI power consumption accounting for only 1% to 5% of its newly added capacity (compared to 50% to 70% in the US).
• US data center development faces mounting bottlenecks in grid capacity, interconnection queues, and local permitting, whereas state-supported initiatives like China's "Eastern Data, Western Computing" integrate renewable power with direct energy and tax subsidies.

Takeaways

• Long-term infrastructure investors should prioritize energy producers, utility providers, and grid transmission equipment manufacturers that solve the primary bottleneck in Western AI data center deployment.
• Evaluate independent power producers and clean energy suppliers with direct access to data center buildouts, as power availability will dictate the pace of large-scale AI cluster scaling.