The recent US ban on exporting high-performance graphic processing units (GPUs) to China has sent ripples through the global tech community. This move, part of the escalating tech war between the two nations, has had varying impacts on Chinese tech companies, from industry giants to burgeoning startups.
Before the ban, anticipation of heightened tensions led major tech firms to stockpile these critical components. Baidu, a frontrunner in China’s race to rival OpenAI, has amassed a significant supply of AI chips. These reserves are expected to sustain the development of its ChatGPT equivalent, Ernie Bot, for the next couple of years, according to CEO Robin Li. During a recent earnings call, Li expressed confidence in their chip reserves and alternative resources, ensuring the continuity of AI-native applications for end users. However, he acknowledged the long-term challenges in AI development due to difficulties in acquiring advanced chips, emphasizing the pursuit of alternatives.
In response to US export controls, Chinese tech behemoths like Baidu, ByteDance, Tencent, and Alibaba made preemptive strikes, ordering about 100,000 units of A800 processors from Nvidia, as reported by the Financial Times. These orders, amounting to an estimated $4 billion, demonstrate the strategic foresight of these companies. Additionally, they secured $1 billion worth of GPUs slated for delivery in 2024.
This trend of heavy investments could pose significant barriers to entry for AI startups. Only those with substantial funding, like 01.AI, founded by prominent investor Kai-Fu Lee, can afford to compete. 01.AI managed to acquire a notable number of high-performance inference chips through loans, which it quickly repaid after securing a valuation of $1 billion.
Baidu’s recent launch of Ernie Bot 4, powered by its GPU reserves, is a testament to its commitment to maintaining a competitive edge in AI. However, rating large language models (LLMs) remains a challenge due to their complexity. Many Chinese AI firms are focusing on meeting the criteria of LLM charts for a better ranking, but the real-world effectiveness of these models is yet to be fully assessed.
Smaller AI enterprises, lacking the financial muscle to stockpile chips, are left with two choices: settling for less powerful processors not affected by US export controls or waiting for potential acquisition opportunities. Li predicts that the confluence of advanced chip scarcity, high demand for data and AI talent, and substantial upfront investments will soon lead the industry towards a consolidation phase.
In conclusion, the US chip export ban has created a complex landscape for China’s AI sector. While tech giants with deep pockets are adapting and thriving, smaller players face an uphill battle. This dynamic underscores the broader implications of the US-China tech war, highlighting the strategic maneuvers and resilience of Chinese tech firms in the face of geopolitical challenges. As the industry navigates through these turbulent times, the evolution of AI in China remains a key area to watch, marked by innovation, adaptation, and a relentless pursuit of technological advancement.
