The demand for AI computing power continues to explode, with 140 trillion tokens driving the full chain upgrade of electronic components

In March, the daily average token call volume in China's AI field officially exceeded the 140 trillion mark. This milestone data not only marks the outbreak of large-scale inference applications in China, but also becomes the core lever to leverage the electronic components industry to shift from "consumer electronics driven" to "AI computing driven", promoting a systematic restructuring of the entire industry chain in terms of demand structure, technological route, competitive landscape, and domestic substitution.

Token, as the basic unit of information processing in AI models, has an average daily call volume of 140 trillion. Essentially, it is the concentrated release of computing power demand brought about by massive inference requests and multimodal interactions. Its high-frequency, continuous, and large-scale inference scenario characteristics have broken the demand cycle and supply logic of the electronic components industry in the past. The demand for full chain components has been redefined: the demand for computing power, inference specific chips, and edge computing chips has surged, and advanced packaging technologies such as Chiplets have become essential; On the storage side, HBM has become the core bottleneck, with iteration accelerating and the supply-demand gap widening. DRAM and NAND Flash are upgrading to higher specifications; Connecting the transmission end, the value and shipment volume of high-end interconnect devices such as high-speed optical modules and optical chips have both increased; Energy and passive end, the high power consumption of AI servers has driven an explosion in demand for third-generation semiconductor power devices, leading to specification upgrades and demand expansion for high-end passive components.

The most significant change in this industry restructuring is the shift in demand focus: in the past, the growth of the electronic components industry was tied to consumer electronics, and demand fluctuated periodically; Nowadays, AI computing infrastructure and inference servers have become the main drivers of new growth, and the demand for computing power related components has shifted from "periodic pulses" to "sustained demand". The industry's growth logic has been upgraded to be driven by computing power density and token processing efficiency. At the same time, the competitive landscape of the industrial chain is accelerating its reshaping: top overseas manufacturers dominate with core technological barriers, leading to an increase in industry concentration; The market gap brought by 140 trillion tokens has opened a breakthrough window for domestic manufacturers, accelerating technological breakthroughs in areas such as domestic computing chips, HBM matching, and advanced packaging. The domestic substitution of high-end devices has entered the stage of large-scale implementation, and the independent and controllable process of the supply chain has accelerated.

Deeper restructuring is also reflected in the coordinated upgrading of technological routes and industrial ecosystems. The extreme requirements for computing power efficiency ratio, integration degree, and cost in token scale processing have forced the electronic components industry to break away from the path of upgrading single devices and shift towards collaborative innovation of "chip packaging material system": third-generation semiconductor materials replace traditional silicon-based devices, advanced packaging integrates multi chip functions, high-frequency and high-speed materials support data transmission, and computing collaborative technology reduces computing power energy consumption, forming a closed loop through technological iteration of the entire industry chain. At the same time, the industry ecosystem has also extended from "hardware supply" to "computing power adaptation". Component manufacturers have begun to deeply bind with large model manufacturers and computing power service providers, developing customized devices around specific scenarios of token processing. The coupling degree between hardware, AI algorithms, and computing power scheduling continues to improve, promoting the upgrade of electronic components from "standardized components" to "core components of computing power infrastructure".

The breakthrough of 140 trillion tokens per day is not only the starting point of short-term dividends in the industry, but also a watershed for long-term transformation. For the electronic components industry, this round of AI driven restructuring is both an opportunity - the demand for computing power has opened up growth space far beyond consumer electronics, and domestic substitution has ushered in the best window period; It is also a challenge - high-end technology bottlenecks, mismatched production capacity, energy consumption pressure, intensified global competition, and other issues still constrain the development of the industry. In the future, with the continuous increase in token usage and the full implementation of multimodal AI and AI intelligent agents, the electronic components industry will continue to deepen its "computing power driven" transformation. Only by grasping the three core areas of technological innovation, domestic substitution, and ecological synergy can it take the lead in the reconstruction of the entire industry chain and support China's AI industry to move from large-scale applications to high-quality development.