Samsung’s troubled foundry business has received a welcome vote of confidence from South Korean chip designer DeepX, which is spending more than $100 million to develop a new AI processor using Samsung’s 2nm manufacturing process.
DeepX has completed the design of its DX-M2 chip and handed the finished plans to Samsung’s foundry division, moving the project towards manufacturing and testing.
According to The Korea Herald, the investment covers chip development, performance verification, and production engineering, making it a substantial commitment for a company trying to establish itself in the crowded AI accelerator market.
The DX-M2 is intended to run generative AI models directly on devices, with robots, industrial equipment, and AI-enabled PCs among its intended applications. DeepX claims the processor will handle generative AI workloads while consuming approximately 5W, an attractive proposition for manufacturers whose products cannot accommodate the electricity bills or cooling requirements of conventional AI hardware.
The company expects to receive its first silicon in March 2027, with mass production scheduled for 2028. For Samsung, the deal offers another chance to show that its advanced manufacturing technology can attract customers beyond its own semiconductor operations.
The Korean giant has spent years attempting to loosen TSMC’s grip on the most lucrative contract chipmaking business, with results that have occasionally tested the patience of its customers. Its early 3nm manufacturing efforts struggled with yields, leaving Samsung fighting to convince chip designers that its newer processes could deliver competitive performance without producing expensive quantities of defective silicon.
Samsung’s second-generation 2nm process, known as SF2P, is central to its attempts to repair that reputation. DeepX had previously selected the process for its DX-M2 processor. The latest announcement moves the project forward, confirming the substantial development commitment and completion of the chip’s design.
The company is working with Korean semiconductor design specialist Gaon Chips on the processor. For DeepX, Samsung’s advanced process is attractive because it could enable a powerful AI accelerator with low enough power consumption for applications outside traditional data centres.
Running AI models locally can reduce delays from sending information to remote servers, particularly in industrial systems where a machine might need to make decisions without waiting for a distant computer to respond.
It can reduce dependence on cloud infrastructure, although the actual advantages will depend on the models being run and the performance of the finished hardware. DeepX’s published specifications describe the DX-M2 as a 5W system-on-chip supporting up to 96GB of external memory, with memory bandwidth reaching 153.6GB/s.
Those figures remain design specifications until production silicon shows the processor performs as advertised. The company is targeting applications involving large language models and vision-language models, which combine visual information with language processing.
Such technology could be useful in industrial robots that need to interpret their surroundings or automated equipment that must analyse information without maintaining a permanent connection to a data centre. Samsung will be particularly interested in whether DeepX can translate its design into commercially successful products.
The foundry business needs more customers willing to commit to its advanced processes, especially as TSMC continues attracting major chip designers. DeepX’s investment is encouraging, but the real test begins when Samsung produces working chips and the company starts measuring performance, power consumption, and manufacturing yields.
DeepX expects the first samples in March 2027, giving Samsung another opportunity to prove that its latest manufacturing technology works outside a presentation.






