Properly earlier than Washington banned Nvidia’s exports of high-performance graphic processing models to China, the nation’s tech giants had been hoarding them in anticipation of an escalating tech conflict between the 2 nations.
Baidu, one of many tech companies constructing China’s counterparts to OpenAI, has secured sufficient AI chips to maintain coaching its ChatGPT equal Ernie Bot for the “subsequent yr or two,” the agency’s CEO Robin Li mentioned on an earnings name this week.
“Additionally, inference requires much less highly effective chips, and we consider our chip reserves, in addition to different options, can be ample to assist a number of AI-native apps for the top customers,” he mentioned. “And in the long term, having difficulties in buying essentially the most superior chips inevitably impacts the tempo of AI improvement in China. So, we’re proactively looking for options.”
Different deep-pocketed Chinese language tech corporations have additionally been taking proactive measures in response to U.S. export controls. Baidu, ByteDance, Tencent and Alibaba collectively ordered round 100,000 models of A800 processors Nvidia to be delivered this yr, costing them as a lot as $4 billion, the Monetary Occasions reported in August. Additionally they bought $1 billion price of GPUs which can be scheduled for supply in 2024.
Such heavy upfront investments might simply deter many startups from getting into the LLM race. Exceptions do exist if the younger enterprise manages to safe good-looking investments rapidly. 01.AI, which was based in late March by outstanding investor Kai-Fu Lee, acquired a considerable variety of high-performance inference chips by loans and has already paid off its debt after elevating capital that valued it at $1 billion.
With its reserve of GPUs, Baidu lately launched the Ernie Bot 4, which Li claimed is “not inferior the least bit to GPT-4.”
Ranking LLMs is hard because of the sheer complexity of those AI fashions. Many Chinese language AI companies have resorted to rating boosting by diligently fulfilling the standards of LLM charts, however the effectiveness of those fashions when utilized to actual purposes real-life continues to be pending judgment.
Smaller AI gamers, missing the money movement to hoard chips, must accept much less highly effective processors that aren’t underneath U.S. export controls. Alternatively, they will await potential acquisition alternatives. Li expects that with a confluence of things, together with the shortage of superior chips, excessive demand for information and AI expertise, and large upfront investments, the business will quickly transition right into a “consolidation stage.”