According to media reports, generative AI leader OpenAI is exploring self-developed AI chips and has even evaluated potential acquisition targets。

People familiar with the matter said that as the shortage of chips to train AI models worsened, discussions within OpenAI about AI chip strategies have been going on since at least last year.。OpenAI is considering a variety of strategies to advance its "self-sufficiency" plan for chips, including working with more chipmakers to diversify suppliers, acquiring AI chipmakers, and stepping up efforts to design chips in-house.。

The report also mentioned that OpenAI CEO Sam Altman has made the purchase of more AI chips a top priority for the company.。Currently, OpenAI, like most of its competitors, relies on GPU-based hardware to develop models such as ChatGPT, GPT-4, and DALL-E 3.。GPU's ability to perform multiple computations in parallel makes it ideal for training today's most powerful artificial intelligence。

The boom in generative artificial intelligence has brought a windfall to GPU makers such as Nvidia, and its chips are in short supply.。At present, more than 80% of the world's most suitable chips for running AI applications are "monopolized" by Nvidia.。It is reported that Nvidia's best AI chips will be sold out by 2024.。Microsoft warned in its second-quarter earnings report that a severe shortage of server hardware needed to run AI could lead to service disruptions。

GPUs are essential for OpenAI to run and service its AI models。OpenAI currently relies on GPU clusters on the cloud to execute customer workloads, which is extremely costly。An analysis by Bernstein analyst Stacy Rasgon found that if ChatGPT queries grew to one-tenth the size of Google searches, about $48.1 billion worth of GPUs would be needed, and about $16 billion worth of chips a year would be needed to keep running.。

OpenAI isn't the first company to experiment with self-developed AI chips。Because OpenAI's chip shortage and cost problems are also common problems faced by other technology companies.。

At this stage, Google has TPU (Tensor Processing Unit) that can be used to train large generative AI models such as PaLM-2 and Imagen.。Amazon offers AWS customers proprietary chips for training and inference。Microsoft is reportedly working with AMD on an in-house AI chip called Athena, which OpenAI is also testing.。

But for tech companies, it's not easy to "shake off" Nvidia by yourself.。

On the one hand is the technical problem。Nvidia, AMD and other chip manufacturers through years of technology accumulation, has built a chip manufacturing technology "moat," if the enterprise from scratch, want to turn over this "moat" is not easy。Meta has been pushing ahead with custom chip-related work, but it's not going well and has led to the company scrapping some of its AI chips。But even so, Meta has not given up, and it is understood that the company is developing an updated chip internally that will cover all types of artificial intelligence work.。

On the other hand, the amount of investment required for self-developed chips is huge.。In addition to large technology companies such as Google and Microsoft, it is difficult for ordinary small and medium-sized technology companies, especially start-ups, to have enough money to support chip research.。

Chipmaker Graphcore, once seen as a British AI "unicorn," is an example of stepping down from the "altar."。Graphcore was founded in 2016 and its main business includes chip design for AI projects.。The company was valued at approximately $2.8 billion in 2020.。But as losses have continued, the company is now struggling to operate.。The company's IPU products were used on Microsoft's Azure platform, but its revenue fell again and again last year after a failed partnership with Microsoft, and its valuation fell by about $1 billion.。Graphcore closed offices in Norway, Japan and South Korea last year and started layoffs。Graphcore isn't the only startup chip company struggling。Intel's artificial intelligence chip company Habana Labs also started layoffs this year in an effort to reduce costs, laying off about 10% of its employees。

However, as a star-rated start-up, OpenAI doesn't seem to need to worry about funding。The company has raised more than $11 billion in venture capital, and the various commercializations the company has been advancing are expected to generate close to $1 billion in revenue annually.。In addition, there have been recent media reports that OpenAI is considering selling shares, and its secondary market valuation could soar to $90 billion.。

However, even if OpenAI self-developed chips are successful and brought to market as desired, the process will take years and the investment required will be as high as hundreds of millions of dollars.。It remains to be seen whether OpenAI investors will make such risky bets。