Meta weighs a $6.5 billion Samsung foundry deal for next-generation AI chips, report says

Meta Platforms is reportedly discussing a large foundry agreement with Samsung Electronics that could involve about $6.5 billion in orders for next-generation AI accelerator chips. The talks, according to the report, center on Samsung’s newer 2-nanometer fabrication process, which would represent a meaningful shift from Meta’s recent reliance on Taiwan Semiconductor Manufacturing Co. for earlier generations of its in-house AI chips.

The chips at issue are part of Meta’s MTIA line. MTIA, which stands for Meta Training and Inference Accelerator, is the company’s internal family of specialized processors built to train machine-learning models and run inference, meaning the computations needed to make those models respond to user queries and other workloads at scale. Because Meta uses these accelerators across its data centers, production capacity and manufacturing yield at leading-edge nodes are strategic bottlenecks.

The specific value cited in the report is $6.5 billion, with the expectation that the new chips would use Samsung’s 2-nanometer process technology. If the arrangement goes forward, it would increase Samsung’s role in supplying chips for Meta’s AI infrastructure and could help Meta reduce supply concentration risk as it ramps up compute demand.

The report’s broader framing ties the potential chip shift to Meta’s push to expand cloud and AI infrastructure. While Meta has disclosed major AI investment plans publicly, the foundry details of where chips are manufactured and how far in advance production is contracted are generally not fully spelled out in those disclosures, making third-party reporting a key source for this particular angle.

Other recent coverage surfaced alongside the Yahoo Finance item, echoing the same core claims: Meta is in talks with Samsung for a large MTIA-related order, and Samsung’s 2-nanometer node is the manufacturing target. Some of that related reporting also points to the idea that the deal would be a deliberate supply-chain step as Meta modernizes its AI chip roadmap.

Still, a key uncertainty remains. The reporting describes discussions and potential agreement terms, not a finalized contract. In the absence of a confirmed filing, official procurement statement, or detailed customer disclosure from either Meta or Samsung, it is not possible to verify the exact number of wafers, the delivery schedule, pricing terms, or how much of Meta’s overall AI accelerator demand the Samsung-produced chips would cover.

Meta also did not provide additional detail in the publicly linked post beyond the discussion reported in the market coverage. For investors and industry watchers, the next tell will likely be any confirmation through Meta’s investor communications, procurement-related disclosures, or additional reporting that specifies whether Samsung would be an added supplier, a partial replacement, or a back-up source for certain MTIA generations. The outcome matters because AI data center expansion depends heavily on chip availability, and changes in leading-edge manufacturing partners can affect both timeline and cost assumptions.