Recent reports indicate that the custom AI chip pre-development project between Samsung Electronics and Meta has been put on hold.
This news sends a significant signal about the challenges facing Samsung's System LSI division in securing major AI clients. This isn't an isolated event; it follows a similar pattern to the recently reported pause in collaboration with OpenAI. The core of this issue lies in a strategic shift by Big Tech companies towards developing their own custom chips and solidifying their supply chains with established players.
Let's look at Meta's recent moves to understand the context. First, Meta has been aggressively pursuing its own AI accelerator roadmap, called MTIA (Meta Training and Inference Accelerator). They are on a rapid 6-month release cycle, demonstrating a clear commitment to internalizing their AI hardware. Second, Meta has solidified its partnership with Broadcom and TSMC, signing a long-term deal that extends to 2029. This strengthens an already proven supply chain, reducing the need to bring in a new partner like Samsung for a high-stakes custom SoC project. Finally, Meta's increasing capital expenditure on AI infrastructure naturally leads them to prioritize partners who can deliver with certainty.
For Samsung, this represents a potential loss of a multi-billion dollar revenue opportunity and a setback to its ambitions in the custom AI chip market. The System LSI division was created to capture exactly this kind of large-scale business from tech giants.
However, the story isn't entirely one-sided. While the logic chip business faces headwinds, Samsung's memory division is experiencing powerful tailwinds. The company recently announced the world's first shipment of HBM4E 12-layer samples, a next-generation memory chip crucial for AI. This positive news has largely offset the logic chip concerns in the stock market. Furthermore, a new potential opportunity has emerged with AI company Anthropic, which named Samsung a 'strategic infrastructure partner' in its latest funding round. This could open a new door for Samsung's foundry business, though it's still too early to count on it as a confirmed order.
Ultimately, Samsung is currently navigating a dual narrative: uncertainty in its logic chip ambitions clashing with strong leadership and momentum in its advanced memory business.
- Custom SoC (System on Chip): A custom-designed integrated circuit that combines all major components of a computer or other electronic system into a single chip. Big Tech companies design these to optimize performance for their specific AI workloads.
- Foundry: A semiconductor manufacturing plant that makes chips for other companies. Samsung operates one of the world's leading foundries, competing with TSMC.
- HBM4E (High Bandwidth Memory 4 Extended): The latest generation of high-performance stacked memory used in GPUs and AI accelerators, offering significantly higher speed and bandwidth than previous versions.