Zhipu and MiniMax shares surge as Anthropic IPO and half-price GPT-6.1 lift Chinese AI

A rally with foreign catalysts

On the morning of October 27, shares of Zhipu — widely described as the first listed pure-play large-model company — climbed as much as 12 percent in Hong Kong trading, while fellow Chinese model maker MiniMax added more than 5 percent. According to a dev.to roundup drawing on Sina Finance, NetEase and other Chinese outlets, the trigger came from abroad: Anthropic's IPO filing and OpenAI's release of GPT-6.1 Terra at half the previous price. Both events signalled to investors that public-market exposure to frontier models is now a real asset class, and served as a reminder that the deepest open-weight ecosystem currently sits in China.

Usage data, not benchmarks, shows the shift

OpenRouter's figures for the week of October 19 to 25, cited in the roundup, make the move measurable. Global model calls reached 188 trillion tokens, up 14.63 percent week over week. Chinese models handled 93.54 trillion of those tokens, an 18.14 percent increase, marking the 26th consecutive week in which Chinese models out-consumed their American counterparts. DeepSeek's Flash 1017 alone processed 33.6 trillion tokens, the highest of any model worldwide — evidence that ecosystem gravity tends to appear in usage statistics long before it appears in benchmark leaderboards.

Zhipu's fundamentals catch up with its ticker

The rally is not purely narrative. Zhipu reported first-half 2026 revenue of 953.9 million yuan, up 399.7 percent year over year, with open-platform and API revenue of 825.2 million yuan — a jump of 2,735.7 percent that now represents 86.5 percent of the total. Full-business annual recurring revenue stands at 1.8 billion dollars, and management raised its year-end ARR guidance from 2.4 billion to 3 billion dollars. Goldman Sachs upgraded the stock to Buy and lifted its own ARR forecast to 3.2 billion dollars. Amazon Bedrock has integrated Zhipu's GLM-5.3 under a revenue-sharing arrangement, and management says similar cloud partnerships at home and abroad will begin contributing recognised revenue from October.

DeepSeek closes a 15 billion dollar round

The same week, DeepSeek closed a funding round of roughly 100 billion yuan — about 15 billion dollars — the largest single AI financing in Chinese history, after twice raising its target from an initial 50 billion yuan. CATL and Tencent led a syndicate heavy with industrial capital, bringing cumulative Chinese AI fundraising this year to roughly 150 billion yuan. The round values DeepSeek near 74 billion dollars against about 1 billion dollars of annualised revenue. CITIC Securities is advising on a STAR Market IPO targeted for early next year which, if the timeline holds, would make DeepSeek the first frontier lab to list anywhere — potentially beating Anthropic to the opening bell.

A US ban answered with open source

The week also produced a pointed juxtaposition. The US Department of Homeland Security issued a federal exclusion order barring not only DeepSeek products but third-party services built on its models. DeepSeek's response was to fully open-source DeepInfer, its production inference engine, with support for both NVIDIA and Huawei Ascend hardware. As a Chinese economic-daily commentary cited by the roundup framed it, an institution can be locked out of an account, but it cannot be locked out of open source.

The open-weight drumbeat continues

The funding news sits atop a relentless stream of open-weight releases. Moonshot promised full open weights for Kimi K3.1 Pro — a 2.8-trillion-parameter mixture-of-experts model with 104 billion active parameters and a million-token context — by November 4, and cut prices by 35 and 32 percent to 13 and 68 yuan per million tokens. MiniMax published the weights of M3.1 Flash, a 428-billion-parameter model with 23 billion active, on Hugging Face and ModelScope. Tencent open-sourced Hy-SimulMT-30B-A3B, a speech-to-speech simultaneous interpretation model with roughly three-second latency and real-time voice cloning, under Apache 2.0. Nanbeige released a 33-billion-parameter agentic coding model, and Hugging Face, working with academic partners, shipped Carbon-X, a 14.8-billion-parameter open DNA foundation model.

Why it matters

Two forces are converging at once. First, frontier AI is becoming an investable public asset: ARR is emerging as the key metric, Goldman is underwriting the story, and a race to list — DeepSeek versus Anthropic — could decide which ecosystem public markets price first. Second, open weights are eroding the leverage of restrictions: a federal exclusion order loses much of its force when the models and now the inference engine itself are freely downloadable and run on both NVIDIA and domestic Chinese chips. For buyers of AI capacity, the practical consequence is falling prices and credible alternatives to closed US APIs. For American labs, the closed-API flywheel that funded their buildout now faces a challenger being rebuilt in public, at commodity prices, in Shanghai, Beijing and Shenzhen.