OpenAI was at one point considering acquiring Cerabrasan AI chip manufacturing company that is in the process of going public, according to new legal documents.

That of Elon Musk ongoing trial against OpenAI features new exhibits that describe how OpenAI planned to take over Cerebras around 2017 – a year after Cerebras was founded and just a few years after it began operations.

In a e-mail addressed to OpenAI CEO Sam Altman and Musk, Ilya Sutskever, one of OpenAI’s co-founders and former chief scientist, floated the idea of ​​buying Cerebras through Musk’s EV company Tesla . At the time, Musk was financially involved in OpenAI and had some influence over its direction.

“If we decide to buy Cerebras, my impression is that it will be done through Tesla,” Sutskever wrote in September 2017. “But why do it this way if we could also do it from OpenAI? Specifically, the concern is that Tesla has a duty to shareholders to maximize shareholder return, which is not consistent with OpenAI’s mission. The overall result might therefore not be optimal for OpenAI.

In an earlier email dated July 2017 to Musk and OpenAI co-founder Greg Brockman (now president of the company), Sutskever mentions several agenda items related to Cerebras: “Negotiate the terms of the merger with Cerebras” and “Further due diligence with Cerebras”.

The merger deal would eventually fail, although the evidence does not clearly show why. And OpenAI would end up shelving its chip ambitions for years.

Cerebras, based in Sunnyvale, California, builds custom hardware to run and train AI models and says its chips are faster and more efficient than Nvidia’s flagship offerings for AI workloads.

After raising $715 million in venture capital, Cerebras is reportedly looking to roughly double its $4 billion valuation through its IPO. But he faces considerable challenges. A single Abu Dhabi company, G42, accounted for 87% of Cerebras’ revenue in the first half of 2024, and U.S. lawmakers have expressed concern about G42’s historic ties to China. Cerebras CEO Andrew Feldman also has a checkered past, having pleaded guilty to circumvent accounting controls while serving as vice president of publicly held Riverstone Networks.

Had the acquisition gone ahead, it could have benefited both companies. Cerebras would have avoided the tricky IPO route, while OpenAI could have had a vital resource in its race to build in-house chips.

OpenAI has long sought to reduce its reliance on Nvidia, which has a massive share of the market for AI-optimized chips. While OpenAI lags behind in internal chipset – Google And Amazon Web Servicesamong others, has long offered chips designed for AI workloads – they are under pressure to reduce the costs of training, tuning and running models. Having his own chips could be a way to get the discounts he needs.

OpenAI hoped at one point to establish a network of factories for chip manufacturing, and was considering an acquisition target. But it is would have scrapped those plans in favor of aggressively building a team of chip designers and engineers and working with semiconductor companies Broadcom and TSMC to create an AI chip to run models. It could arrive as early as 2026.