Artificial Intelligence

Big Tech's AI Investments

Major tech companies are heavily investing in AI large language models, viewing them as potential solutions for growth. These companies possess advantages in funding and talent compared to startups. Different strategies are emerging, with Huawei focusing on "industry-specific" applications and Baidu initially prioritizing general-purpose AI models. Core business alignment is evident in the chosen applications. Market strategies vary, with ByteDance, Kuaishou, and iFlytek opting for closed-source models, while Baidu, Tencent, and Alibaba are embracing open-source approaches. Top-tier models from major players are primarily closed-source, making overall capability comparisons challenging, but Baidu's Ernie Bot, Tencent's Hunyuan, and Alibaba's Tongyi Qianwen are considered leading models.

Here's the news brief based on the provided information:

AI Infrastructure

DeepSeek API Access Paused by LuChen Tech

LuChen Tech, a domestic AI Infra company, has suspended DeepSeek API access services. Founder You Yang cited cost considerations as the primary reason. He argues that DeepSeek's published system theoretical cost profit margin data is derived under overloaded resource conditions, making it difficult for conventional MaaS products to achieve. You Yang also noted frequent disconnections with DeepSeek products and suggested that the cost of developing large models may be significantly higher than figures presented in their papers. Several industry insiders have indicated that the MaaS industry is generally operating at a loss, and LuChen Tech's decision to suspend DeepSeek services is due to the high cost of providing stable access, making it challenging for small and medium-sized MaaS providers to compete with large cloud vendors. Hundreds of companies have announced access to DeepSeek, requiring MaaS service providers to reserve substantial redundant resources to handle peak user traffic, creating significant cost pressures.

AI Legislation

California's SB53 for AI Safety

California State Senator Scott Wiener has introduced AI bill SB53, designed to protect employees at leading AI labs who raise concerns about AI systems posing a “critical risk” to society. The bill also proposes creating a public cloud computing cluster, CalCompute, to provide researchers and startups with the necessary computing resources for developing AI for public benefit. SB53 aims to prevent retaliation against employees of frontier AI model developers who disclose risks potentially leading to death or serious injury to over 100 people, or damages exceeding $1 billion.

AI Model

DeepSeek-R1 Model's Popularity

The DeepSeek-R1 model rapidly gained popularity after its release, becoming a phenomenon. Its rise to prominence over four key days is analyzed: from its launch on January 20, which garnered attention in tech circles, to January 27, when comments from Black Myth developer Feng Ji triggered widespread discussion. Technical innovation, national sentiment, active discussions, and an open ecosystem collectively propelled DeepSeek's spread. Zhihu established its technical credibility, while Weibo became the main force behind its broader recognition. This event has become a significant collective memory in the era of the intelligence revolution and offers new insights for the AI industry.

Artificial Intelligence

AI Integrated Machine Market Outbreak

The demand for domestic computing power is strong, and the AI integrated machine market is booming. Many manufacturers such as Huawei Ascend and Sugon have increased the deployment of integrated machines. iFlytek has also jointly upgraded the Spark integrated machine with Huawei and launched products for specific scenarios. The open source and model capability upgrade of DeepSeek is the main driving force. Its integrated machine can be equipped with domestic AI chips to meet the needs of governments and enterprises for data security and independent control. However, the cost of millions and users' willingness to pay are challenges for market promotion. It is estimated that the demand for integrated machines will increase significantly from 2025 to 2027.