OpenAI, the company that develops ChatGPT, is considering its own AI chips and has begun evaluating a potential acquisition target.
According to recent internal discussions revealed to Reuters, the company has not yet decided on further action. However, since at least last year, the company has been discussing various options to solve the shortage of expensive AI chips that OpenAI relies on, according to people familiar with the matter.
These include making its own AI chips, working more closely with other chipmakers, including Nvidia, and diversifying suppliers (not just Nvidia).
CEO Sam Altman has previously made purchasing more AI chips a top priority for the company. He has publicly complained about GPU shortages.
GPUs are the best chips to run a variety of AI applications, a market dominated by NVIDIA, which controls more than 80 percent of the global market share.
The effort to acquire more chips is related to two major issues Altman points to: the shortage of advanced processors to power OpenAI software and the "ridiculously high" cost of running the hardware needed to power its workloads and products.
Since 2020, OpenAI has developed its generative AI technology on a large supercomputer built by Microsoft, one of its biggest proponents, which uses 10,000 NVIDIA GPUs. Running ChatGPT is very expensive for this company.
Analysis by Bernstein analyst Stacy Rasgon shows that each query costs about 4 cents. If ChatGPT's query volume grows to one-tenth of Google's search volume, it will initially require about $48.1 billion worth of GPUs and about $16 billion worth of chips per year to keep running.
The era of custom chips
A handful of big tech companies, such as Amazon and Alphabet's Google, seek to have control over the effort: designing chips that are critical to their businesses, and trying to develop their own AI chips will put OpenAI among them.
It's unclear whether OpenAI will focus on plans to build a custom chip.
According to industry veterans, doing so would be a major strategic move and a huge investment, potentially costing billions of dollars a year.
Even if OpenAI invests a lot of resources into this heavy task, there is no guarantee of success. Acquiring a chip company could speed up OpenAI's efforts to develop its own chips, as Amazon did with Israeli chip startup Annapurna Labs in 2015.
According to a person familiar with OpenAI's plans, OpenAI considered this path before and even conducted due diligence on a potential acquisition target.
The identity of the company that OpenAI is considering acquiring is unknown.
Even if OpenAI continues to pursue plans for custom chips, including acquisitions, the work could take several years, during which time the company continues to rely on commercial vendors such as Nvidia and AMD.
Some big tech companies have been building their own processors for years, with little success.
According to Reuters, Meta's work to develop custom chips has been problematic, prompting the company to eventually scrap some AI chips. Meta is now developing a new chip that will cover all types of AI workloads.
According to The Information, Microsoft, the main proponent of OpenAI, is also developing a custom AI chip, and OpenAI is being tested.
Since ChatGPT's launch last year, there has been a surge in demand for dedicated AI chips.
To train and run the latest generative AI technologies, specialized chips or AI accelerators are essential. Nvidia is one of the few chipmakers that produce practical AI chips and dominate the market.