OpenAI, the Power Behind ChatGPT, Explores Manufacturing Its Own AI Chips

Oct 7, 2023

OpenAI, the mastermind behind the revolutionary language model ChatGPT, is reportedly contemplating the creation of its own AI chips. The company is said to be considering a potential acquisition target as part of this initiative, according to insiders privy to OpenAI’s plans.

The decision to proceed with this ambitious venture is still up in the air, as per recent internal discussions leaked to Reuters. However, for at least the past year, OpenAI has been mulling over various strategies to tackle the scarcity of high-cost AI chips it heavily relies on. These deliberations have included the possibility of manufacturing its own AI chip, strengthening ties with other chipmakers including Nvidia, and expanding its supplier base beyond Nvidia.

OpenAI has refrained from commenting on these reports. However, CEO Sam Altman has been vocal about the need for more AI chips, making it a priority for the company. He has expressed discontent about the dearth of graphics processing units (GPUs), a sector largely monopolized by Nvidia that commands over 80% of the global market for chips optimally designed to run AI applications.

Altman’s push for more chips is driven by two primary concerns: the shortage of advanced processors that fuel OpenAI’s software, and the hefty costs associated with operating the hardware necessary to power its products and efforts.

Since 2020, OpenAI has been developing its generative artificial intelligence technologies on a colossal supercomputer built by Microsoft, one of its biggest supporters. This supercomputer utilizes 10,000 of Nvidia’s GPUs.

Operating ChatGPT comes with a significant price tag for the company. According to Bernstein analyst Stacy Rasgon, each query costs approximately 4 cents. If ChatGPT queries were to reach one-tenth the scale of Google search, it would necessitate an initial investment of about $48.1 billion in GPUs and roughly $16 billion per year in chips to maintain operations.

The move to develop its own AI chips would position OpenAI among a select group of major tech companies, including Google and Amazon, that have sought to gain control over the design of chips integral to their operations.

However, it remains unclear whether OpenAI will proceed with plans for a custom chip. Such a move would represent a significant strategic undertaking and a substantial investment potentially running into hundreds of millions of dollars per year, as per industry insiders. Even dedicating resources to this endeavor would not guarantee success.

Acquiring a chip company could expedite the process of creating OpenAI’s proprietary chip, similar to Amazon’s acquisition of Annapurna Labs in 2015.

OpenAI had considered this route to the extent that it performed due diligence on a potential acquisition target, according to a source close to its plans. The identity of the company OpenAI was considering purchasing remains unknown.

If OpenAI decides to pursue plans for a custom chip, including a possible acquisition, the venture is likely to span several years, leaving the company reliant on commercial providers like Nvidia and Advanced Micro Devices in the interim.

Other major tech firms have attempted to build their own processors for years, often with limited success. Meta’s custom chip project has faced setbacks, causing the company to discard some of its AI chips, as per a Reuters report. The Facebook owner is now focusing on a newer chip designed to cater to all kinds of AI work.

OpenAI’s primary backer, Microsoft, is also working on a custom AI chip that OpenAI is currently testing, as reported by The Information. This development could indicate a growing divergence between the two companies.

The demand for specialized AI chips has skyrocketed since the launch of ChatGPT last year. These specific chips, or AI accelerators, are crucial for training and running the latest generative AI technology. Nvidia, one of the few chipmakers that manufacture useful AI chips, continues to dominate the market.