March 26, 2026 /SemiMedia/ — Samsung Electronics has secured an exclusive deal to supply its next-generation HBM4 memory to OpenAI, for use in the company’s first in-house AI chip known as Titan, according to industry sources.

The supply is expected to begin in the second half of the year, with volumes reaching up to 800 million gigabits of 12-layer HBM4. This represents a notable share of Samsung’s planned HBM output, with a higher allocation within its HBM4 product line.

The HBM4 will be placed close to the Titan processor to support high bandwidth and efficiency needed for large AI models. Titan is being developed by OpenAI in partnership with Broadcom, and is expected to enter mass production at TSMC in the third quarter, with a launch planned by year-end.

The move highlights a broader shift by OpenAI from relying on general-purpose GPUs to building its own chips. As AI workloads move more toward inference, custom hardware is becoming more important for performance and cost control.

Industry sources said Samsung was able to meet OpenAI’s requirements for its latest HBM generation, helping it win the deal. There are also expectations that Samsung could remain a key memory supplier for future versions of the Titan platform.