On 13 June 2025 the Hangzhou municipal government issued its implementation plan for building the city into an AI innovation highland in 2025. The plan sets out a set of measurable targets: intelligent computing scale, the number of foundation models and the number of vertical application models.
Defining a highland by metrics, not slogans
The plan aims to bring more than 50 EFLOPS of intelligent computing to market by the end of 2025, and to cultivate two foundation models and at least twenty-five vertical application models. The value of quantified targets is that they give both policy support and assessment a common reference point.
The gap between “two foundation models” and “at least twenty-five vertical application models” also reveals a positioning judgement: Hangzhou is not trying to compete on the number of general-purpose foundations, but is concentrating on the vertical application layer.
How it connects to Hangzhou’s legislative rhythm
Before the plan was issued, Hangzhou had already completed dedicated legislation in the embodied AI robot field. Legislation supplies certainty and the plan supplies resources; together they are two sides of local industrial policy.
Hangzhou’s courts have also been active in shaping the rules for commercial AI applications. See our notes on Hangzhou’s first unfair competition case involving commercial AI use and on the Hangzhou AI hallucination tort case.
The compliance questions behind compute targets
Expanding intelligent computing brings questions of data concentration and energy consumption, and it widens the exposure to data compliance. When a company connects to public compute resources, it needs to check data processing and outbound transfer rules at the same time. See our notes on the national integrated compute system and large models and on the action plan for building high-quality industry datasets.
Where companies can start
First, translate local targets into an application path, deciding whether the company is a candidate for foundation-model cultivation or vertical application-model cultivation.
Second, assess the data compliance cost of connecting to compute, factoring data processing rules into the compute procurement decision.
Third, watch how local and central policies stack. Local plans have to align with national deployment. See our note on Shanghai’s AI application subsidy.
Competition between local AI policies ultimately comes down to whether targets can be turned into resources companies can use and rules they can predict.
This article is based on the Hangzhou municipal government’s Implementation Plan for Accelerating the Building of an AI Innovation Highland (2025 Edition) and public information and is for reference only; it does not constitute legal advice.