NVIDIA Welcomes Korean Chips into Its Ecosystem to Lower AI Costs
55s
2
13:10, 10.11.2026
NVIDIA is taking an unprecedented step by opening its proprietary MGX server rack standard to third-party developers. Under a memorandum of understanding with SK Telecom and startup Rebellions, American graphics processing units (GPUs) will work side-by-side with Korean neural processing units (NPUs) in the same data center.
Smart Labor Division
Instead of executing all tasks on ultra-expensive GPUs, the computing workloads will be split: NVIDIA's costly processors will handle the most complex computations, while Rebellions' cheaper Korean R200 chips will take over inference (everyday AI requests).
NVLink Fusion, To Bring It All Together
Another technological breakthrough is the use of Nvidia's proprietary NVLink Fusion interconnect.
This technology will allow third-party R200 chips to connect with the Nvidia system so seamlessly that the entire combination will operate as a single unit without data transfer delays.
Cheaper AI Tokens by 2028?
The full commercial launch of the infrastructure on SK Telecom's network is scheduled for 2028.
According to Rebellions CEO Park Sung-hyun, the primary goal of this alliance is to set a new global standard and make the production of "tokens" (units of AI-generated text) as cost-effective as possible.
Was this article helpful to you?
Yes
No
VPS popular offers
-10%
Windows
€ 28.44 /mo
Billed annually
-4.5%
Windows
€ 17.67 /mo
Billed annually
-7.2%
Linux
€ 6.4 /mo
Billed annually
-9.5%
Windows
€ 73.99 /mo
Billed annually
-10%
Linux
€ 69.99 /mo
Billed annually
-10%
Linux
€ 12.55 /mo
Billed annually
-10%
€ 149.04 /mo
Billed annually
-10%
Linux
€ 8.8 /mo
Billed annually
-8.1%
Windows
€ 31.25 /mo
Billed annually
-10%
Linux
€ 25.85 /mo
Billed annually
