At a recent AI investment summit held in Beijing under the theme "Certain Opportunities in the AI Infrastructure Era," the chief product officer and senior vice president of Moxin (沐曦) shared critical perspectives on the evolving landscape of AI computing power, offering investors a practical framework for evaluating chip-sector opportunities.
The executive outlined a fundamental structural shift underway in AI infrastructure. Traditional single-card, single-chip computing architectures are gradually being phased out, with multi-card clustered "superserver" configurations emerging as the dominant evolutionary direction. Concurrently, the industry's research and application focus has migrated from traditional AI infrastructure to agent-centric infrastructure, marking a decisive pivot in how computing resources are deployed and optimized.
As large language models and industry-specific intelligent agents accelerate their market penetration, the executive predicted that real-world business demands from the broader economy will increasingly shape and redefine the standards for computing infrastructure development. With model iteration cycles compressing rapidly, chip manufacturers now operate at the forefront of industry transformation. To remain competitive, they must conduct daily adaptation work alongside original equipment manufacturer partners, ensuring hardware evolution keeps pace with—or exceeds—the release cadence of upper-layer models. This底层adaptation capability has become the core pillar supporting the successful deployment of AI applications across industries.
Addressing the market's current divergence in AI value distribution, the executive proposed a differentiated pricing logic that challenges the prevailing homogeneous view of AI token worth. Token values vary dramatically across use cases, with casual conversational Q&A tokens commanding relatively low value, while tokens serving vertical industry scenarios—such as mineral exploration and industrial research—carry substantially higher intrinsic worth. Tokens powering intelligent agents applied to complex medical problem-solving and other high-end domains possess exceptional industrial and social value. This value hierarchy indicates that the core dividends from AI computing power concentrate in specialized, vertical application niches rather than in generic, foundational computing infrastructure.
On the domestic autonomy front, the company remains firmly committed to a fully homegrown technology pathway. Navigating an increasingly competitive global semiconductor landscape, the firm operates a "currently shipping, in development, and pre-research" three-generation parallel R&D system, continuously strengthening its domestic computing foundation while deploying critical infrastructure such as superserver clusters. The executive emphasized that computing infrastructure carries core sovereignty attributes; achieving autonomous and controllable domestic computing power is a challenging long-term endeavor, yet it remains the foundational cornerstone for the sustainable, healthy development of China's AI industry.
Offering practical investment guidance, the executive delivered what emerged as a key highlight of the summit: while domestic AI models and computing hardware still trail global best-in-class standards, the crux of industrial development and investment lies not in pursuing flawless, 100-point technology but in the ability to close the loop and achieve commercial deployment with mature, 70-to-80-point solutions. This pragmatic assessment reframes competitive dynamics, prioritizing real-world implementation over theoretical perfection in the race to capture value from the AI revolution.