Broadcom Reportedly Conditions OpenAI Custom Chip Production on Microsoft Purchasing 40 Percent
OpenAI, the AI research organization behind ChatGPT, is pushing forward with ambitious plans to develop its own custom AI chips as a means to lessen its dependence on Nvidia’s dominant graphics processing units. However, recent reports indicate a significant hurdle in these efforts involving chip designer Broadcom. According to sources cited by Nikkei Asia, Broadcom will not proceed with building OpenAI’s bespoke silicon unless Microsoft, OpenAI’s largest investor and cloud computing partner, commits to purchasing 40 percent of the produced chips.
This development underscores the complex dynamics at play in the semiconductor industry, particularly as AI demand surges and supply chain constraints intensify. OpenAI has been in discussions with Broadcom for the design phase of its custom chip project and Taiwan Semiconductor Manufacturing Company (TSMC) for the actual fabrication. The custom chips aim to optimize inference and training workloads for large language models, potentially offering cost savings and performance tailored specifically to OpenAI’s needs.
Broadcom’s stipulation reflects a pragmatic approach to risk management in an industry fraught with uncertainties. Producing custom chips requires substantial upfront investment in research, development, and production tooling. Without assured demand, such ventures carry high financial exposure. By requiring Microsoft to absorb 40 percent of the output, Broadcom seeks to guarantee a baseline volume that justifies the commitment. Microsoft, which has poured billions into OpenAI through equity investments and Azure cloud infrastructure deals, already hosts much of OpenAI’s computing workload. This makes it a logical candidate to integrate the custom chips into its data centers.
The report highlights that OpenAI envisions multiple iterations of these chips. Initial versions would focus on inference tasks, with subsequent generations incorporating training capabilities. Production timelines suggest mass manufacturing could begin as early as late 2025, aligning with OpenAI’s aggressive roadmap to scale AI capabilities amid competition from rivals like Anthropic and Google DeepMind.
Geopolitical tensions further complicate the landscape. US export controls on advanced semiconductors to China have heightened caution among chipmakers. Broadcom, which derives significant revenue from global markets, may view the Microsoft commitment as a safeguard against potential disruptions or shifts in demand. OpenAI’s chip initiative is part of a broader industry trend where hyperscalers and AI firms seek to insource silicon design. Companies like Google with its Tensor Processing Units, Amazon with Trainium and Inferentia, and Meta with its MTIA chips have paved the way, demonstrating that custom silicon can deliver efficiency gains over off-the-shelf GPUs.
For OpenAI, reliance on Nvidia has been both a boon and a burden. Nvidia’s Hopper and Blackwell GPUs power much of the company’s training runs, but soaring prices and supply shortages have prompted diversification. CEO Sam Altman has publicly advocated for new chip architectures, even floating ideas for national-scale semiconductor investments. The Broadcom partnership positions OpenAI to leverage expertise from a firm renowned for networking and custom ASICs, including work for Google’s TPUs.
TSMC’s involvement is equally critical. As the world’s leading foundry, it would handle the 3nm or finer process nodes necessary for cutting-edge AI accelerators. However, Broadcom’s design leadership ensures the chip’s architecture aligns with OpenAI’s software stack, including optimizations for its proprietary models.
Microsoft’s potential role adds another layer of intrigue. The software giant has its own Maia chips for AI workloads but continues heavy investment in OpenAI. A 40 percent purchase commitment would integrate OpenAI’s silicon into Azure, potentially strengthening their symbiotic relationship. Reports suggest OpenAI approached Microsoft early in the chip project, recognizing the necessity of its buy-in for commercial viability.
Neither Broadcom, OpenAI, nor Microsoft has publicly confirmed the details. OpenAI has acknowledged pursuing custom hardware but provided no specifics on partners or timelines. Broadcom’s spokesperson declined comment, while Microsoft emphasized its ongoing collaboration without addressing chip production.
This impasse could delay OpenAI’s silicon ambitions if unresolved. Alternative foundries like Samsung exist, but Broadcom’s design prowess makes it a preferred partner. Success here could accelerate OpenAI’s path to greater autonomy, reducing per-token compute costs and enabling novel AI architectures.
Industry analysts view the condition as standard practice for custom chip deals. Guaranteeing offtake from a creditworthy buyer like Microsoft de-risks the project for Broadcom, potentially paving the way for scaled production. As AI hardware evolves, such strategic alliances will shape the next wave of innovation.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.