Artificial Intelligence thread

[PopularScience]

Temu is Headquarter in Boston from a long time but the parent company is headquarter in Shanghai, Mihoyo is headquarter Shanghai and Shein pretty much every important decision is done in Nanjing China.
SMIC is headquarter in Cayman Island, the company that is making the chips of the most sanctioned company in the world.
ACM Research is headquarter in California, the company that make one of the most advanced wafer cleaning equipment in the world. Probably one of the most important semiconductor equipment makers in China.

Manus is basically a grift company with a mediocre product that got some press because the novelty back them. They move to Singapore because their product is powered by US cloud based models. Alien looking Zuckerberg buy them for the "talent", personally they would have go to Tsinghua and they would have get better talent for cheaper.

I heard Shein moving to HK for listing in hkse.

 

[Nevermore]

The situation is somewhat different. Manus had already planned to go overseas when it had very little presence on the mainland.

Shein is still expanding its business in China, and there were previous rumors about relocating its headquarters back to China and listing in Hong Kong. Temu is essentially in the same position, PDD's headquarters also based in Guangzhou.

Mihoyo actually resulted from a founder split. The Singapore studio is essentially a separate operation started by Cai Haoyu with his own shares, and it has little connection to the existing business and has a very small staff.

Mihoyo maintains an R&D base in Singapore and continues to expand its operations there. The development headquarters for Mihoyo's next-generation game Varsapura is based in Singapore, though project development is divided between teams in Shanghai and Singapore.

 

[GulfLander]

Actually good acquisition by Meta

Manus was a quickly growing startup and had good R&D and working product on agentic workflows but still had the common issue that startups have, user acquisition

Meta has a huge established user base, but is behind AI and agents.

Meta gains Manus' very valuable data, r&d in agents and ai integration, and established user base of experienced developers/users to test and quickly iterate it's AI models.
And Manu will gain access to the huge user base and vast resources of Meta.

Imo perfect acquisition. Alexander Wang doing good work

"Very valuable data".. does it include data obtained in China ?

 

[tokenanalyst]

"Very valuable data".. does it include data obtained in China ?

They don´t offer services in China.

 

[GulfLander]

They don´t offer services in China.

I mean those "collected" if there were before relocating?

 

[CMP]

I mean those "collected" if there were before relocating?

They left precisely because they had almost no market penetration, and hence little to no data collected.

 

[bsdnf]

Moonshot has completed a $500 million Series C funding round, valuing the company at $4.3 billion post-money. Chinese AI companies are receiving more investment or going public.

 

[meedicx]

It has been a very eventful December for Chinese AI releases, probably the busiest ever. Chinese AI labs are gaining momentum and accelerating their releases.

From my tracking, here is the releases I found notable

Deepseek	Deepseek-3.2 Deepseek-3.2-Speciale	Speciale is especially notable for being SOTA on reasoning onpar with the best closed models
Alibaba	Qwen3-TTS Wan-2.6 Qwen-Image-Layered Qwen-Image-2512	SOTA text-to-speech models. New Video generation model with native audio gen. Improved image generation and interesting image segmentation model
Bytedance	Seed1.8 VLM Seedream 4.5 Image Gen Seedance 1.5 Pro Video Gen	Released the best Chinese multimodal LLM. Improved image generation model. New Video generation model with native audio
Tencent	HY 2.0 HY World 1.5 Hunyuan OCR	Rebrand Hunyuan LLM to HY. Interesting world model (WASD game engine using AI to generate frames). SOTA OCR model w/ only 1B parameters
Kuaishou	Kling Video 2.6 Kling-O1	New Video generation model with native audio and support for motion control with reference video. Experimental omni model for video generation
Xiaomi	Xiaomi MiMo-V2-Flash	First release for Xiaomi MiMo lab; very efficient and fast MOE with strong agent abilities (to support future AI phone)
Z.AI	GLM-4.6V GLM-4.7	Released current best open weight model on AA benchmarks. Strong coding model with rave reviews in the coding community.
Minimax	Minimax-M2.1	Very fast follow to M2 (released only one month ago) with big improvements. Strong coding model with rave reviews in the coding community.
StepFun	Step-GUI	SOTA smartphone GUI Agent that can run locally on device

In this month, the major Chinese video generation models (Wan-2.6, Seedream-1.5, Kling-2.6) have been updated to support native audio generation catching up with Veo. Deepseek released the finalized version of 3.2 showing the strength of its new Deepseek Sparse Attention algorithm.

Z.AI and Minimax both released strong coding models and used the momentum to file for IPO on HKEX in 2026 Jan. Z.AI, Minimax, Stepfun were in major trouble earlier this year when Deepseek R1 stole their thunder and made their future direction uncertain. By the end of year, these 3 all have successfully pivoted and refocused on vertical niches - Z.AI / Minimax on AI coding and StepFun on smartphone Agentic AI.

On the hardware side

• Bytedance partnered with ZTE to release the Doubao AI Smartphone, which made a huge stir and could potentially be the next revolutionary consumer electronic product.
• Moore Threads announced it's new Lushan gaming GPU and Huashan AI GPU slated for release next year
• Lisuan started shipping it's Lisuan G100 gaming GPU which is on par with nvidia GTX 4060
• Science published an article on the LightGen Photonic Chip, which is a potential revolutionary technology to obsolete the current electron-based AI chips computing paradigm