At one time, OpenAI was considering the acquisition of Cerebras, a company specializing in AI chip manufacturing that’s currently preparing to go public, according to new legal documents.
As part of Elon Musk's lawsuit against OpenAI, recent exhibits show that OpenAI considered acquiring Cerebras around 2017, about a year after the chipmaker was founded and shortly after OpenAI’s launch. In an email to OpenAI’s CEO Sam Altman and Musk, OpenAI co-founder and former chief scientist Ilya Sutskever suggested buying Cerebras through Tesla, Musk’s electric vehicle company. At the time, Musk had financial ties to OpenAI and held some influence over its strategy.
“If we decide to buy Cerebras, it would likely be done through Tesla,” Sutskever wrote in September 2017. “However, it might be better to proceed from within OpenAI, given Tesla’s shareholder obligations that may conflict with OpenAI’s mission, which might not yield optimal results for OpenAI.”
An earlier email from July 2017 from Sutskever to Musk and OpenAI co-founder Greg Brockman (now OpenAI’s president) listed Cerebras-related action items, including “Negotiate merger terms with Cerebras” and “Conduct further due diligence with Cerebras.”
However, the merger plans were ultimately abandoned, though the exhibits don’t clarify why. OpenAI subsequently put its chip-related ambitions on hold for years.
Cerebras, headquartered in Sunnyvale, California, designs specialized hardware for AI model training and execution, claiming their chips are more efficient and faster than Nvidia’s leading AI chip options. Cerebras has raised $715 million in venture capital and is expected to aim to double its $4 billion valuation with its IPO. Nonetheless, the company faces hurdles: a significant 87% of its 2024 first-half revenue came from G42, a single Abu Dhabi-based firm, which has historical ties to China that concern U.S. lawmakers. Additionally, Cerebras CEO Andrew Feldman has a controversial background, having previously pled guilty to circumventing accounting controls while he was a VP at Riverstone Networks.
If the acquisition had materialized, both companies could have benefited. Cerebras might have sidestepped a challenging IPO, while OpenAI would have gained a key resource in its pursuit of in-house chip capabilities.
Reducing dependence on Nvidia, a dominant player in AI chip technology, has long been a goal for OpenAI. As it lags behind in developing in-house chips, especially compared to Google and Amazon Web Services, OpenAI faces pressure to cut the costs of training, refining, and deploying models. Owning its own chips could be a way to achieve the necessary cost savings.
OpenAI once aspired to create a network of chip production facilities and was exploring acquisition options. However, it now appears to have shifted course, instead focusing on building a team of chip designers and engineers and collaborating with Broadcom and TSMC to develop an AI chip tailored to run models, expected as early as 2026.
Post a Comment