The US Really Doesn’t Want China To Get Their Hands On Advanced AI Chips
Trends

October 17, 2023
In the hands of ByteDance, advanced US AI chips could make us more addicted to TikTok. Obtained by the Chinese military, they could be used to develop killer robots. Either way, it’s a lose-lose for Americans.
And that’s a situation the US doesn’t want. Yesterday, the US tightened export controls of advanced AI chips — making it even harder for US chipmakers like Nvidia (NASDAQ:NVDA), AMD (NASDAQ:AMD) and Intel (NASDAQ:INTC) to sell to China.
What’s changing? The first set of restrictions was announced a year ago. Since then, chipmakers have created modified versions of their AI chips to bypass these rules. The new changes would make it even harder for chipmakers to work around the rules.
- Shares of all three companies fell yesterday, with Nvidia dropping the most — as 20-25% of its data center sales came from China.
- But AI chip demand has been so strong that earlier this year, Nvidia said it wouldn’t see any “immediate material impact” from the restrictions.
Instead, Chinese companies have rushed to buy AI chips this year. Intel, which made 27% of its sales from China in 2022, has seen a surge in demand for its chips made explicitly for the region in recent months.
“War of a hundred models”
That’s how one Tencent exec described China’s current state of AI competition. While China was late to AI, their tech companies have rapidly deployed over 130 large language models (LLMs), making up ~40% of the global total, compared to the US’ share of ~50%, according to brokerage CLSA.
Even with the controls, Baidu (NASDAQ:BIDU), the ‘Google of China,’ said yesterday that its latest AI chatbot, ERNIE 4.0, achieved a performance level comparable to OpenAI’s GPT-4. If true, that would put Baidu’s AI model ahead of Meta.
- Industry experts say there are too many similar models without viable business cases. This, along with increasing competition and US chip restrictions, risks driving up costs.
- Macquarie Group’s Esme Pau told Reuters, “LLMs with lower capacities will gradually be eliminated” in the next six to twelve months.
Forward-looking: Gobi Partners GBA’s managing director said early investors have regretted investing in LLM firms at the peak of the hype. While the initial rush to build AI models has helped double Nvidia’s sales in the recent quarter, demand could slow if funding dries up or export restrictions worsen.