OpenAI, the creator of ChatGPT, is designing a custom AI chip in partnership with Broadcom, according to a Motley Fool report. The effort represents a direct challenge to Nvidia's near-monopoly on AI training and inference hardware, specifically targeting the pricing leverage Nvidia has enjoyed amid surging demand.
The chip is reportedly intended for both training and running AI models, a segment where Nvidia's GPUs currently command over 80% market share. By developing its own silicon, OpenAI aims to reduce its dependence on a single supplier and potentially lower the enormous compute costs behind products like ChatGPT.
Analysts note that Broadcom brings extensive experience in custom chip design through its ASIC (application-specific integrated circuit) business, which already serves clients like Google and Meta. OpenAI's move comes as the AI industry grapples with supply constraints and rising expenses tied to Nvidia's premium-priced hardware.
However, experts caution that a custom chip is unlikely to topple Nvidia's dominance overnight. The Santa Clara-based company benefits from deeply entrenched software ecosystems like CUDA, which developers rely on for AI workloads. Building equivalent software support would take years and significant investment.
Nvidia's core advantage remains its integrated hardware-software stack, a barrier that custom chips alone may not breach. For now, OpenAI's project is more about cost control than market disruption, though it signals growing unease among AI leaders about vendor lock-in.