DIGITIMES Asia reported on 9 November 2026 that Nvidia, SK Telecom and the South Korean AI chip startup Rebellions have agreed to develop AI inference systems at rack scale. Inference means running a finished AI model to answer requests. Nvidia's GPUs would sit alongside next-generation chips from Rebellions, known as NPUs (neural processing units), which are built for AI work. A photo caption places staff from all three firms, plus South Korea's science ministry, at a signing event in Seoul.
The opening of the report gives no product name, launch date, price or customer. It does not say how the systems would be sold, or how work would be divided between the two chip types. The summary also breaks off mid-sentence while describing Nvidia's side of the deal, so more detail may sit in the full article.
"Rack-scale" refers to a whole rack of hardware built to work as one system. For teams that buy or run AI inference, the deal points to systems that use more than one kind of chip. Until details appear, buyers can only ask vendors which chip handles which work.