OpenAI Launches Its First AI Chip to Reduce Dependency on Nvidia

OpenAI is developing its first internal AI chip to reduce its reliance on Nvidia and accelerate the diversification of its chip supply. This chip, focused on training tasks, is expected to complete its design in the next few months and will be manufactured by TSMC.

The process, known as "tape-out", can cost tens of millions of dollars, and if the first tape-out fails, further troubleshooting and re-tape-out will be required.

OpenAI aims to use this project to enhance its negotiating power with chip suppliers and lay the foundation for future chip development.

According to sources, OpenAI plans to begin mass production at TSMC in 2026, but the tape-out timeline may vary depending on the manufacturing process.

Developing its own chip not only helps reduce reliance on major suppliers like Nvidia, but also provides more customization in AI model training and inference tasks.

OpenAI's partnership with Broadcom and TSMC aims to create a chip focused on AI inference tasks, which will support OpenAI's large-scale AI systems.

Although there is currently a greater demand for chips in training tasks, analysts predict that demand for inference chips will gradually increase as more AI applications are deployed.

Broadcom will assist OpenAI in refining the chip design and providing key components to ensure efficient data transmission between chips, which is critical for OpenAI's AI systems.

OpenAI had originally planned to establish its own chip manufacturing network to achieve full chip production independence, but this plan has been shelved due to the costs and time required to establish such a network.

Instead, OpenAI is focusing on internal chip design and collaborating with industry partners like Broadcom and TSMC to share the burden of manufacturing and R&D. This strategy reduces investment pressure and ensures a strong position in chip supply.

Additionally, OpenAI's partnership with AMD is progressing steadily. The company plans to use AMD's new MI300X chip through Microsoft's Azure service to compete with Nvidia.

As the AI market grows, OpenAI and other companies have realized the risks of relying solely on suppliers like Nvidia and are exploring alternative chip suppliers to diversify risk. AMD's MI300X chip is expected to be used in 2024 and aims to capture a portion of the AI chip market.

Currently, Nvidia's GPUs dominate the AI market, but due to chip shortages and rising prices, major companies like Microsoft, Meta, and OpenAI are searching for alternative solutions.

While OpenAI works to reduce dependence on Nvidia, it is also focusing on the costs of computing power and hardware, which are its largest operational expenses. To control costs, OpenAI is optimizing resource utilization and strengthening collaboration with other chip manufacturers.

OpenAI expects a $5 billion loss in 2023, but plans to reduce hardware and computing costs by diversifying its chip suppliers. This strategy will not only benefit OpenAI's long-term development but could also have a profound impact on the AI industry's chip supply chain.

As AI applications become more widespread, the demand for chips is expected to grow exponentially, and OpenAI's diversified supply strategy will enhance its competitive edge in the market.

As OpenAI's chip development progresses, the AI chip market is likely to experience new competition and technological breakthroughs in the coming years.

OpenAI's internal chips will not only drive innovation in its products and services but may also lead to technological advancements that change the current chip market landscape.