[Interview] DeepX Captivates Taiwan: "Aiming to Be Number One in AI Chips"

Taiwan, the Hardware Manufacturing Hub, Targeted Proactively
Kim Nokwon: "To Win, We Must Fight and Triumph in the Tiger's Den"
Producing NPU Semiconductors to Surpass Nvidia GPUs

DeepX, a Korean fabless company developing artificial intelligence (AI) semiconductors, is drawing attention at Computex 2025, Asia's largest information technology (IT) exhibition, held from May 20 to 23 in Taipei, Taiwan, by setting up its booth next to Foxconn, a symbol of Taiwanese manufacturing. DeepX's neural processing unit (NPU) is evaluated as delivering equal or superior inference performance to Nvidia's graphics processing unit (GPU) while consuming less power.

DeepX has been proactively targeting the Taiwanese market and Computex since 2023, before the exhibition gained global prominence. While the United States and China are leading the advancement of AI, Taiwan remains the global hub for 'computer hardware' manufacturing. According to Kim Nokwon, CEO of DeepX, "We entered the heart of the tiger's den based on the belief that if we cannot succeed in Taiwan, we cannot succeed anywhere in the world with system semiconductors."

Kim Nokwon, CEO of DeepX, is taking a commemorative photo after an interview with Asia Economy at the DeepX booth inside the Nangang Exhibition Center in Taipei, Taiwan, where Computex 2025 is being held on the 20th. Photo by Jang Heejun

On May 20, at the Nangang Exhibition Center in Taipei, Kim Nokwon told Asia Economy, "Taiwan is both a crucial battleground we must conquer and a mecca where we can succeed," adding, "It is the starting and ending point of semiconductors, with customers and cutting-edge technology all gathered here, making it our second most important market after the United States."

Unlike other fabless companies such as FuriosaAI and Rebellions, which manufacture inference semiconductors for data servers, DeepX focuses on the NPU market for edge computing. The company develops system semiconductors that efficiently process AI computations. Computing systems based on NPUs deliver higher efficiency and lower power consumption compared to GPUs.

At this year's Computex, DeepX showcased its flagship AI semiconductor 'DX-M1' and related modules. The DX-M1 is an AI accelerator in the form of a system-on-chip (SoC) that uses the 'M.2 M-Key' interface, boasting excellent performance and versatility. It is built on an NPU platform for on-device AI inference.

DeepX focused on emphasizing collaboration with Taiwanese companies at 'Computex 2025'. Eleven Taiwanese companies, including Inventec, MSI, AIC, and DFI, jointly showcased DeepX's demo. DeepX

DeepX placed a strong emphasis on collaboration with Taiwanese companies. Eleven Taiwanese companies, including Inventec, MSI, AIC, and DFI, jointly exhibited DeepX's demos, and DeepX secured a spot on the fourth floor of Hall 1, where major companies such as Foxconn and Asus had their booths. CEO Kim stated, "This demonstrates that global top-tier companies are working with DeepX," and confidently added, "We are not just doing well; we are on our way to becoming number one."

Even before the launch of the DX-M1, DeepX had been collaborating with global companies to develop application products. These global partners are utilizing DeepX's DX-M1 to create industrial embedded computers and other products. In the workstation and server sector, DeepX has partnered with companies such as Ktous and AIC, while in the industrial PC sector, it has joined forces with Asus, Inventec, IEI, and DFI. Supermicro, led by a Taiwanese-American like Nvidia, is also collaborating with DeepX.

The DX-M1 is highly regarded for being optimized for 'edge AI' environments that require both high performance and low power consumption. Edge AI refers to technology that performs computations directly on devices, without sending data to the cloud. For example, it enables vehicles to detect driver drowsiness. Another example, demonstrated at the booth, is a 'smart factory' solution that detects whether factory workers are wearing safety helmets.

AI accelerator 'DX-M1' and module developed by DeepX. Photo by Heejoon Jang

DeepX is also developing the successor to the DX-M1, the DX-M2. CEO Kim stated, "The design of the DX-M2 model has been completed, and a prototype has already been produced." He emphasized, "The DX-M1's mass production yield has surpassed 90%, which demonstrates DeepX's deep understanding of yield and its outstanding design optimization capabilities." The DX-M1 utilizes Samsung Electronics Foundry's 5-nanometer process. Achieving a yield of over 90% at the initial mass production stage is considered exceptional.

Although the DX-M2 is expected to use a 2-nanometer process, Samsung Foundry has not yet secured stable yields for 2-nanometer production. CEO Kim dismissed concerns by saying, "We have optimal solutions to address yield issues."

Reflecting on Nvidia CEO Jensen Huang's keynote speech, CEO Kim stated, "The Korean semiconductor industry needs a leader." He explained that Nvidia received support from TSMC through low-cost wafer supply during difficult times, and other companies such as Realtek and MediaTek also grew under similar support. He urged, "I hope that large corporations and fabless companies in Korea will collaborate to enhance the market's competitiveness together."

Meanwhile, CEO Kim Nokwon is a former Apple application processor (AP) designer. He later worked at global leaders such as Broadcom, Cisco, and IBM before founding DeepX in 2018.