AI chips are out of stock globally!

Google’s CEO likened the AI ​​revolution to humanity’s use of fire, but now the digital fire that fuels the industry—AI chips—is hard to come by.

The new generation of advanced chips that drive AI operations are almost all manufactured by NVIDIA. As ChatGPT explodes out of the circle, the market demand for NVIDIA graphics processing chips (GPUs) far exceeds the supply.

Sharon Zhou, co-founder and CEO of Lamini, a startup that helps companies build AI models such as chatbots, said:

"Because there is a shortage, the key is your circle of friends."

"

It's like toilet paper during the epidemic. "

Due to the existence of this situation, customers such as OpenAI, the creator of ChatGPT, are limited by the computing power of cloud service providers such as Amazon and Microsoft. To develop and operate increasingly complex models and help other companies build AI services, AI developers need access to server computing power.

Even the world’s most well-connected tech entrepreneurs are working to secure supply. At the AI ​​Congressional hearing on May 16, OpenAI CEO Sam Altman said that due to computing power bottlenecks, it would be better if fewer people used ChatGPT.

On May 23, Tesla CEO Musk said at the Wall Street Journal CEO Council Summit:

“

Currently, GPUs are harder to obtain than drugs.”

However, Musk is still much better.

Startups that said they were seeking Oracle's computing power earlier this year were suddenly told that a buyer had snapped up much of Oracle's spare server space, the Wall Street Journal reported. According to people familiar with the matter, these new startups have been informed that the buyer is Musk, who is developing a competing product called X.AI to challenge OpenAI’s market position.

Startups are trying various ways to solve the shortage

Without advanced graphics processing chips, the execution speed of large language models in artificial intelligence will be greatly reduced. This is a common view among the founders of startup companies. Nvidia's advanced graphics processing chips have excellent parallel computing capabilities, which is very critical for the operation of AI.

UBS analysts estimate that an early version of ChatGPT will require approximately 10,000 NVIDIA graphics processing chips; while Musk estimates that an updated version will require 3 to 5 times that number of advanced processors.

Nvidia recently said it was expanding supply to meet growing demand.

Nvidia CEO Jensen Huang said on Sunday that the company has increased production of its new flagship chip H100 for generative artificial intelligence.

AI startups and investors are doing their best to solve the chip shortage problem. Some investors are combing their computing networks for spare computing power, while others are arranging orders for high-capacity processors and servers to share with other AI startups.

Some startups are shrinking their AI models to increase efficiency; other founders are seeking to connect with salespeople at Amazon and Microsoft.

Sharon Zhou from Lamini stated that they have the necessary chips.． Lamini was co-founded with a former NVIDIA engineer. But she and many other founders declined to say how they obtained the chips.

Many AI founders expect the shortage of AI chips to last at least until next year.

Founders and investors say that

Even if there is an established company for AI chips, they still have to wait weeks to use them. The CEO of an AI startup said:

"Even if you have paid in advance, it does not mean that the GPU will be delivered to you the next day or next week, you can only wait."

The CEO of Supermicro, one of the world's largest server makers, said the company's backlog of graphics chip systems has reached an all-time high and the company is rushing to increase production capacity.

The secondary market is on fire, and Nvidia is also going crazy

The shortage has ignited the secondary market for AI chips, some of which involve large encryption companies. These companies purchased chips for mining during the boom of the currency circle, but now no longer do so during the downturn of the digital currency market. Chip required.

Demand for Nvidia’s products has driven the company’s stock up about 167% this year. AI chip costs vary, Some retailers are selling Nvidia's advanced AI chips for about $33,000, although they may sell for more on the secondary market due to high demand.

Kanjun Qiu, CEO of artificial intelligence research firm General Intelligent, has been buying advanced graphics chips for her servers since last year, allowing her to weather the current shortage.

A venture capitalist recently sent her a message asking if there was any excess capacity that could be leased to other startups. Qiu hasn't decided yet whether to give up her chip.

The above is the detailed content of AI chips are out of stock globally!. For more information, please follow other related articles on the PHP Chinese website!