Samsung Electronics has suffered a rare one-two setback to its custom AI silicon ambitions, with Meta and OpenAI both stepping away from chip projects at the company's System LSI division within days of each other. Korean media first reported on June 4, 2026 that Meta had requested a hold on its custom AI chip program with Samsung — coming just days after a separate effort to build an inference processor for OpenAI was said to have "cooled off."
Neither Meta nor Samsung has publicly confirmed the reasons. Meta reportedly framed its move as a "temporary hold" rather than an outright cancellation, but the timing — two marquee AI customers pulling back within the same week — has put a spotlight on Samsung's bid to become a go-to designer and fabricator of bespoke AI chips.
Key Highlights
- Meta requested a pause on its custom AI chip project with Samsung's System LSI division on or around June 4, 2026, halting advanced development work.
- The Meta effort had reportedly been underway since last year and, if it had reached mass production, the long-term contract was expected to be worth tens of trillions of won.
- Days earlier, Samsung's project to build an inference neural processing unit (NPU) for OpenAI was reported to have stalled amid strategic disagreements.
- Samsung has described neither project as definitively dead, but two high-profile customers stepping back within the same week is a visible blow to its custom-silicon strategy.
Details
The custom chip work sat inside Samsung's System LSI division — the arm responsible for logic chip design, including the company's Exynos mobile processors — rather than its memory business. Designing bespoke accelerators for hyperscalers has been a strategic goal for Samsung, which has watched rivals capture the most lucrative slices of the AI hardware boom.
For Meta, the project fit a broader pattern. The company has been aggressively building out its own custom silicon, including its MTIA accelerators, while simultaneously signing huge supply deals with Nvidia and AMD. A pause on the Samsung program does not necessarily signal a retreat from custom chips overall — it may reflect a reshuffling of priorities or a shift toward other manufacturing partners.
The OpenAI situation appears rooted in strategic disagreements rather than technical failure. OpenAI has been exploring multiple paths to secure compute, from custom accelerators to enormous infrastructure commitments, and a cooling NPU partnership is one data point in a fast-moving negotiation landscape.
Impact
The back-to-back retreats are an uncomfortable look for Samsung at a moment when it is trying to prove that its foundry and logic-design businesses can compete for the highest-value AI work. Samsung's second-generation 2nm GAA process (SF2P) is slated for production in 2026, with capacity standing up at its Taylor, Texas fab, and the company recently deepened a multi-year collaboration with Cadence to certify AI-optimized 2nm and 3D-IC design flows.
Winning marquee custom-silicon contracts would have validated that investment. Losing momentum on two at once raises questions about whether Samsung can convert its manufacturing capacity into the design wins that command premium margins.
Background
There is a counterweight to the gloom. Samsung recently joined Anthropic's Series H funding round — the financing that valued the AI lab at roughly $965 billion — as a strategic infrastructure partner. Because Anthropic specifically referenced logic chips in describing the partnership, industry watchers see a potential opening for Samsung's System LSI and foundry units to pick up exactly the kind of custom AI silicon work that the Meta and OpenAI projects represented.
In other words, the customer roster may be reshuffling rather than collapsing. The AI hardware market remains acutely supply-constrained, and chip designers with proven capacity retain leverage even when individual deals stall.
What's Next
The immediate questions are whether Meta's "temporary hold" thaws and whether the OpenAI talks resume on new terms. More broadly, Samsung's ability to land — and keep — custom AI chip business will be a key test of whether its multibillion-dollar foundry buildout pays off in the highest-margin corner of the semiconductor market.
For the MENA region, where Saudi Arabia and the UAE are pouring capital into AI data centers and sovereign compute, the episode is a reminder that custom-silicon supply chains remain volatile. Buyers planning long-horizon infrastructure should assume that today's chip-design partnerships can shift quickly, and build flexibility into procurement accordingly.
Source: SamMobile