According to reliable sources, OpenAI, the developer of ChatGPT, is advancing its plan to develop its first self-designed artificial intelligence chip, which is expected to complete its design and be delivered to TSMC for tape-out in the coming months. The project plans to utilize TSMC's advanced 3-nanometer process technology and will be equipped with high-bandwidth memory, with an overall architecture similar to NVIDIA products.
This strategic move began in October 2023, aiming to address the issues of AI chip supply shortages and high costs. Sources indicate that OpenAI views this training-focused processor as an important tool to enhance its negotiation power with other chip suppliers. The company's engineering team plans to continuously optimize the processor's performance in subsequent iterations.
If all goes well, OpenAI is expected to achieve mass production targets by 2026. However, the road to chip development is not smooth: each tape-out can cost tens of millions of dollars, with a standard production cycle of about six months, and the first tape-out may face the risk of failure, requiring additional diagnostics and re-tape-out processes.
It is worth noting that, despite significant investments by tech giants like Microsoft and Meta in self-developed chips, they have not achieved notable breakthroughs. Meanwhile, the market landscape is subtly changing: Chinese AI startup DeepSeek has lowered hardware barriers through algorithm optimization, sparking discussions about the future direction of AI computing demand.
Against this backdrop, the enthusiasm for investment among tech giants remains strong. Microsoft, Amazon, Google, and Meta have announced a total investment of $320 billion in AI technology and data center construction by 2025, further increasing from last year's record spending.
This series of developments suggests that the competitive landscape in the AI chip sector may soon undergo significant changes.
