In a strategic move, Broadcom has teamed up with OpenAI to bolster its ambitious AI infrastructure endeavors. The collaboration, announced today, involves the development of custom silicon designed specifically for AI accelerators, with a monumental capacity of 10 GW. This joint venture is set to commence deployment late next year. The initiative allows OpenAI to directly incorporate its frontier model learnings into tailored hardware solutions, unlocking new potentials. As part of the agreement, infrastructure scaled with Broadcom’s Ethernet and connectivity solutions will be integrated across OpenAI’s and partner data centers. OpenAI’s CEO, Sam Altman, emphasizes that creating proprietary accelerators is crucial for advancing AI to benefit humanity universally. An accompanying podcast featuring Altman and Broadcom CEO Hock Tan highlighted the vision of a world where AI can meet the global demand for advanced intelligence. Altman noted that the assembly will focus on inference capabilities, describing 10 GW as a remarkable addition to their existing infrastructure. Broadcom’s Semiconductor Solutions Group President, Charlie Kawwas, commented on their portfolio’s completeness with Ethernet, PCIe, and optical solutions leading the AI infrastructure space. With its models assisting in the chip design process, OpenAI sees this partnership expediting the path to an ecosystem where each individual could access their dedicated AI accelerator. Nevertheless, these ambitious projects leave OpenAI relying on weaving intricate economic interdependencies with tech giants, despite not expecting to achieve cashflow positivity for several years. Market observers remain wary of potential overvaluation in the AI sector, drawing parallels with past tech bubbles.