Samsung Foundry is reportedly in talks over major deals to produce 2nm AI chips for Meta and Anthropic. The discussions matter because they signal that Samsung is strengthening its position in the race to build the most advanced AI chips.
At the same time, new demand is said to be flowing into Samsung’s foundry business. If the talks end in contracts, Samsung would expand from being a chip supplier for select partners into a key manufacturer for two major names in AI infrastructure.
Meta’s shift away from TSMC
Meta is said to have selected Samsung Foundry to manufacture the third-generation Meta Training and Inference Accelerator, or MTIA. According to SEDaily, the production value of the AI chip could exceed KRW 10 trillion using Samsung’s 2nm manufacturing process.
This would mark an important change in Meta’s supply chain. Its first and second MTIA generations were previously made by TSMC.
The reported move to Samsung comes as TSMC’s advanced manufacturing capacity is said to be fully booked. Major customers such as AMD, Apple, MediaTek, Nvidia, and Qualcomm are described as having already secured much of the Taiwanese foundry’s cutting-edge node capacity.
| Company | Reported chip plan | Key detail |
|---|---|---|
| Meta | Third-generation MTIA | May use Samsung’s 2nm process; production value could exceed KRW 10 trillion |
| Anthropic | 2nm AI chips | Exploring Samsung Foundry for custom AI hardware |
Meta’s own AI hardware strategy is designed to reduce dependence on third-party accelerators. The company is trying to rely less on suppliers such as AMD and Nvidia while speeding up its internal hardware development.
It is also targeting AI data center construction with a combined capacity of 5GW by 2030. To support that pace, Meta reportedly wants to release a new generation of AI chips every six months.
That aggressive cycle requires a manufacturing partner that can move quickly. During chip development, Meta is said to be working closely with Samsung’s System LSI division on design matters.
Anthropic’s search for a more integrated supply chain
Anthropic, the company behind Claude, is also reportedly exploring Samsung Foundry’s 2nm process. Like Meta, it wants to reduce dependence on external AI chip suppliers.
The company is said to be preparing an AI data center with a total capacity of 1GW, with investment estimated at $50 billion.
About half of that investment is reportedly earmarked for hardware. The spending would cover custom AI chips or ASICs, DRAM, and NAND flash memory.
That is where Samsung stands out. It is one of the few companies able to manufacture advanced chips while also supplying memory products such as DRAM and NAND flash.
Because of that, Anthropic is reportedly interested in building a strategic partnership with the South Korean company. Such a setup could create a more integrated supply chain for large-scale AI infrastructure.
Samsung is also said to have invested $65 billion in Anthropic during its Series H funding round in May 2026. That financial link adds context to the two companies’ recent discussions about chip manufacturing.
Fresh orders are changing Samsung Foundry’s outlook
The talks with Meta and Anthropic come as Samsung Foundry is said to be moving back toward profitability. New orders are one of the clearest signs of that recovery.
Samsung had earlier secured a $16.5 billion contract from Tesla. That deal covers production of Tesla’s AI6 chip using Samsung’s 2nm process.
The company also manufactures the latest LPU for Groq, which is described as Nvidia’s strategic technology partner. The growing order book shows that Samsung is gaining traction in the next wave of AI hardware development.
Beyond that, Samsung is reported to have received chip manufacturing orders from Chinese automaker BYD. The company is also said to be in line to produce part of Google’s next-generation TPU on a 2nm node.
Taken together, these developments point in the same direction. Samsung is not only selling factory capacity, but also positioning its 2nm process as a core offer for large AI customers.
If the talks with Meta and Anthropic move into mass production, the impact could go beyond a pair of major contracts. It would reinforce the idea that advanced AI chip capacity is no longer concentrated around a single player, especially as demand keeps rising and major customers look for manufacturers that can supply both chips and memory components.