TL;DR

·During WAIC, UBS conducted research on 12 tech companies and further expressed optimism about the Chinese AI industry chain. The core reason is not just the concept's popularity, but the simultaneous enhancement of model iteration, local AI computing power utilization, and infrastructure demand.

·Semiconductors and equipment became the fastest-growing sector in institutional research interest in July, reflecting funds starting to flow towards domestic chips, equipment, and advanced packaging due to AI demand.

·Domestic AI infrastructure is transitioning from single-chip competition to SuperPods, network interconnection, and system optimization, essentially bridging the performance gap through cluster capability.

·ZhuiFu and MiniMax have both set their annual recurring revenue targets at $1 billion, but whether revenue growth can translate into profit still depends on inference optimization and computing power leasing costs.

·Longsys Technologies maintains high utilization of AI-related capacity and is ramping up 2.5D packaging, indicating that the volume production of domestic AI chips is moving towards the packaging stage.

·UBS believes that the software and services sector is still in a low-crowded zone and may retain more upside potential compared to the already heated semiconductor sector.

·AI data center demand is also spilling over to energy storage and power equipment, but the order fulfillment cycle, rising lithium prices, and low-price market competition will still suppress short-term margins.

On August 14th, UBS released "China On the Ground - July 2026." The report combines WAIC 2026 tech research, management discussions with companies, and institutional research data to provide a clearer assessment: investment clues in the Chinese AI industry are spreading from large models themselves to AI computing chips, servers, network interconnection, advanced packaging, and power infrastructure.

During WAIC, UBS organized approximately 40 investors to research 12 tech companies, covering large models, GPUs, networks, analog chips, packaging testing, server ODMs, and semiconductor equipment. The core conclusion after the research is that Chinese large models are still iterating rapidly, the product form of domestic AI infrastructure is also improving, and downstream demand for AI computing power has not significantly weakened.

Domestic AI Competition Shifts from Single Cards to Systems

UBS observed at WAIC that domestic manufacturers are no longer focusing solely on individual chips but are showcasing system-level solutions composed of chips, servers, interconnection networks, and cluster management.

Huawei showcased the Atlas 950 SuperPod, consisting of 1024 chips; Inspur presented the Sugon 8000 solution supporting a hundred-thousand-card-level interconnection; Matrix Time, ZTE, and Alibaba also launched their own super node products. On the networking side, relevant vendors began showcasing a 51.2T CPO switch prototype and a complete optical interconnection solution.

This shift signifies that domestic computing power is reducing the impact of single-chip performance differentials through expanding cluster scale, improving interconnection efficiency, and optimizing system architecture. For investors, the benefits are expanding from GPUs to include switching chips, optical interconnects, server ODMs, storage interfaces, and advanced packaging.

Domestic AI accelerator manufacturers reported back to UBS that demand from cloud service providers and other customers continues to strengthen. On one hand, high-end computing power from overseas remains scarce in the Chinese market; on the other hand, domestic cloud providers are increasingly willing to adopt domestically produced computing cards. Some second-tier manufacturers are developing products to compete with the H100 and H200, while also iterating on SuperPod technology.

Model Revenue Growth, Computing Power Cost Remains Key Variable

The commercialization progress of large-scale models is also accelerating, but the quality of growth still depends on computing power costs.

According to the SmartPeak management team, the company's annualized recurring revenue reached $1 billion as of July, surpassing the original target of the end of 2026 ahead of schedule, mainly driven by growth in Token usage and API price increases. The company recently completed a HK$31 billion H-share placement, with the funds supporting computing power expansion while also considering acquiring larger-scale computing resources through leasing.

UBS believes that API price increases and inference optimization may drive SmartPeak's profit margin improvement, but the tight domestic computing power supply and rising leasing costs remain important variables for profitability. In other words, being ahead in revenue does not necessarily mean that the unit economics of model companies have stabilized.

MiniMax's Annual Recurring Revenue (ARR) increased from around $100 million at the end of 2025 to approximately $150 million disclosed in the first-quarter performance report and further grew by over double before the M3 release. The management team remains confident in achieving the $1 billion ARR target by the end of 2026. The proportion of API and Token Plan business revenue has increased from around 30% in 2025 to nearly half by May 2026, indicating that growth momentum is shifting further from consumer subscriptions to developer and enterprise calls.

However, MiniMax's new generation M3 model has a parameter scale close to twice that of M2.7. Larger models typically mean higher inference costs, and the company needs to rely on inference engineering optimization, cluster operations, and pre-purchased computing power to maintain gross margins.

Semiconductor Research Heating Up, Advanced Packaging Leading the Way

Institutional interest in China's AI is also shifting towards the hardware side.

According to UBS Quantitative Research, the top three sectors with the fastest month-over-month increase in institutional research coverage in July were semiconductors and semiconductor equipment, technology hardware and equipment, and banks; while the research coverage of industrials, energy, and food & beverages sectors saw the most significant declines.

Among them, research interest in the software and services sector has increased, but the crowding indicator remains in negative territory; research interest in materials, food & beverages, and industrials has decreased, yet the positions are relatively crowded. This indicates that semiconductors have become the most prominent direction benefiting from AI-related trading activity, while the software sector may still have some lingering expectations gap.

Looking specifically at the industry chain, Land Semiconductor stated that the third and fourth generation DDR5 memory interface chips are ramping up quickly, with the fifth-generation products expected to begin shipping in the second half of 2026. MRDIMM-related products have become a significant source of new business revenue, and PCIe 6.0 Retimer validation is also making progress.

The prosperity of the advanced packaging stage is more direct. STATS ChipPAC's second-quarter capacity utilization rate is about 80%, with AI-related businesses, wafer-level packaging, and TSV businesses all running at full capacity. The company plans to invest 7.8 billion yuan in building a new factory focusing on 2.5D packaging, with construction expected to be completed in the second half of 2028 and production starting in 2029.

UBS believes that this packaging cycle is different from past cycles driven by consumer electronics, and the structural demand brought by AI may support further improvements in utilization rates and gross margins. However, there is still a considerable amount of time before the new factory goes into operation, and the ultimate returns will depend on the volume production of domestic AI chips, customer adoption, and the pace of industry expansion.

AI Demand Beginning to Spill Over into Power and Energy Storage

The report also extends the observation scope of AI infrastructure to energy storage and power equipment.

Sunshine Power has stated that since the first quarter of 2026, the company has been selling AI data center energy storage systems to developers and integrators, who then supply the products to U.S. cloud service providers. As energy storage is usually a later-stage procurement for data center construction, the order confirmation process is relatively slow, and the likelihood of direct engagement with cloud service providers is low.

The company is also collaborating with U.S. cloud service providers to advance the R&D and validation of solid-state transformers, with product launches expected in the coming months. Solid-state transformers and energy storage systems feature a modular design, which can shorten on-site deployment time and reduce installation costs.

However, AI demand alone cannot fully offset the cyclical pressure on the energy storage business. JA Solar expects that the increase in lithium prices may continue to suppress the gross margin in the second quarter of 2026, and the company's expansion into low-margin markets will also drag down the overall profit level. Therefore, UBS believes that AI data center storage and solid-state transformers provide long-term growth opportunities, but short-term performance still depends on order flow, raw material prices, and overseas supply chain arrangements.

Overall, the signal released by this report is not that the Chinese AI industry has completely overcome all constraints, but that the growth path is becoming more complete: model inference drives the demand for computing power, domestic computing power adoption promotes server and interconnect upgrades, and chip mass production further extends to packaging, equipment, and power infrastructure.

What really needs to be verified in the next stage is no longer just whether domestic models can continue to improve their capabilities, but whether this gradually forming AI technology stack can transform demand into stable income, manageable costs, and sustainable profits.

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia