Meta weighs shift in AI chip manufacturing partners, according to a report

Meta Platforms is reportedly considering changes to how it manufactures key AI chips, with at least part of its next-generation accelerator production potentially shifting away from Taiwan Semiconductor Manufacturing Co. and toward Samsung’s foundry operations.

According to a market report carried by TradingView, Meta is in discussions with Samsung Electronics’ foundry unit and is exploring a deal that could be worth more than 10 trillion Korean won (about $6.5 billion), tied to production of Meta’s in-house AI accelerator family known as MTIA.

The same report said the contemplated arrangement would use Samsung’s 2-nanometer (2nm) process technology, and that the effort would cover future MTIA generations. It specifically cited the possibility of MTIA 450 and MTIA 500 as the products that could be built using Samsung’s 2nm manufacturing approach.

Meta’s first two MTIA generations were described as having been manufactured by TSMC, which would make the reported direction change notable for investors tracking how aggressively Meta is building its own AI infrastructure at scale. The MTIA line is Meta’s custom silicon designed to accelerate AI workloads used across its services and data-center buildout.

Why the change could matter, the reports suggest, is timing and capacity. TradingKey’s analysis attributed the shift largely to supply constraints and competition for advanced chip capacity, stating TSMC’s production is tight through 2027 and that Meta would rather not delay mass production while it cycles through new chip generations on a roughly six-month cadence.

Beyond supply, the analysis also connected Meta’s chip strategy to a broader push to scale “Meta Compute,” described as an infrastructure effort intended to monetize surplus AI computing capacity. The underlying theme is that the faster Meta can secure reliable manufacturing for its custom accelerators, the more quickly it can expand its data-center and service mix around AI.

Meta has not, in the material reviewed for this story, confirmed the specific terms of any Samsung arrangement or any change to its foundry strategy, and the cited reporting appears to be based on industry accounts rather than a direct company announcement.

Still, the reports highlight a live tension in the AI hardware market: hyperscalers are racing to lock in advanced nodes and production slots for custom chips, while foundries compete for limited wafer capacity at the leading edge. For Meta, any move that improves throughput or reduces delivery risk for MTIA accelerators could ripple into its broader AI infrastructure timetable.