China's biggest livestream host moved 100,000 packs of AI compute credits in 30 minutes, priced at ¥0.01. It was never about selling — it registered one idea in 100,000 heads: tokens are daily consumables now. If you're a founder who watched that and thought about stockpiling credits for the team, read this first.
A ¥28 ($4) monthly pass went for ¥0.01; total take ≈ $150 — less than one slot on his show. Li Jiaqi wasn't selling compute. He was teaching a mass audience that tokens are grabbable, hoardable, ordinary — the same psychological registration lipstick once needed.
They've long sat on corporate purchasing lists. Unconfirmed leaks put Tencent's employee AI perk at ¥220k ($30k) a head — internal Claude, Cursor, their own CodeBuddy. Ours runs shared pools at ~$280 a head.
In June, Tencent reversed: "credits for everyone" became needs-based, judged "on output, not consumption." The reversal is the answer. Tokens went out; efficiency didn't grow on its own.
The token lands in my account, I finish quietly, saved time is mine. You paid for 8 hours and wonder how few it could be. Goals twist; real gains get hidden. You'll never see the full account.
Zally, change-management lead at Anker, hears it from staff: "My efficiency doubled — but the processes between us are still broken." And: "When the novelty fades, everyone reverts."
Taylor proved it in 1898: Schmidt hauled 12.5 tons of pig iron a day; stopwatches and redesigned motions took it to 47. Nobody touched his muscles. Better gloves? He thanks you — and hauls 12.5 tons forever.
A small squad, one task impossible with current headcount, pressure to invent an AI-native way through. When the old method can't cope, every real gap surfaces — the task becomes your process audit.
Anker runs this at scale, the "Rocket Class": business, IT and AI people per group, each with one real problem, distilling scattered SOPs and veteran judgment into agents that run on the platform. Their metric: willingness and output — "how many tokens were consumed is a secondary factor."
Which process to rebuild is a founder's decision — it doesn't outsource to a tool or a training budget.
1. Workflow mapped into "actions" (rule-based) vs "judgments" (human calls)? Without that line, tokens are money in water.
2. What impossible task will you hand your squad? Can't name one — you haven't decided what AI is for.
3. Save headcount, or build assets? Saved hours are one-off; Anker's agents stay when people leave. Expertise in a head walks out the door.
Tip if you did stockpile: Zhipu and Qwen credits plug into third-party tools — one subscription, several tools. Ours does.
Anyone can afford ¥0.10. The real spend is getting AI into the workflow.
Ask how the squad-and-impossible-task method maps onto your China operations.
Start a conversation →