Report: Samsung Locks Exclusive HBM4 Supply Deal With OpenAI to Power Its First AI Chip

The high demand for high-bandwidth memory across the AI industry has been a headache for the hardware market in general. However, it appears to be a favorable situation for Samsung. According to a new report emerging from South Korea, Samsung Electronics has secured a deal to exclusively supply next-generation HBM4 memory to OpenAI.

Samsung’s HBM4 memory to reportedly power OpenAI’s in-house ‘Titan’ chip

The details come from The Korean Economic Daily, which reports that Samsung will start supplying its 12-layer HBM4 memory to OpenAI in the second half of 2026 (via Reuters). This memory is expected to power OpenAI’s first in-house AI chip, reportedly called Titan.

The report further notes that the said chip is being developed in collaboration with Broadcom. TSMC is likely to handle production starting in the third quarter of this year, with a launch expected by the end of 2026.

Samsung’s involvement in the supply of HBM4 memory appears substantial. The company is reportedly allocating more than 5.5 billion GB of HBM4 to OpenAI, which could make it one of its largest customers after NVIDIA and AMD. If the details are accurate, this would mark a significant turnaround for Samsung, especially after earlier struggles with HBM3 and HBM3E validation.

To catch you up, OpenAI has long relied on NVIDIA so far. But developing its own chip could help OpenAI improve inference performance and reduce dependency on general-purpose GPUs. HBM4, with its stacked memory design, is expected to play a key role here by offering faster data transfer and higher efficiency for large AI models.

All that said, the deal remains unconfirmed at this point, and details are still limited. Let’s not forget that Samsung has also signed a deal with AMD to develop next-generation memory solutions, designed for AI apps.