OpenAI Officially Announces Development of Custom AI Chips

Details

• CEO Sam Altman revealed that OpenAI is creating its own chips, marking the first time the company publicly confirmed plans for bespoke silicon.
• The announcement was made during Episode 8 of the OpenAI Podcast on October 13, 2025, and distributed via Spotify, Apple, and YouTube.
• Altman positioned this initiative as complementary to existing partnerships with Nvidia, AMD, Microsoft Azure, and other manufacturing partners.
• While technical specifications and a launch timeline remain undisclosed, the effort builds on OpenAI’s experience with training advanced models like GPT-5 and Sora.
• Earlier reports from Reuters referred to this initiative as “Project Tigris,” which emerged after 2023 GPU shortages delayed key model deployments.
• The new accelerators are being designed for both training and inference, with goals to reduce token costs and enhance energy efficiency.
• By controlling both hardware and software, OpenAI believes it can iterate on model architectures faster than when relying solely on off-the-shelf GPUs.

Impact

This move directly challenges Nvidia’s stronghold in AI datacenter hardware, arriving shortly after Google’s TPU v6 release. By developing proprietary chips, OpenAI could achieve cost savings and gain a competitive edge through tighter integration—echoing strategies from Apple and Google. If successful, this approach could not only fine-tune OpenAI’s cost structure but also prevent the supply chain disruptions that hindered previous major model launches.