How Token Design Influences AI Access and Cost
Carmen Li, founder and CEO of Silicon Data, stated at the CITIC Securities International Investors' Forum in Hong Kong on September 22, 2026, that the global AI competition is moving beyond chip production toward the economic models governing GPU usage and data center operations. She emphasized that investors and firms must now prioritize how computing resources are allocated, priced, and incentivized rather than solely focusing on semiconductor output.
Li explained that while early AI development centered on who could build the fastest or most efficient chips, the current phase involves optimizing the return on investment for AI infrastructure. She noted that tokenomics—the design of incentive systems around computational tokens used to access GPU power—has become a critical factor in determining which companies can scale AI applications sustainably. According to Li, this shift reflects a maturing market where access to computing power is as important as the power itself.
Li detailed that Silicon Data has begun experimenting with token-based models that allow clients to purchase or earn access to GPU time through contribution to network stability or data sharing. These tokens, she said, function not just as payment tools but as mechanisms to balance supply and demand in real time. By tying access to usage behavior and network health, the model aims to prevent hoarding and underutilization of expensive AI hardware. She added that early trials show a 20% improvement in GPU efficiency when token incentives are aligned with workload scheduling.
What Role Will Regulation Play in GPU Markets?
When asked about potential oversight, Li acknowledged that as GPU access becomes commoditized through token systems, regulators may need to examine whether such models create unfair advantages or systemic risks. She warned against treating computational tokens like financial securities without clear distinctions in function and purpose. Li advocated for industry-led standards before government intervention, suggesting that transparency in token issuance and redemption could build trust without stifling innovation.
Li concluded that the winners in the next phase of AI will not be those with the most chips, but those who design the most efficient economic layers around their use. She predicted that firms integrating tokenomics with real-time workload management will gain a durable edge in cost, scalability, and environmental impact. The focus, she said, is shifting from silicon to systems.
Frequently Asked Questions
What does Carmen Li mean by the AI race shifting to GPU economics? She means that competition is no longer just about manufacturing advanced chips but about how GPU resources are priced, distributed, and incentivized through models like token systems that affect access and efficiency.
How does Silicon Data’s token model work in practice? Clients can earn or spend tokens based on their contribution to network stability or data sharing, with token access tied to real-time GPU availability and usage behavior to improve utilization rates.
Why does Li caution against regulating GPU tokens too quickly? She argues that treating computational tokens as financial instruments could misapply rules and hinder innovation, preferring industry transparency standards before government involvement.