Chinese AI Firm DeepSeek Pursues Investment Amid Next-Gen Model Delays

Chinese AI startup DeepSeek is reportedly taking steps to secure external investment for the first time while delaying the launch of its next-generation model. DeepSeek garnered industry attention 15 months ago with its R1 inference model, but questions are mounting over the investment drive as the release of its new V4 model faces multiple postponements.

A potential shift to Huawei chips is being cited as a reason behind the delay of DeepSeek's next-generation V4 model and its new investment push. While DeepSeek used Nvidia's H800 GPUs for its previous V3 model training, reports suggest the V4 model will be optimized for Huawei's latest AI chip, the 'Ascend 950PR.'

In addition to the model delays, DeepSeek is also facing talent attrition, with Guo Daya (郭大雅), the lead researcher behind the DeepSeek R1 model, reportedly moving to ByteDance. Although DeepSeek has received ample funding from its parent company Liangwenpeng's hedge fund HighFlyer, this investment is seen as an effort to retain departing key personnel and establish stock option valuations.

DeepSeek's next-generation model delays and its pursuit of investment are heightening industry interest in the company's future direction amidst the rapid-fire competition in AI technology development. Nvidia CEO Jensen Huang recently commented that a day when DeepSeek optimizes first for Huawei would be 'terrible' for the US, expressing concern over AI models developed globally performing best on hardware other than US-made products.