OpenAI is actively pursuing the development of custom AI chips to address the GPU shortage and reduce its reliance on Nvidia hardware. The company has partnered with Broadcom to design an inference chip, expected to be produced by TSMC and released by 20261. This move aligns OpenAI with other tech giants like Amazon, Google, and Microsoft, who have also developed custom AI processors1.

To support this initiative, OpenAI has assembled a team of around 20 engineers, including experts from Google's Tensor Processing Unit (TPU) project1. The company is also hiring for hardware/software co-design roles to work with vendors on future AI accelerators2. While OpenAI initially considered building its own fabs, it has since abandoned this plan due to high costs and long timelines1. Instead, the focus remains on chip design and collaboration with established manufacturers to meet the growing demand for AI computation.

OpenAI's rollout of GPT-4.5 has been significantly impacted by GPU shortages, forcing the company to limit access to its most expensive subscription tiers. The new model, described by CEO Sam Altman as "giant" and "expensive," requires substantial computational resources, with OpenAI charging $75 per million tokens for input and $150 per million tokens for output - a 30x and 15x increase respectively compared to their previous GPT-4 model1. To address the shortage, OpenAI plans to add tens of thousands of GPUs next week, gradually expanding access to ChatGPT Plus customers and other tiers2.

The GPU shortage highlights broader challenges in the AI industry, where demand for high-performance chips far outpaces supply. This scarcity has led to increased costs and limited accessibility of cutting-edge AI technologies, potentially stifling innovation and concentrating AI benefits among wealthy corporations3. In response, some companies are exploring alternative solutions, such as developing their own AI chips or optimizing existing hardware to reduce reliance on expensive GPUs43.

TSMC, a leading semiconductor manufacturer, has emerged as a key player in the AI chip manufacturing landscape. The company has formed strategic partnerships with major AI companies to address the growing demand for advanced AI hardware. Notably, OpenAI is reportedly collaborating with TSMC to develop its first generation of in-house AI chips, aiming to reduce dependency on Nvidia and finalize the design in the coming months12. This partnership leverages TSMC's cutting-edge manufacturing processes, including 5nm and 3nm nodes, which are crucial for producing high-performance AI chips3.

• TSMC has introduced innovative technologies like the TSMC A16™ process and System-on-Wafer (TSMC-SoW™) to enhance AI chip performance4.

• The company's Chip on Wafer on Substrate (CoWoS) technology has been instrumental in enabling the AI revolution by allowing more processor cores and high-bandwidth memory stacks on a single interposer4.

• TSMC's collaborations extend beyond OpenAI, with partnerships involving other tech giants and AI-focused companies, positioning it as a central figure in the evolving AI hardware ecosystem32.