DEEPX Commits Over $100 Million to Samsung-Made DX-M2 AI Chip
DEEPX will spend over $100 million on its DX-M2 AI processor, with Samsung's foundry building the 2nm chip. First silicon lands in March 2027; mass production is targeted for 2028.
By Amara Osei
3 min read

What's News
- DEEPX commits over $100 million to develop the DX-M2 AI processor
- Samsung's foundry will fabricate the chip on a 2-nanometer process
- First silicon expected March 2027; mass production targeted for 2028
- The DX-M2 is a 5-watt SoP supporting 96GB external memory and 153.6GB/s bandwidth
- The chip won a CES 2027 Innovation Award in the AI category
South Korean fabless chipmaker DEEPX will spend more than $100 million developing its next-generation DX-M2 artificial intelligence processor, with Samsung Electronics' foundry business manufacturing the part on a 2-nanometer process. The startup has completed the chip's design and handed it to Samsung for fabrication, according to a Yonhap report published by The Korea Herald, citing a company statement.
The reported budget covers design verification and production engineering alongside development work. It's a sizable bet for a startup whose product has not yet reached first silicon.
What is the timeline?
DEEPX expects to receive its first silicon in March 2027. From there, the company plans to begin evaluation and preparations for mass production. Full-scale production is targeted for 2028, according to a separate Yonhap report.
The initial chips serve a specific purpose: DEEPX will test whether the manufactured processor matches its design targets before committing to higher-volume production. That qualification stage typically includes checks of function, power use and reliability. Neither cited report discloses a customer order tied to the 2028 production target — a gap investors will watch closely.
What does the DX-M2 actually do?
The processor targets on-device generative AI and physical AI systems, including robotics and smart mobility applications. DEEPX's official product page describes it as a 5-watt system-on-chip supporting:
- up to 96 gigabytes of external memory
- 153.6 gigabytes per second of memory bandwidth
- M.2 and PCIe module formats
The company notes those specifications remain subject to change.
Keeping AI inference on the device, rather than in the cloud, can cut reliance on connectivity and let systems process data closer to where it's generated. DEEPX positions the DX-M2 for workloads that demand local response times and power efficiency. Performance and energy claims will still need independent assessment once production silicon exists — nothing shipping today confirms them.
Has the chip won recognition yet?
Yes, on paper. The DX-M2 received a CES 2027 Innovation Award in the artificial intelligence category, Yonhap reported. But the award was announced before the first silicon milestone, so the commercial timeline still depends entirely on successful fabrication, testing and qualification. An award doesn't manufacture wafers.
How does the deal fit Korea's chip strategy?
The project links DEEPX to Samsung's advanced foundry capacity and to South Korea's broader push into edge and physical AI hardware. A domestic fabless startup taping out on a domestic 2nm line is exactly the kind of pairing Seoul's semiconductor policy aims to encourage.
DEEPX has also been building its distribution footprint across Asia-Pacific. The company previously signed an agreement with global technology distributor Avnet covering South Korea, Japan, Australia, Singapore, Taiwan and Southeast Asia. The DX-M2 program adds the domestic advanced-manufacturing component through Samsung's foundry process while the company prepares for evaluation in 2027.
The next hard checkpoint is March 2027, when first silicon arrives and DEEPX finds out whether a $100 million design works as drawn.
Source: technode.global
More from Amara Osei
Show full bio
Senior reporter covering companies, deals and strategy in entrepreneurship, startups and venture funding at Business Bearings.
2,136 articles