Meta is reportedly in talks for a $6.5 billion Samsung AI chip deal as it reshapes its in-house silicon roadmap

Meta Platforms is reportedly in discussions with Samsung Electronics to develop and manufacture its next generation of in-house AI accelerator chips in a deal valued at more than $6.5 billion, according to multiple market reports published July 3.

The reported figure, which is framed as a multi-year arrangement, centers on Samsung Foundry producing chips based on Meta’s “MTIA” accelerator lineup. MTIA stands for Meta’s in-house AI accelerator chips, designed to run machine-learning workloads more efficiently and potentially at lower cost than relying solely on third-party processors.

Benzinga reported that Samsung Foundry would be responsible for production of Meta’s proprietary AI accelerator “MTIA,” starting with the third generation. It also said the third-generation MTIA would be mass-produced using Samsung Foundry’s advanced 2-nanometer process, citing the Seoul Economic Daily. The same report said Meta, Samsung, and Anthropic did not immediately respond to requests for comment from Benzinga.

In a separate repost on TradingView, citing Sedaily, the chip talk is described as a shift in Meta’s manufacturing partners. The article said TSMC produced the first two generations of Meta’s in-house AI accelerators, but Meta’s growing computing needs appear to be pushing it toward a wider group of chip partners.

The discussion also ties to how fast Meta wants its custom silicon to refresh. TradingView said Meta is targeting the introduction of a new AI chip every six months, and that third- and fifth-generation products are targeted for next year. That cadence, if it holds, would increase pressure on packaging, yields, and capacity across manufacturing partners.

On scale, the TradingView report also referenced Meta’s infrastructure planning, including a target of 5 gigawatts of data center capacity by 2030. Custom accelerators are a key part of that approach because AI training and inference demand large data center power footprints and significant spend on hardware, and Meta has been building its own silicon to manage performance and cost.

For Samsung’s chipmaking business, winning a large Meta design and manufacturing relationship would represent a high-profile customer win in the AI accelerator supply chain. Samsung’s foundry unit has been pursuing more share in advanced-node manufacturing, and market participants often view major customer announcements as indicates about confidence in technical and operational execution.

The reports stopped short of announcing a final contract. As of publication, there was no detailed confirmation from Meta or Samsung about pricing, contract timing, or whether Meta’s AI chip roadmap will rely on Samsung for all future generations or only specific batches. Investors and infrastructure buyers are likely to watch for any company statements, chip tape-outs tied to 2-nanometer timelines, and updates on whether the MTIA manufacturing split between TSMC and Samsung becomes permanent.