AnK Launches High-Powered GPU Infrastructure for Nepal’s AI Startups and Students

Kathmandu. As artificial intelligence gains momentum in Nepal, startups and students working on AI projects continue to face a major challenge: access to powerful computing infrastructure. Training AI models requires high-capacity GPUs and servers, but purchasing and maintaining such systems remains costly for most innovators.

To address this gap, Nepal-based startup AnK (Accelerated Komputing) has launched an Infrastructure-as-a-Service (IaaS) platform that allows students, researchers, and AI startups to rent high-performance GPU infrastructure remotely at affordable rates.

According to co-founder Amit Regmi, the company has set up a computing facility in Sipadol, Suryabinayak, Bhaktapur, where 19 high-performance systems are currently in operation.

Users can remotely access the infrastructure to train AI models, conduct engineering simulations, and perform advanced computational research without needing expensive hardware.

Regmi said the infrastructure initially emerged from the company’s own technical needs rather than from a business plan.

“AnK works on creating AI models for mechanical simulations. Ordinary desktop computers could not handle the load required for AI training, so we gradually started adding powerful GPUs and servers for our own work,” he said.

As the infrastructure expanded, the company realized that many startups and students in Nepal were struggling to access advanced GPU systems. This led AnK to expand its internal setup into a commercial Infrastructure-as-a-Service platform.

Infrastructure worth over Rs 30 million 

According to the company, its computing infrastructure is currently spread across Nepal and Japan. Chief Operating Officer Shuvangi Adhikari said 19 computing devices are currently installed in Nepal.

The setup includes eight NVIDIA RTX 5090 GPUs, four RTX 5080 units, five RTX 3090 systems, and two Mac Mini servers. In Japan, the company operates four DGX Spark Systems designed for enterprise-level AI workloads and large-scale model training. Plans are also underway to add NVIDIA RTX 6000 Pro systems there.

To support uninterrupted operations in Nepal, the company has installed a 40 kVA inverter, a dedicated 25 kVA electricity line, a cooling and thermal management system, and a 70 TB network-attached storage system for handling large datasets.

Regmi said more than Rs 20 million has been invested in Nepal for hardware, cooling, servers, and power infrastructure, while an additional Rs 15 million has been spent on infrastructure in Japan.

According to Chief Operating Officer Adhikari, users are provided with dedicated virtual machines that allow them to remotely access GPUs and storage from ordinary laptops. The infrastructure can be rented on hourly or monthly plans, depending on users’ needs.

The platform primarily targets AI startups training their own models, engineering and data science researchers, and university students.

“Instead of relying on expensive foreign infrastructure, we want Nepali companies to access the GPU resources they need from within Nepal,” Regmi said.

Connecting Mechanical Engineering and AI

Regmi, a graduate of Pulchok Engineering Campus in Mechanical Engineering, moved to Japan in 2011 for higher studies and employment. During his six-year stay there, he worked at companies including Honda, E&T Japan, and Misumi before later moving to the United States in 2017.

While in Japan, Regmi met Akio Tanaka, an electrical engineering graduate from Waseda University and founder of the Japanese company Godel Block.

Drawing on their combined experience in Computer-Aided Design (CAD), Computer-Aided Engineering (CAE), and manufacturing simulations, the duo launched AnK in 2023. 

According to Regmi, simulations are widely used in industries to test products such as car engines, aircraft components, or industrial equipment virtually before physical manufacturing begins. However, these simulations are computationally expensive and often require powerful computers running continuously for hours.

To reduce that burden, AnK is developing AI-based surrogate models capable of learning from thousands of previous simulation results. Once trained, these AI systems can predict simulation outcomes much faster, significantly reducing both time and cost in engineering workflows.

From AI-powered 3D design to automated dental systems 

The company has also developed an AI system called MEDA (Mechanical Engineering Design Agent), which can generate 3D designs from text prompts or simple 2D drawings. A research paper on the project was presented at the prestigious ASME conference in the United States.

AnK has additionally worked on AI-driven dental automation systems for hospitals in Japan. The system can automatically design replacement dental crowns by analyzing surrounding tooth structures and checking alignment during biting and jaw movement.

The company has also developed “PressNet,” a surrogate model used for glass-forming simulations in Japanese manufacturing industries. 

“There are no large-scale manufacturing industries in Nepal,” Regmi said. “That is why we are developing these surrogate models for advanced manufacturing markets like Japan and the United States.”

Seeking investment to expand the company’s capacity

Regmi currently works as a project manager at Carnegie Mellon University while remotely overseeing AnK ’s operations from the United States.

The company currently employs 14 people, most of whom are based in Nepal. According to Regmi, AnK has adopted a remote-working culture to attract skilled Nepali talent living abroad.

The company says its current revenue comes mainly from simulation and surrogate model projects, although significant investments are still being made into infrastructure development and research.

Regmi said external investment will be necessary if the company aims to scale its operations further.

He also stressed that Nepal has no shortage of skilled technical manpower and argued that lower labor costs could help position the country as a competitive destination for AI research and development.

About three months ago, students from the Institute of Engineering completed the first-phase test of a Vertical Takeoff and Vertical Landing (VTVL) rocket project, which received financial support from AnK. The company says it plans to continue supporting future phases of the project.

पछिल्लो अध्यावधिक: जेठ ३, २०८३ १२:५०

टिप्पणीहरू