According to the Zhitong Finance and Economics APP, Guosheng Securities released a research report stating that behind Nvidia's financial report is the long-term sustainability and growth of AI hardware demand. From various perspectives such as technology drive, demand structure, and competitive landscape, the long logic of AI narrative remains unshaken. The report points out that in the short term, capital tends to flow into the AI application field, suppressing AI hardware valuations. However, from the perspective of industry cycles, the value of hardware will eventually be revalued to a reasonable level. The report also mentions that AI applications and hardware complement each other, with the growth in hardware demand and the implementation of AI applications forming a positive feedback loop.

Zhongsheng Securities' main points are as follows:

Financial Highlights: Nvidia's FY25Q3 revenue was $35.1 billion, a year-on-year increase of 94%, a quarter-on-quarter increase of 17%, higher than the company's previous guidance of $32.5 billion (±2%); net income was $19.3 billion, a year-on-year increase of 109%. Among them, Datacenter business revenue was $30.8 billion, a year-on-year increase of 112%, a quarter-on-quarter increase of 17%, driven mainly by strong demand for Hopper and Blackwell chips; Gaming, Professional Visualization, Autos & Robots businesses all achieved year-on-year growth. The company's GAAP gross margin was 74.6%, despite short-term fluctuations, long-term profitability remained stable. In addition, the company's Q4 revenue guidance is $37.5 billion (±2%), further confirming the continuity of demand in the AI hardware market.

AI hardware has substantial growth potential, with a solid narrative logic. Behind Nvidia's financial report is the long-term sustainability and growth of AI hardware demand. From various perspectives such as technology drive, demand structure, and competitive landscape, it can be argued that the long logic of AI narrative remains unshaken:

Exponential growth in computational demand.

The expansion of AI applications is significantly increasing computational demand. From training to inference stages, the scale of generative AI models (such as the GPT series) continues to expand. For example, the latest GPT-o1 released by OpenAI demonstrates that the growth in AI model parameters is leading to higher demand for high-performance computing hardware (strengthening the demand for hardware upgrades and replacements), and the computational demand in the inference stage is further expanding in commercial applications, keeping the demand for datacenter GPUs highly robust.

The popularity and high frequency demand of inference applications.

Different from training, inference applications have higher frequency and wider coverage. As AI extends from central datacenters to edge computing scenarios (such as real-time language translation, autonomous driving, and generative content recommendation), the demand for computing power shows a diffusion from centralized to distributed, driving a structural transformation that will boost the global demand for AI hardware and further stimulate the need for low-latency, high-bandwidth connectivity technologies (such as nvidia's Spectrum-X).

The fundamental position in the ecosystem chain.

AI hardware is the cornerstone of the entire ecosystem. Both CSP manufacturers and AI companies require strong hardware support. Currently, nvidia's products cover training GPUs (H100, A100), inference GPUs (L40, Blackwell), network connectivity (Spectrum-X), and supporting facilities for optimizing AI software. On one hand, an end-to-end solution is forming the 'nvidia ecosystem', while on the other hand, it tightly binds hardware requirements with AI technological developments.

The market's pessimism towards hardware stems more from short-term emotional and financial factors rather than changes in industry fundamentals. In the long term, AI hardware, especially custom chips and communication optical modules, are key components supporting AI training and inference. The widespread adoption of AI applications will further drive the demand for high-performance hardware, accelerating the upgrade iteration trend of optical modules. Furthermore, the competition landscape in the hardware industry is relatively stable, with leading companies having significant advantages in technological barriers and market share. Their profitability and growth potential will not change due to market sentiment fluctuations. The continuous increase in hardware demand driven by AI development, further highlights the advantages of leading companies.

In the short term, funds tend to flow into the AI application field, suppressing the valuation of AI hardware (not due to fundamental reasons), essentially resulting in a market 'high-low switch'. However, from an industry cycle perspective, the value of hardware will eventually be re-evaluated and return to a reasonable level. Therefore, AI hardware and application software are not in a competitive relationship, but rather symbiotic. The growth in hardware demand forms a positive feedback loop with the implementation of AI applications.

Therefore, looking at the current situation and focusing on the layout for spring 2025, we recommend gradually allocating to leading optical module companies represented by 'EasyMandarin', which are expected to welcome a valuation switch next year.

Recommendations:

Computing Power - Optical Communication: zhongji innolight, eoptolink technology inc., suzhou tfc optical communication, t&s communications, tengjing technology, advanced fiber resources, accelink technologies, deten, liant technology, huagong tech, yuanjie technology, cig shanghai, dongguan mentech optical & magnetic. Copper Links: shenzhen woer heat-shrinkable material, tongling jingda special magnet wire. Computing Power Equipment: zte, unisplendour corporation, rege netcom, shengketong communication, feilingxisi, foxconn industrial internet, wus printed circuit, horizon robotics, hkust information. Liquid Cooling: shenzhen envicool technology, shenling environment, guangzhou goaland energy conservation tech. Edge Computing Carrying Platform: meig smart technology, fibocom wireless inc., quectel wireless solutions. Satellite Communication: china satellite communications, china spacesat, genew technologies, guangzhou haige communications group incorporated.

Data Elements - Operators: china telecom, china mobile, china united network communications. Data Visualization: haohan depth, embedway technologies, shenzhen sinovatio technology.

Risk Warning: AI development falling short of expectations, computing power demand falling short of expectations, market competition risks.