DEEPX commits $100 million to 2nm edge AI chips with Samsung
South Korean startup DEEPX is spending over $100 million to build 2nm edge AI processors with Samsung, expanding hardware options for businesses in Asia.
On November 17, 2026, South Korean fabless semiconductor company DEEPX announced a project budget exceeding $100 million for its next-generation DX-M2 artificial intelligence processor. The company completed the chip design and transferred it to Samsung Electronics for manufacturing on a 2-nanometer process. DEEPX also expanded its distribution footprint across the Asia-Pacific region through an agreement with Avnet. This development matters to hardware operators and technology leads because it signals a maturing regional supply chain for advanced edge computing. Businesses building physical systems will soon have more local options to process data directly on their devices instead of paying for cloud inference.
Moving artificial intelligence to the edge
The DX-M2 processor is designed for on-device generative artificial intelligence and physical systems like robotics and smart mobility. According to TechNode Global (opens in a new tab), the processor is a 5-watt system-on-chip that supports up to 96 gigabytes of external memory and 153.6 gigabytes per second of memory bandwidth. The company lists standard M.2 and PCIe module formats for the device, though it notes that these specifications remain subject to change.
Keeping inference workloads on a local device reduces reliance on constant cloud connectivity. Systems can process data exactly where it is generated. This approach is necessary for workloads that require immediate local response times and high power efficiency. When a machine on a factory floor or a delivery vehicle needs to make a routing decision, waiting for a cloud server to process the data introduces unacceptable delays. Local processors solve this latency problem.
For a mid-sized manufacturing or logistics business, the ability to run complex models locally changes how automated systems are deployed. Facilities in remote areas or warehouses with poor network coverage can maintain full operational capacity without a stable internet connection. The focus on physical systems indicates a shift away from text-based assistants toward machines that interact with the real world in real time.
The cost of advanced manufacturing
Developing silicon on a 2-nanometer node requires significant capital. The reported project budget covers design verification, production engineering, and ongoing development work. By partnering with the Samsung foundry business, DEEPX is securing advanced manufacturing capacity within South Korea. This localizes the production of critical hardware components that have historically been concentrated in a few global hubs.
For the broader technology market, a startup committing over $100 million to a single chip design illustrates the high stakes of the hardware industry. The investment reflects a calculation that demand for local processing will grow enough to justify the massive upfront costs of modern semiconductor fabrication. The 2-nanometer process allows for higher transistor density. This translates directly to better energy efficiency and higher performance within the strict thermal limits of a small device.
Building hardware at this scale also requires a robust testing phase. The initial chips will allow DEEPX to test whether the manufactured processor matches its design targets before committing to higher-volume production. This step is critical for ensuring that the final product can handle the rigorous demands of industrial and commercial environments.
What it means for your business
For a company of 50 to 500 people building hardware products in Asia, this announcement changes the supply chain landscape. DEEPX signed a regional distribution agreement with Avnet that covers South Korea, Japan, Australia, Singapore, Taiwan, and Southeast Asia. This means hardware startups and system integrators in these markets will have established, localized channels to source advanced processors. Working through a major distributor simplifies procurement, warranty support, and inventory management for a smaller business that cannot negotiate direct purchasing agreements with semiconductor manufacturers.
The choice of standard M.2 and PCIe module formats is particularly useful for mid-sized engineering teams. These are standard interfaces used in standard computers and industrial motherboards. A technology lead does not need to design a custom circuit board to test this new processor. They can plug the module into existing industrial equipment or prototype boards to evaluate its performance. This lowers the barrier to entry for companies wanting to add local inference capabilities to their existing product lines.
Furthermore, moving processing to the edge changes the cost structure of software products. Instead of paying recurring monthly fees to cloud providers for every inference request, a business pays a one-time capital expense for the local hardware. This shifts the financial model from variable operating expenses to fixed capital costs. A finance lead can often forecast and depreciate these capital costs more easily over the lifespan of the equipment.
What to watch
The commercial timeline depends entirely on successful fabrication and testing. DEEPX expects to receive its first silicon in March 2027. Watch for the results of this initial delivery. The first chips will allow the company to test whether the manufactured processor actually matches its design targets.
This qualification stage includes rigorous checks of function, power use, and reliability. The 5-watt power specification is a target. Performance and energy claims will need to be assessed when production silicon becomes available for independent testing. Hardware operators should wait for these verified benchmarks before committing to architectural changes in their own products.
If the evaluation succeeds, DEEPX targets full-scale mass production for 2028. Watch for customer orders tied to this 2028 production target. The company has not yet disclosed any specific customer orders in the cited reports. The transition from a successful prototype to mass market adoption will dictate whether this processor becomes a reliable standard for regional hardware builders.
The push into physical and edge computing requires hardware that can run complex models efficiently outside of a data center. As regional semiconductor companies invest heavily in advanced manufacturing nodes, businesses gain more options for building intelligent local systems. Technology leads should evaluate their current cloud inference costs and consider how standard edge modules might change their product architecture in the coming years.