Computing Power and Tokens Emerge as Quasi‑Assets, Opening New Frontiers for AI‑oriented Finance

According to China Economic Net, rapid advances in artificial intelligence have brought computing power and tokens distinctive quasi‑asset characteristics across the tech sector. The Ministry of Industry and Information Technology has recently released policy guidance that for the first time explores equity investment backed by computing‑power resources. A suite of innovative financial products including “computing‑power loans”, “model‑based loans” and “token‑linked loans” have been rolled out in Beijing, Shanghai, Guangdong and other locations. Market participants are starting to take computing power and token performance as fresh yardsticks when assessing the operational quality of AI‑focused businesses.

Variations in operational arrangements can generate several‑fold gaps in pay‑back cycles even for identical levels of computing capacity. Equally, token quality fluctuates subject to model capability and day‑to‑day operational management. For computing power and tokens to evolve from quasi‑assets into widely‑accepted hard‑currency‑style resources, sustained technological progress must lift their intrinsic value. Businesses also need to establish dedicated financial management routines for computing resources and token consumption, feeding operational insights back to boost overall energy‑efficiency ratios.

Policy initiatives at multiple administrative levels are unlocking the latent asset value embedded within computing resources and tokens. Under the *Action Plan for Supporting Small and Medium‑Sized AI Enterprises (2026‑2028)*, the central ministry explicitly advocates new supportive mechanisms such as equity investment via computing‑power contributions, data‑backed equity and integrated investment‑incubation frameworks. This marks the very first mention of computing‑power‑based equity investment within an official MIIT document.

Local authorities have been examining such innovative investment models ahead of the national‑level release. Back in March 2025, the Shanghai municipal government published documents to study equity investment supported by computing assets and test hybrid investment structures combining fund injections with computing‑power equity stakes for intelligent computing hubs. Further elaboration appeared in Shanghai’s 14th Five‑Year‑Period software and information services industrial plan issued in August 2026, confirming ongoing trials for asset appraisal and equity contribution using computing resources.

Regional policymakers are also recognising tokens as a new category of digital asset. At the end of July, Anhui’s policy framework for AI‑driven individual‑founder entrepreneurship encourages commercial banks to launch token‑linked credit facilities and model‑based loan offerings. Inner Mongolia subsequently issued targeted measures to foster the token‑driven economy in August, urging financial institutions to expand credit provision for relevant projects on market‑oriented and legal foundations while permitting technology firms to explore feasible pathways towards formal token asset recognition.

Domestic cases of completed computing‑power equity deals remain limited for the moment. Major Chinese AI corporations including Alibaba, ByteDance and SenseTime have run internal incubation programmes, offering computational resource support and token‑related subsidies to start‑ups. In May, SenseTime opened its new industrial base in Shanghai, leveraging its large‑language models and intelligent digital‑creation agents to nurture an innovation community for short AI‑generated drama content.

Overseas technology players are also experimenting with comparable investment structures. In May this year, OpenAI announced arrangements for start‑ups admitted to Y Combinator’s programme, granting token credits worth up to 2 million US dollars in exchange for equity commitments. Industry commentators regard this framework as an emerging paradigm where tokens themselves function as capital.

In the domestic banking sector, token‑linked and computing‑power‑focused lending products continue to expand. The Shanghai branch of Bank of China has unveiled its dedicated computing‑power loan and token‑linked loan packages. The computing‑power‑oriented facility serves operators of computing hubs and larger industrial‑chain enterprises, financing data‑centre construction, intelligent computing centre operations and bulk procurement of cloud‑based computational services. The token‑anchored lending product is tailored for small‑scale innovative AI enterprises; it draws on industrial operational datasets to strengthen corporate credit standing, with token metrics forming a core valuation reference. Agricultural Bank of China, China Construction Bank and Bank of Jiangsu have already trialled equivalent loan schemes across selected cities, bringing tangible financial benefits to leading domestic large‑model developers.

Marked value divergence has become visible amid the rising status of these digital quasi‑assets. Disparities in underlying technology and operational proficiency mean that the economic returns yielded by identical computational hardware or token volumes can differ substantially. Research from Xinbao Investment Research has demonstrated that one standard intelligent computing server may deliver static pay‑back cycles varying by more than eight‑fold when running different flagship open‑source large models. The research took official pricing schedules, server utilisation rates, infrastructure outlay, operational maintenance, power expenditure and taxation into consideration during its modelling.

Analysts point out that revenue generated by computing hardware results from a combination of processing throughput and unit service pricing. Models boasting extensive parameter sets tend to run comparatively slowly while commanding higher unit prices; faster‑running alternatives frequently face intense downward pressure on service charges. Well‑balanced large‑model implementations therefore represent the most suitable workload choices for intelligent‑computing hardware.

Senior industry figures from Gartner highlight that surging market appetite for AI computing resources does not automatically guarantee high returns on every capital outlay. Discrepancies in overall commercial outcomes arise principally from infrastructure utilisation, effective computational output and the comprehensive performance of the complete software‑hardware stack. Significant gaps in value density also exist between different batches of generated tokens. Within forum discussions held at the Bund Summit, academics noted that tokens operate as a fresh unit of exchange within artificial‑intelligence workflows, yet individual token outputs are not inherently equivalent. 

Large‑scale token production facilities cannot automatically guarantee high‑quality token generation. Variations in underlying models and practical business tasks create differing levels of embedded intelligence within token outputs, which means valuation frameworks for tokens need supplementary quality weighting coefficients alongside raw quantitative measurements. Market participants are now adopting more rational attitudes, paying closer attention to measurable economic gains generated from token expenditure.

Industry practitioners keep exploring practical routes for these digital resources to evolve from quasi‑assets into broadly accepted hard‑currency instruments. The competitive landscape for artificial intelligence is shifting away from pure capacity expansion towards efficiency‑driven and subsequently low‑carbon‑oriented competition. Higher intelligence density for each token unit together with reduced carbon footprints will facilitate deeper adoption of AI solutions across traditional industry sectors. Key performance benchmarks used to assess computing‑power investment returns will gradually move beyond simple GPU hourly rental rates. 

Evaluation frameworks will focus increasingly on how much valid AI computation each unit of capital investment can sustainably deliver. Businesses ought to shift their strategy from straightforward procurement of basic computing resources towards holistic economic optimisation of the complete AI infrastructure stack, covering resource orchestration, hardware utilisation, heterogeneous computing and high‑speed inter‑server networking. Internal financial workflows specifically designed for large‑model operations and computing‑resource expenditure also need to be put in place across AI enterprises in the period ahead.