In 2026, a microscopic yet profoundly significant term defined global wealth dynamics: Token. At NVIDIA's GTC conference that year, Jensen Huang proposed shifting the technology industry's core metrics from FLOPS and TOPS to Token, a unit better suited to measuring AI productivity in the current era. NVIDIA positioned itself as the manufacturer responsible for Token production and efficiency definition.
Half a month later, on April 1, Lenovo Group Chairman and CEO Yang Yuanging addressed another critical question at Lenovo's new fiscal year kickoff meeting: where, how, and by whom would the Tokens defined by Huang be consumed efficiently, securely, and reliably. The previous day, Lenovo had launched two "Agent PCs" designed specifically for silicon-based employees—screenless computers intended as dedicated workspaces for AI assistants. Yang described 2026 as Lenovo's "AI Delivery Year," with plans to develop wearable devices, new-form-factor PCs, tablets, smartphones, and personal computing hubs for "personal intelligence," alongside deployable enterprise agents for workflow transformation.
Huang focused on "how Tokens are produced," while Yang defined "where Tokens are consumed." NVIDIA drove a supply-side revolution, whereas Lenovo bet on demand-side access points. If NVIDIA aspired to be the world's largest Token factory, Lenovo aimed to become the most extensive Token consumption terminal provider.
Huang personally acknowledged this symbiotic relationship at GTC 2026, telling Yang, "This year belongs to you. We did it." NVIDIA's evolution from a graphics card company to an acceleration computing firm now extended to becoming an "AI factory," offering integrated systems from CPUs and GPUs to networks, storage, and full-stack software optimized for Token output. Huang declared the end of the chip company era, emphasizing that factory efficiency would be measured by Tokens per watt.
Token efficiency is becoming a critical metric for corporate scale and even national GDP. Huang's "Token economics" reflected a structural shift from AI training to inference, where each query, task decomposition, and agent execution continuously consumes Tokens. The rise of agents like OpenClaw—hailed by Huang as the "next Linux"—amplified Token demand through autonomous task execution. Data from China’s National Data Bureau showed daily Token usage exceeding 140 trillion by March 2026, with OpenClaw alone contributing 20% of weekly platform consumption.
Huang's corporate growth formula for the AI era tied revenue to Token throughput per watt, available power, and Token pricing tiers. However, his narrative remained production-centric, leaving consumption scenarios undefined.
Lenovo filled this gap. Yang divided AI development into three phases: computing power engines, model engines, and the current data-engine-driven era, particularly reliant on private data. He emphasized that AI must leverage personal, business, and device data—often processed locally for privacy and efficiency. Lenovo's Agent PCs, like YOGA AI Mini and Think AI Tiny, provided dedicated terminals for silicon-based employees, mitigating data leakage risks and resource contention. These devices featured one-click deployment of DingClaw agents via Lenovo's DingOS, voice-controlled task execution, and a Skill factory with 8,000+ tools by year-end, safeguarded by a security mechanism.
As agents proliferate, "one AI PC per human" could create a 50-million-unit market worth $25–50 billion. Each agent PC would consume Tokens predictably through local or cloud inference, forming manageable, auditable nodes. Lenovo's strategy extended beyond PCs to a "computing power matrix" of wearables, tablets, phones, and sensors tailored to personalized data scenarios for individuals and enterprises.
Lenovo CTO Tolga Kurtoglu stressed transitioning AI from LLM-driven to privately data-driven systems, ensuring reliability and security. Executive Vice President Luca Rossi projected Lenovo and Motorola would operate one of the world's largest personal AI platforms across billions of devices. This scale, combined with enterprise servers and edge devices, would magnify Token consumption exponentially.
Lenovo's unique strength lies in bridging B2B and B2C scenarios. In China alone, it developed 46 domain-specific agents in the past year. Its agent platform, Qira (Tianxi in China), enabled cross-model, cross-device AI integration, driving record PC market shares of 25% globally and 36% in China, with Motorola phones reaching top-four global rankings.
Kurtoglu outlined three focus areas: semantic data transformation for contextual knowledge, accurate and secure AI responses, and hybrid AI orchestration for seamless compute/model switching. Lenovo's agent system aims to close the gap between AI capabilities and practical application—ensuring Tokens deliver value to form consumption loops.
Without Lenovo, NVIDIA's Tokens would remain unrealized potential; with it, Tokens enter a cycle of high-quality consumption. Yang and Huang's partnership binds Token production to real-world usage across households and industries.
A critical layer remains: the foundation models and agent systems that determine Token consumption efficiency. While model capabilities set the baseline for silicon-based employees, Token scheduling—deciding which model to use, where to compute, and how to access data—controls cost and profitability. Major AI firms are pivoting from model prowess to agent systems, but users need a neutral调度层 (scheduling layer) to optimize choices.
Lenovo's hybrid AI architecture, developed over three years, seeks this调度权 (scheduling authority). Its "control center" vision simplifies complexity for users, automating optimal paths for model, compute, and data selection. Lenovo targets not just hardware margins but authority over Token scheduling, execution, and experience definition.
Model power sets the starting point, but Token scheduling determines whether AI integrates sustainably into workflows. NVIDIA leads Token production; model vendors prove agent viability; yet, the rarest player is one that orchestrates Tokens across multi-terminal, multi-scenario deployments. Lenovo may not have fully closed the loop, but it is among the most committed to this path.
Comments