OpenAI was once considering the possibility of acquiring Cerebras, an AI chip manufacturer that is currently in the process of going public, as shown in new legal documents.

Elon Musk’s lawsuit against OpenAI has revealed that the organization was thinking about acquiring Cerebras around 2017, shortly after Cerebras was founded and a few years after OpenAI was established.

In an email to OpenAI CEO Sam Altman and Musk, Ilya Sutskever, one of OpenAI’s co-founders and former chief scientist, suggested the idea of purchasing Cerebras through Tesla, Musk’s electric vehicle company. Musk was financially involved with OpenAI at the time and had some influence over its decisions.

Sutskever wrote in September 2017, “If we decide to acquire Cerebras, my intuition is that we should do it through Tesla.” However, he expressed concerns about aligning Tesla’s duty to its shareholders with OpenAI’s mission. Ultimately, the deal did not materialize, and OpenAI put its chip ambitions on hold for a period.

Cerebras, located in Sunnyvale, California, specializes in building custom hardware to operate and train AI models, claiming that its chips are faster and more efficient than Nvidia’s leading offerings for AI tasks.

Having raised $715 million in venture capital, Cerebras aims to double its $4 billion valuation through an IPO. However, the company faces significant challenges, including its reliance on a single Abu Dhabi firm for the majority of its revenue and concerns about historic ties to China. Cerebras CEO Andrew Feldman also has a controversial history, having previously admitted to circumventing accounting controls.

If the acquisition had occurred, both companies could have benefited. Cerebras might have avoided the complexities of an IPO, while OpenAI could have gained a crucial asset in its efforts to develop in-house chips and reduce reliance on Nvidia.

OpenAI had previously considered establishing a network of chip manufacturing factories and acquiring a target company. However, the organization has shifted its focus to building a team of chip designers and engineers and collaborating with semiconductor firms to create an AI chip for model operations, possibly available by 2026.