The US chip export ban to China is causing a stir in the tech world, but it seems the country's tech giants were one step ahead. Anticipating a tech war escalation, companies like Baidu have been stockpiling high-performance graphic processing units (GPUs) from Nvidia.
On a recent earnings call, Baidu’s CEO Robin Li revealed that the firm has enough AI chips to keep training its ChatGPT equivalent, Ernie Bot, for the next couple of years. He also noted that inference requires less powerful chips, and the company's reserves should support numerous AI-native apps for end users.
However, the ban isn't without consequences. Li admitted that difficulties in acquiring advanced chips could slow down AI development in China, prompting the firm to seek alternatives.
Other tech heavyweights, including ByteDance, Tencent, and Alibaba, have also been proactive. These companies collectively ordered around 100,000 units of Nvidia's A800 processors to be delivered this year, costing them up to $4 billion. They also invested $1 billion in GPUs scheduled for delivery in 2024.
Such hefty upfront investments could deter startups from joining the Large Language Model (LLM) race. However, exceptions exist. For instance, 01.AI, founded by prominent investor Kai-Fu Lee, quickly secured substantial investments, allowing it to acquire high-performance inference chips and pay off its debt.
Despite the chip reserves, rating LLMs remains challenging due to their complexity. While many Chinese AI firms resort to ranking boosting by fulfilling LLM chart criteria, the effectiveness of these models in real-life applications is yet to be proven.
For smaller AI players lacking the financial muscle to hoard chips, the options are to settle for less powerful processors not under US export controls or await potential acquisition opportunities. Li predicts that factors like advanced chip scarcity, high data and AI talent demand, and significant upfront investments will soon trigger a "consolidation stage" in the industry.
Made with TRUST_AI - see the Charter: https://www.modelprop.co.uk/trust-ai
Comments