The US ban on chip exports is hurting Chinese AI startups, not so much the giants yet

By RockedBuzz 3 Min Read

Well earlier than Washington banned Nvidia’s exports of high-performance graphics processing items to China, the nation’s tech giants had been stockpiling them in anticipation of an escalating tech struggle between the two nations.

Baidu, one in every of the tech firms constructing OpenAI’s Chinese counterparts, has secured sufficient AI chips to proceed coaching its ChatGPT equal Ernie Bot for “the subsequent yr or two,” firm CEO Robin Li mentioned in a earnings call this week.

“Additionally, inference requires less powerful chips, and we believe our chip reserves, as well as other alternatives, will be sufficient to support many native AI apps for end users,” he mentioned. “And in the long term, having difficulty acquiring the most advanced chips inevitably affects the pace of AI development in China. So, we are proactively looking for alternatives.”

Other deep-pocketed Chinese tech firms have additionally taken proactive measures in response to U.S. export controls. Baidu, ByteDance, Tencent and Alibaba have collectively ordered about 100,000 items of Nvidia A800 processors to be delivered this yr, at a price of about $4 billion, in line with the Financial Times reported in August. They additionally bought $1 billion value of GPUs, scheduled for supply in 2024.

Such massive preliminary investments may simply dissuade many startups from coming into the LLM race. There are exceptions if the younger firm can rapidly safe main investments. 01.AI, based in late March by outstanding investor Kai-Fu Lee, has acquired a large variety of high-performance inference chips by loans and has already paid off its debt after elevating capital that valued it at $1 billion of {dollars}.

With its GPU reserve, Baidu lately launched the Ernie Bot 4, which Li says is “not inferior in any approach to the GPT-4.”

Evaluating LLMs is tough attributable to the sheer complexity of those AI fashions. Many Chinese AI firms have resorted to bettering rankings by diligently assembly the standards of LLM graphs, however the effectiveness of those fashions when utilized to actual purposes is nonetheless awaiting judgment.

Smaller AI operators, which haven’t got the money move to stockpile chips, should make do with much less highly effective processors that are not topic to U.S. export controls. Alternatively, they will anticipate potential acquisition alternatives. Li predicts that, with a confluence of things, together with the shortage of superior chips, excessive demand for information and AI expertise, and enormous upfront investments, the trade will quickly transfer right into a “consolidation phase.”

Valued at $1 billion, Kai-Fu Lee’s LLM startup unveils open supply mannequin

Share This Article
Leave a comment