Enterprises want AI but often don't know which model to pick, where to find compute, or how to put it to work. On October 22, the Sichuan AI Supermarket launch conference was held at Chengdu Science & Innovation Eco-Island, hosted by the Sichuan Provincial AI Office, with Lu Songming, director of the AI Office and head of the Department of Science and Technology, in attendance. A provincial platform that puts models, compute and agents on the “shelf” is now open.

Digital platform building with data shelves and cart icons, tech illustration
A provincial marketplace standardizes models, compute and agents for sale

1. Five elements on the shelf, matched in 3 seconds

The platform spans five service systems—compute, tokens, data, agents and terminals—targeting the three pain points SMEs face in adopting AI. For “how to use”: ① over 170 mainstream large models and 40+ vertical industry models; ② 60+ industry agents and 200+ industry services covering manufacturing, healthcare, government and energy; ③ online “smart shopping guide” matches solutions in 3 seconds, backed by 500+ engineers offline for end-to-end delivery.

Platform at a glance

· 170+ mainstream LLMs, 40+ vertical models

· 60+ industry agents, 200+ industry services

· 118.8 EFLOPS heterogeneous compute connected (70.8 EFLOPS in-house)

· 200+ suppliers in the first batch

2. The compute ledger: latency down 20%, cost down 40%

For “can't afford it”, the platform offers a clear cost-down path: ① it connects 118.8 EFLOPS of heterogeneous compute on the national compute network, including 70.8 EFLOPS of its own; ② 16 scheduling strategies dispatch jobs to green-power regions or off-peak pools, cutting call latency by 20% and total cost by 40%; ③ traditional monthly billing is replaced with per-card and per-hour flexible billing, lowering the entry barrier for SMEs.

3. A trusted-token zone answers “dare not use”

Data security is the deepest concern of government and manufacturing users. The platform builds a trusted Token zone: ① physical isolation, network whitelists and a self-developed stack keep core data on-premise end to end; ② graded security protection and dedicated delivery teams for government and other critical scenarios; ③ the first batch of 200+ suppliers includes China Telecom AI Technology, Tianfu Jiangxi Lab, Huakun Zhenyu and Xiaoduo Tech, with a second batch applying.

Enterprises shift from “hunting for vendors” to “browsing shelves one-stop”—AI capabilities can be combined on demand and tried before purchase.

4. A provincial template for inclusive AI

By standardizing scattered models, compute, data and delivery services onto one “shelf”, Sichuan offers a provincial template for inclusive AI: supply aggregated into a market, demand assembled on demand, safety backstopped by a dedicated zone. For government and enterprise buyers, this is the first public market where AI can be comparison-shopped. NineZenith takes the same “scenario-first” approach in industry LLM delivery—Zenith-Act supports private deployment and trial-before-buy, echoing the philosophy of such element marketplaces.

(Sources: Sohu, Chuanguan News, China Science & Technology Net and other public reports)