A Chinese state fund ran 6.07 billion tokens through its deal desk

Share
A Chinese state fund ran 6.07 billion tokens through its deal desk

Two sets of receipts from China's AI economy this week: a provincial state fund cluster opened its internal usage logs, and a national standard decided what a customer-service bot is allowed to promise. Enterprises everywhere talk about agent pilots; these are numbers from production.

Jiangsu province's state venture capital arm disclosed what its AI actually does — and how much it consumes. At a September 10 provincial SASAC exchange held at Jiangsu High-Tech Investment Group, the manager of the province's Strategic Emerging Industries fund cluster said the group's investment staff have made 124,054 AI service calls this year, burning roughly 6.072 billion tokens on locally deployed models. The workloads are concrete: an AI-assisted due-diligence report reviewer has processed 139 projects under its official build and 37 under a staff build; 347 investment agreements and 86 sets of meeting minutes have been pushed through model checks at the deal stage; and 91 funds now get weekly and quarterly report review with AI assistance. The operating rule, stated plainly: AI aggregates information, screens documents and flags anomalies — humans make the investment decision.

What's underneath the usage is the other half of the story. Gaotou said it built its own GPU pool with terabyte-class memory, deployed Zhipu's GLM and Alibaba's Tongyi Qianwen locally, and wired the models into business systems through an interface layer kept deliberately separable from the applications, so no single model version can lock in the workflow. The group says it catalogued 145 internal AI needs with the state think tank CAICT and narrowed them to seven live scenarios. On the capital side, the cluster has 301.2 billion yuan of established fund scale, of which 36.647 billion yuan is aimed squarely at AI — seven AI industry funds (21.8 billion yuan) plus 15 AI-focused sub-funds (14.847 billion yuan); the system has backed more than 110 companies across the broad AI stack and took part in a private placement at chip-packager Tongfu Microelectronics. Two further 1 billion yuan AI-themed funds are being set up. This is the deal-side companion to the fund-level bets we covered in August — China's state AI fund just took a direct stake in Kling — and it follows the deployment quotas in China's MIIT sets hard targets for an AI service-provider army. The read: a state fund answers to a ministry, not to LPs, which is why it published token counts no US venture firm would volunteer. Those numbers are now a benchmark other provincial funds will be quietly measured against.


China's first national standard for AI customer service took effect on September 1, and its core rule is about authority, not capability. GB/T 47746-2026, "Customer Contact Service — Requirements for the Coordination of Human and Intelligent Customer Service," allows a bot to answer routine questions and explain rules, but says that wherever price, discounts, refunds, compensation or contract changes are involved, the final confirmation must come from a human operator or a fixed, auditable process. It also requires a visible, reachable "talk to a person" entry and demands that context survives the switch, so customers stop re-explaining themselves to each new agent.

The backdrop is complaint data: the China Consumers Association flagged AI customer service in its first-half 2026 analysis for exactly the two failure modes the standard names — hollow promises the company later disowns, and engineered difficulty reaching a human. The standard is recommendatory, so nothing about it is directly enforceable; but it hands regulators and courts a written yardstick, which is the same soft-law playbook behind China's top court sets the rules of the road for AI deepfake lawsuits.

What to watch: whether any platform dispute cites GB/T 47746 as its test, and whether other provincial state funds copy Gaotou's disclosure — the token count is the part that will get compared.

If a bot is trusted to negotiate but not to promise, is that a sane human-AI division of labor or just liability theater? Write to us via our contact page.

Get SiliconNoon in your inbox

One email a day with the AI stories that matter — what happened, the numbers, and why it matters. No spam, unsubscribe in one click.

Read more

DeepSeek plans a 160,000-chip Huawei cluster in Inner Mongolia

DeepSeek plans a 160,000-chip Huawei cluster in Inner Mongolia

Monday's stories are about what happens after the models ship: a Chinese lab committing to domestic silicon at a scale nobody has attempted, a record graduating class walking into a hiring market AI is quietly reshaping, and a retailer giving away its AI stack to move more hardware. DeepSeek plans to install at least 160,000 Huawei Ascend 950DT accelerators at a data center it is building in Inner Mongolia, according to people familiar with the matter — a deployment that would rank among the…

Alibaba plans a 10T-parameter model and a 500,000-chip cluster

Alibaba plans a 10T-parameter model and a 500,000-chip cluster

Alibaba's cloud conference in Hangzhou opened this morning with a three-part commitment that is harder to wave off than the usual keynote arithmetic: a model bigger than anything trained so far, the chip to run it, and the power to feed both. Huawei spent the same week putting a coding model inside its own operating system, and a small project on the local-AI forums quietly showed that a model's refusals can be switched off without touching a single weight. Alibaba says its next Qwen models…

Musk: Grok 4.7 trails Opus 5.5, expects to catch up next year

Musk: Grok 4.7 trails Opus 5.5, expects to catch up next year

The most candid AI scorecard of the weekend came from the head of one of the labs in it — and he gave it to Chinese state television. Elon Musk told China Media Group that SpaceXAI's flagship model, Grok 4.7, is "not as good as, say, Opus 5.5 that just got released," and put a timeline on closing the gap: "I would expect that we probably catch up to Frontier sometime next year, most likely." The interview was conducted on 23 September at Tesla's Global Engineering Headquarters and published by…