Summary of Key Developments in the AI Industry
Recently, the AI industry has seen a surge in activity: companies are strengthening their competitiveness through large-scale recruitment and technological breakthroughs; industry leaders have divergent opinions on the direction of AI infrastructure development; the demand for AI has driven up prices of supply chain components; capital markets have shown mixed reactions to AI companies; and major players are accelerating the rollout of AI products and investing in 3D generation technologies. These developments reflect the rapid progress and strategic adjustments in the AI sector across various aspects, including technology, market, and capital.
I. AI Companies: Expanding Scale While Enhancing Efficiency Through Technological Advances
- DeepSeek's Systematic Expansion
DeepSeek is not just hiring core talents individually but is recruiting for 33 positions across seven major categories, including algorithms, products, and operations, in both Beijing and Hangzhou. Interns are also welcome, with the goal of doubling the size of all departments. New employees can directly participate in key projects, such as the recently released DSpark inference acceleration framework. This framework addresses the speed bottleneck associated with large models during high-concurrency scenarios, increasing the generation speed on the user side by 60%-85%. The code has been made open-source, which not only improves their own service efficiency but also contributes to industry progress.
- OpenAI's Self-developed Chips Reduce Dependence
OpenAI has collaborated with Broadcom to create a custom inference chip called “Mexican Chili” for large language models. Previously, OpenAI relied on cloud providers for computing power. Now, by developing its own chips, they have achieved better performance per watt than the most advanced chips available on the market. The chip went from design to production in just 9 months, marking a crucial step in building their AI infrastructure, which will help reduce costs and gain control over core technologies.
II. AI Infrastructure: Long-term Optimism vs. Short-term Controversy Among Industry Leaders
- NVIDIA's Sergey Brin: A Decade-long AI Infrastructure Cycle
NVIDIA CEO Sergey Brin believes that the “era of useful AI” has begun, and the construction of AI infrastructure will take several decades, involving critical facilities like power grids and the internet. His concept of a “token economy” is easy to understand: tokens represent units of content generated by AI (e.g., words in a sentence). The more content is generated, the more revenue companies earn. NVIDIA can offer tokens at the lowest cost and with the highest throughput, making them attractive to businesses looking for computing power.
- Masayoshi Son: Limited Relevance of Space Data Centers
SoftBank founder Masayoshi Son has questioned the usefulness of space data centers. While they may save on electricity costs, these costs account for only a small portion of total operational expenses, and the savings are not enough to cover the costs of launch, maintenance, and communication delays. He believes that the current AI competition is more important than efforts in space.
III. Supply Chain: AI Demand Driving Price Increases
- Micron's Tight Storage Supply Until 2027
Micron reported a 345% year-on-year increase in revenue and a 14-fold increase in net profit for its latest quarter, but executives noted that the shortage of storage chips will continue until 2027. The reason is the surge in demand for AI servers, coupled with long production cycles and a lack of skilled workers, resulting in a supply that cannot keep up with demand.
- MLCC Prices Soaring Higher than Gold
MLCCs (multilayer ceramic capacitors) have seen dramatic price increases, with one product from Murata rising from less than 0.2 yuan to 0.65 yuan. The main factor is the much higher usage of AI servers compared to traditional servers, along with speculation by manufacturers, leading to more volatile prices than in the stock market.
IV. Capital Movements: Some Companies Hesitate to Go Public, Others Invest Heavily
- OpenAI Postpones IPO to Next Year
OpenAI originally planned to go public in the second half of this year but has decided to delay it until next year. They are concerned about the potential drop in stock price after SpaceX’s initial public offering and the market's growing skepticism about high valuations for AI companies. CEO Sam Altman still aims for a valuation of $1 trillion (up from the previous $73 billion) and prefers to wait.
- MiniMax Offers Equity to All Employees, Yingmou Technology Raises Millions
MiniMax is giving all its employees equity at no cost, worth nearly HK$600 million in market value. The goal is to retain and attract talent. Yingmou Technology, a company focused on 3D generation models, has recently raised millions in funding, with overseas revenue accounting for 80% of its total income. Customers include ByteDance and Unity, indicating the strong interest in 3D generation technologies from investors.
V. Major Players' Initiatives: AI Integration into Everyday Products and 3D Generation as a New Trend
- WeChat's AI Assistant “Xiaowei” in Trial Use
WeChat’s AI assistant “Xiaowei” is undergoing limited trials, allowing users to place orders, hail taxis, write copy, and edit images. Although it cannot send messages yet, its fast response times and smooth operation mark a significant advancement for WeChat, which previously only had sporadic AI features such as search and summary tools in comment sections.
- Major Players Investing in 3D Generation
Companies like ByteDance, Tencent, and Alibaba are investing heavily in 3D generation technologies. ByteDance released Seed3D 2.0, Tencent open-sourced its Hunyuan 3D model, and Alibaba and Gaode have developed 3D city models. Analysts predict that the global 3D generation AI market will reach $7.59 billion by 2032, indicating significant potential for this field.
These developments indicate that the AI industry is moving rapidly from research and development to practical applications, with supply chains and capital markets adapting to this rapid growth. The next few years will be critical for AI competition.