OpenAI, responsible for ChatGPT, is delving into the idea of producing its own AI chips and has even considered potential acquisition prospects, sources close to the company reveal.
Since last year, OpenAI has been weighing various solutions to address its dependency on high-priced AI chips. The options considered range from creating a proprietary AI chip and intensifying collaborations with chip manufacturers like Nvidia, to diversifying its supplier base beyond Nvidia.
The organization chose not to respond when asked for comments.
For CEO Sam Altman, securing more AI chips has become a chief concern. He has been vocal about the dearth of graphics processing units (GPUs), with Nvidia dominating over 80% of this global market, essential for AI tasks. Altman’s urgency stems from two primary issues: the shortage of advanced chips for OpenAI’s applications and the staggering costs of the necessary hardware.
From 2020, OpenAI leveraged Microsoft’s colossal supercomputer, with 10,000 Nvidia GPUs, to advance its generative AI tech. Operating ChatGPT doesn’t come cheap for OpenAI. A single query averages around 4 cents, as Bernstein’s Stacy Rasgon estimates. If ChatGPT queries approach a fraction of Google search’s scale, the initial GPU investment could touch $48.1 billion, with an annual $16 billion upkeep.
By looking into designing their own AI chips, OpenAI joins tech giants like Google and Amazon, who are trying to control chip designs crucial for their operations. However, whether OpenAI will commit to a custom chip project remains uncertain, as it represents a substantial commitment that could incur hundreds of millions annually, warn industry insiders. Success isn’t guaranteed even with dedicated resources.
An acquisition might expedite OpenAI’s journey to chip independence, reminiscent of Amazon’s acquisition of Annapurna Labs in 2015. Insiders say OpenAI has already investigated a potential acquisition, though the company remains unknown. Even if OpenAI greenlights its chip plans or an acquisition, realizing this vision might span several years. In the interim, reliance on commercial suppliers like Nvidia and AMD is inevitable.
Some major tech firms have ventured into chip creation but with mixed outcomes. For instance, Meta’s custom chip initiative encountered setbacks, leading them to abandon specific AI chips and pivot to a newer universal AI chip, as per Reuters. Interestingly, Microsoft, OpenAI’s principal supporter, is also working on a custom AI chip, which OpenAI is currently examining, as reported by The Information. This might hint at a growing divergence between the two entities.
The introduction of ChatGPT last year has exponentially driven the demand for AI-specific chips. Nvidia stands out among the few companies producing these essential AI chips, holding a dominant market position.