From Tesla, Meta to Anthropic: Samsung's chip orders reach 50 trillion won

Author: Bao Yilong; Source: Wall Street Insight

Samsung Electronics is rapidly establishing its core position in the AI semiconductor market through its foundry business.

On July 3, according to South Korean media reports, Samsung's foundry division has amassed a long-term order backlog approaching 50 trillion Korean won. After securing Tesla's AI chip order last year, two global tech giants—Meta and Anthropic—have also shifted their ASIC production needs to Samsung.

An industry insider stated:

Since Samsung's foundry won Tesla's AI chip order last year, semiconductor foundry orders for AI servers have entered a phase of full acceleration.

Driven by this, market expectations are that Samsung's foundry business will break even as early as the fourth quarter of this year. The key support for this round of order expansion lies in Samsung's most advanced 2-nanometer process technology.

Meta's third-generation AI accelerator chip, MTIA 3, and Anthropic's custom ASIC are both planned to be produced using this process. External inquiries about Samsung's 2nm process have surged sharply, further solidifying its competitive position in the advanced process foundry market.

Meta Shifts Orders to Samsung, Betting on 2nm Mass Production

According to reports citing industry insiders, Meta is in talks with Samsung's foundry division for next-generation ASIC design and production cooperation valued at over 10 trillion Korean won.

Meta's self-developed AI accelerator MTIA, with its first two generations produced by TSMC, has locked in Samsung as its core manufacturing partner starting from the third generation released this year.

It is reported that the MTIA 3 will be manufactured using Samsung's most advanced 2nm process, with production volumes reaching tens of thousands of wafers. Samsung Electronics stated, "No decision has been made yet."

Behind Meta's shift to Samsung is its strategic need for large-scale self-built AI infrastructure.

Meta is studying a cloud service business that rents AI computing power to external companies, with MTIA serving as the core chip to support this business.

At the same time, Meta has set a goal of building data centers totaling 5 gigawatts by 2030, meaning simply relying on external chip supply is unsustainable.

To this end, Meta has launched a high-speed development cycle in which a new chip generation is iterated every six months, planning to complete consecutive releases from the third to the fifth generation by next year.

To support this ultra-fast development cycle, Meta has also established a joint design mechanism with Samsung's System LSI division. According to reports, the collaboration begins at the early stage of chip architecture design to compensate for the capacity gap in Meta's own engineering team under such a compressed timeline.

Anthropic Internalizes AI Infrastructure, Samsung May Be the Biggest Beneficiary

U.S. AI company Anthropic is also reportedly evaluating the development of custom ASICs using Samsung's 2nm process. This move is seen as part of its strategy to reduce reliance on Nvidia GPUs and Google TPUs, advancing its "AI infrastructure autonomy" strategy.

In terms of investment scale, Anthropic plans to build approximately 1 gigawatt of AI data centers in the long term, with total related investment estimated at about $50 billion (approximately 77 trillion Korean won).

Industry analysis suggests that about half of this amount will flow into AI semiconductor procurement, with semiconductor investments covering ASICs, DRAM, and NAND flash estimated at around $25 billion (approximately 39 trillion Korean won).

Samsung Electronics, with its integrated semiconductor capabilities in memory, foundry, and advanced packaging, is widely regarded as the biggest potential beneficiary of Anthropic's upcoming AI chip procurement.

In May of this year, Samsung participated in Anthropic's $65 billion (approximately 100 trillion Korean won) Series H financing round, thereby establishing a strategic partnership between the two companies.