OpenAI has partnered with Broadcom and Taiwan Semiconductor Manufacturing Company (TSMC) to develop its first custom AI chip, while adding AMD chips alongside its existing Nvidia infrastructure. The company examined various options to diversify chip supply and reduce costs, including an initial plan to build its own network of chip manufacturing foundries. That foundry plan got shelved due to extensive costs and time requirements.
The global AI chip market sits at $14.9 billion as of 2022 and will reach $383.7 billion by 2032, according to allied market research. Nvidia holds 80% of current market share. OpenAI CEO Sam Altman explained that by collaborating with industry leaders like Broadcom and TSMC, OpenAI aims to accelerate its AI development and ensure a more resilient supply chain.
A team of 20 engineers now works on the chip project at OpenAI, including Thomas Norrie and Richard Ho, who previously built Tensor Processing Units at Google. Through the Broadcom partnership, OpenAI has locked in manufacturing capacity at TSMC, aiming for first chip production in 2026, though sources note this timeline might shift.
The numbers tell a clear story about why OpenAI needs this move. The company projects $5 billion in losses this year against $3.7 billion in revenue. Computing costs eat up the largest portion of expenses. The market responded immediately to the news – Broadcom shares increased 4.5% while AMD gained 3.7%.
Similar posts
The new chip focuses on inference tasks – where AI makes predictions from trained models. Broadcom provides essential components that help move information quickly between chips, critical for systems where thousands of processors work together. The company continues using Nvidia’s chips while developing alternatives, maintaining careful relationships to ensure access to Nvidia’s new Blackwell chip line.
The partnership mirrors similar moves by Amazon, Meta, Google and Microsoft to control their chip supplies. AMD’s entry through Microsoft’s Azure platform with its MI300X chips adds another player to the field. AMD projects $4.5 billion in AI chip sales for 2024 after launching in late 2023.
Sources indicate OpenAI still weighs whether to develop additional chip design elements internally or seek more partners. The company wants its custom chip to handle the heavy computational demands of training AI models and running services like ChatGPT. This careful balancing act between innovation and practical constraints shows how even the most ambitious tech companies must work within market realities.
