OpenAI decided to build cores and “grab the cake” with Nvidia

As a popular science fact, for every question you ask ChatGPT, OpenAI, the parent company behind it, spends 4 cents (about 0.29 yuan). 4 cents seems like a small amount, but what if you multiply it by 13 million?

According to a UBS report, ChatGPT had 100 million monthly active users in February, with about 13 million unique visitors every day. To put it simply, if every user asks one question (and only asks one question) when logging in on a single day, the single-day operating cost of ChatGPT will be at least US$520,000. An INSIDER analyst estimated that the daily cost of ChatGPT is as high as $700,000.

Hardware overhead accounts for the majority of AI operating costs. The importance of chips to the technology industry is self-evident, especially for AI companies. Therefore, companies have to operate with high hardware costs. The only way to get rid of this situation is to rely on self-developed chips.

"Creating cores" has become something that AI companies must consider when they develop to a certain level.

According to Reuters, OpenAI may be exploring the independent development of a smart chip designed for AI. The benefits of doing so are obvious and can effectively reduce AI operating costs. Take Google, OpenAI's biggest competitor in the AI ​​field, as an example. It plans to exclude Broadcom from the ranks of AI smart chip suppliers in 2027. This move can save Google billions of dollars every year.

In addition to reducing costs, getting rid of dependence on other companies is another factor in OpenAI's decision to develop its own research.

The AI ​​chip market is currently dominated by Nvidia, which accounts for about 80% of the global AI chip market. With the rapid development of AI, AI companies' demand for chips has increased, and Nvidia's supply capacity seems to be somewhat inadequate at this time. Even if Nvidia increases the production of H100 chips to 1.5 million to 2 million units in 2024, it will not be able to reverse the shortage of supply. OpenAI CEO Sam Altman has publicly complained that the shortage of Nvidia chips has affected the company's expansion speed.

But "making a chip" is not an easy task, and it requires high time and money costs. Alex White, general manager of the AI ​​company SambaNova Systems, said, "Designing and manufacturing chips cannot be completed overnight. It requires a lot of professionalism." knowledge and increasingly scarce resources. OpenAI took over five years to develop GPT-4. I wouldn’t be surprised if the hardware took a similar amount of time.”

Even if it invests manpower, financial resources and time, OpenAI may not be able to successfully develop a chip that meets the requirements. This is the difficulty of self-developed chips. To reduce costs and increase success rate, acquiring a mature chip company is the best solution. OpenAI has this intention. Recently, OpenAI is conducting due diligence on potential acquisition targets, but the specific identities of these companies are still unknown.

In addition to OpenAI, both Meta and Microsoft have spent many years working on developing chips.

According to Reuters, although Meta has successfully developed a custom chip, there have been problems with the chip, causing Meta to abandon some of its AI chips. Zuckerberg is working on developing a newer chip and plans to make the chip support Meta. All types of AI under its umbrella.

Microsoft's process is relatively smoother. As early as 2019, Microsoft was secretly developing an AI chip codenamed Athena. According to reports, Microsoft plans to use this chip more widely within Microsoft and OpenAI as early as next year. Microsoft insiders said that Microsoft does not expect to replace Nvidia chips with Athena chips in the short term. Its main purpose is to reduce its dependence on Nvidia.

It can be seen that giant companies such as OpenAI and Microsoft are unwilling to rely on others. Reducing their dependence on other companies is one of their reasons for "making cores." Reuters speculates that OpenAI’s consideration of developing its own chips is the latest sign of the company parting ways with its partner Microsoft.

It is true that OpenAI's core-building plan is still in its early stages, and the market will not see the results of OpenAI's research and development until at least a few years later. Before this, OpenAI had only two options, use Nvidia chips at high prices or be highly dependent on Microsoft.

# Welcome to follow the official WeChat public account of aifaner: aifaner (WeChat ID: ifanr). More exciting content will be provided to you as soon as possible.

Ai Faner | Original link · View comments · Sina Weibo