OpenAI has formed a major partnership with Broadcom to co-develop powerful custom chips and networking systems. The goal is to build faster, smarter, and more energy-efficient infrastructure to support OpenAI’s growing suite of artificial intelligence (AI) products.
The ambitious project, which begins in 2026 and aims for completion by 2029, targets a massive 10 gigawatts of deployment capacity.
Why OpenAI is Building Its Own Hardware
This collaboration allows OpenAI to directly embed the performance and efficiency insights gained from developing frontier models like GPT into the hardware itself.
- Sam Altman, CEO of OpenAI, called the partnership a “critical step” in building the infrastructure needed to unlock AI’s potential.
- The deal involves Broadcom supplying its high-speed Ethernet networking technology to connect the AI systems (racks), while OpenAI designs the specialized accelerators and systems.
- Hock Tan, CEO of Broadcom, noted that the co-development of “next-generation accelerators” marks a pivotal moment in the pursuit of artificial general intelligence (AGI).
Context: OpenAI’s Aggressive Expansion
This move into custom chip design is part of OpenAI’s massive expansion streak in 2025, which has solidified its position as one of the world’s most valuable private tech firms (recently valued at around $300 billion).
Other major strategic deals include:
- Computing Power: A $300 billion cloud computing deal with Oracle to secure 4.5 gigawatts of computing capacity for training large AI models.
- Talent and Hardware: The acquisition of hardware startup io Products (founded by former Apple design chief Jony Ive) in a $6.5 billion deal, with Ive joining as Creative Head to design AI-powered devices.
- Developer Tools: Plans to acquire the AI coding platform Windsurf (formerly Codeium) in a reported $3 billion deal.
