The token economy: a structural leap in computing demand
AI computing · 2026-07
The commercialization of generative AI has accelerated sharply over the past two years. Token volume has become a key barometer of industry momentum — from chat assistants to coding tools, from AI agents to multimodal applications, each maturing use case drives a step change in consumption. The industry broadly expects global token demand to keep growing at multiples per year.
The structure of demand is shifting as well: inference now accounts for a rising share of total computing, while long context windows and multi-turn interactions significantly increase the compute per task. Notably, Asia-Pacific has become the fastest-growing region in the global computing landscape.
For infrastructure investors, this means demand is no longer a single 'training arms race' but a shift toward continuous, distributed, application-adjacent inference supply. Those who can deliver stable inference capacity in the right locations at the right energy cost will hold the advantage in the token economy.
Global AIDC supply and demand: opportunity in a tight balance
Data centers · 2026-06
Major technology companies continue to expand capital expenditure on AI infrastructure, driving a historic build-out cycle for AI data centers. Yet industry observation suggests that compliant land, power interconnection and long equipment lead times prevent supply from fully catching up with demand, leaving most core markets in a tight balance.
Against this backdrop, projects with secured power capacity have become scarce assets: obtaining and expanding power interconnection is often a higher barrier than construction itself. Meanwhile, high-density racks and liquid cooling create a generational gap between new AIDCs and legacy data centers.
We believe AI infrastructure will remain a high-conviction investment theme for years to come. Seizing the window, locking in energy and location resources, and realizing asset value through professional operations are the keys to this cycle.
Where green power meets computing: the path to sustainable AI
Green energy · 2026-05
The exponential growth of computing has made energy the binding constraint of the AI industry. Electricity typically represents a substantial share of data center operating costs, while regulatory requirements on carbon emissions keep tightening across major markets — green computing is shifting from a bonus to a prerequisite.
The cost of solar generation and battery energy storage has fallen dramatically over the past decade, making 'direct green supply plus storage balancing' economically viable in more and more markets. Co-planning renewable plants and AI data centers locks in long-term energy costs and materially improves carbon performance.
Looking ahead, integrated development of energy and computing will become the industry's mainstream paradigm. Investors and operators who control energy resources and possess cross-domain integration capability will gain a first-mover advantage in sustainable AI.
From bare metal to GPU cloud: the evolution of computing services
Computing services · 2026-04
Early GPU supply was dominated by machine sales and long-term leases, leaving customers to handle deployment and operations complexity themselves. As large-model applications proliferate, the market increasingly demands elasticity, usability and high availability — pushing the industry toward GPU-as-a-Service.
A multi-layer service system is taking shape: bare-metal leasing serves leading customers who demand maximum performance and control; hosting lowers facility and operations barriers; and cloud-based platforms serve a much broader base of developers and enterprises with flexible billing and rapid delivery.
In this evolution, simply 'having GPUs' is no longer a moat. Comprehensive operating capability — cluster stability, network performance, workload scheduling and service responsiveness — is the durable source of value for computing service providers.