Anthropic is looking to diversify its chip supply chain to reduce its heavy reliance on market leader Nvidia. In a fresh development, the Claude-maker has initiated talks with Samsung seeking an AI chip partnership with the South Korean giant. For now, these discussions are in preliminary stages, sources familiar with the metter has disclosed to the media.
Samsung is an established name in the semiconductor business. It is touted among the top leaders in memory chips that power smartphones, computers, and AI servers. Additionally, the company also operates a foundry business to manufacture custom chips in its advanced factories. Research firm Counterpoint Research estimates that Sansung holds around seven percent of the foundry market.
For now, Anthropic is still in the process of chalking out the exact AI chip specifications they would need for Samsung to design and manufacture. Other requirements like exact performance requirements are also under deliberations, reports say.
Anthropic is also yet to finalize if the chip it wishes Samsung to work on would be used for training AI models or running them after deployment.
As of now, no final manufacturing agreement has been mapped out between the two entities.
A report published by The Information said Anthropic still plans to keep its active chip contracts with Amazon, Google, and Nvidia entact. The AI company procures Trainium processors from Amazon, tensor processors from Google, and graphics processors from Nvidia to power its computing strategy.
Earlier in May, Anthropic had opened dialogue with Microsoft looking to secure the software giant’s native Maia 200 chip which can power AI models more cost-effectively.
If Anthropic does cut these deals with Microsoft and Samsung, the two will gain major momentum in their chip businesses.
The tech industry has constantly been observing that the demand for AI chips within the U.S. itself, is far higher than the chip supply industry worth over $13 trillion. This has resulted in frontier labs starting to tap every viable chip stack they can get.
OpenAI also made a big move in breaking free from the Nvidia dependency last month. On June 24, the ChatGPT-maker launched its first “intelligence processor” called “Jalapeño” — built to optimize data flow and resource balance for near-peak performance. The initial testing of the chip showed its per watt substantially has shown to be better than the current state-of-the-art system.
