July 21, 2024
AI chips

OpenAI Considers Development of AI Chips Amidst Chip Shortage

OpenAI, a highly funded AI startup, is reportedly exploring the possibility of developing its own AI chips. The discussions regarding AI chip strategies have been ongoing within the company since last year due to the worsening shortage of chips required to train AI models. OpenAI is considering several strategies to fulfill its chip ambitions, including acquiring an AI chip manufacturer or undertaking internal chip design efforts.

According to CEO Sam Altman, acquiring more AI chips is a top priority for OpenAI. Currently, OpenAI relies on GPU-based hardware, like its competitors, to develop models such as ChatGPT, GPT-4, and DALL-E 3. GPUs are well-suited for training today’s most advanced AI systems due to their capability to perform parallel computations. However, the surge in generative AI technology has placed significant strain on the GPU supply chain, leading to shortages in the market. Microsoft has warned that it may face service disruptions due to a severe shortage of server hardware required for AI, while Nvidia’s high-performing AI chips are reported to be sold out until 2024.

OpenAI heavily relies on GPUs to run and serve its models, using clusters of GPUs in the cloud to handle customer workloads. However, this comes at a considerable cost. An analysis by Bernstein analyst Stacy Rasgon estimated that running ChatGPT queries at one-tenth the scale of Google Search would initially require approximately $48.1 billion worth of GPUs, with an annual cost of around $16 billion to remain operational.

Although OpenAI would not be the first company to pursue the development of its own AI chips, with Google, Amazon, and Microsoft already venturing into this area, it is in a strong position to invest in research and development. Having raised over $11 billion in venture capital, the company is approaching $1 billion in annual revenue. This has led to considerations of a share sale that could potentially value the company at $90 billion, according to the Wall Street Journal.

However, developing AI chips is a challenging endeavor. Last year, AI chipmaker Graphcore faced substantial setbacks, including a significant valuation cut after a deal with Microsoft fell through. As a result, the company announced job cuts due to the difficult macroeconomic environment. Similarly, Habana Labs, an AI chip company owned by Intel, laid off approximately 10% of its workforce. Meta’s custom AI chip efforts also faced issues, leading to the abandonment of some experimental hardware.

Even if OpenAI decides to pursue the development of custom chips, it would be a time-consuming and costly process. Such an endeavor could take years and cost hundreds of millions of dollars annually. It remains to be seen if OpenAI’s investors, including Microsoft, are willing to take on the risks associated with this venture in the unforgiving hardware industry.

1. Source: Coherent Market Insights, Public sources, Desk research
2. We have leveraged AI tools to mine information and compile it