Five South Korean companies — Orchestro Cloud, FuriosaAI, Upstage and two others — have launched development of a "Sovereign AI Full-Stack" built on South Korean neural processing units (NPUs), BigGo Finance reports.

The consortium frames the project as a push for technological self-reliance: an AI stack in which compute, silicon and software layers originate domestically rather than from foreign suppliers.

Who is building the stack?

• Orchestro Cloud — cloud infrastructure participant
• FuriosaAI — the Seoul-based NPU designer whose chips anchor the hardware layer
• Upstage — the large-language-model developer contributing model capability
• Two additional partners named in the launch

FuriosaAI supplies the domestic NPU base. The participation of Upstage, known for its Korean-language LLM work, places a model developer directly inside the stack alongside the silicon vendor and the cloud operator.

What does "sovereign AI" mean here?

The five companies aim to cover the full AI pipeline — chips, infrastructure and models — with South Korean technology. The launch aligns with a broader policy current in Seoul and other capitals: reduce dependence on foreign accelerators and foreign-hosted models for nationally relevant workloads.

A full-stack approach matters because sovereignty claims fail when any single layer depends on imported technology. Pairing FuriosaAI NPUs with a domestic cloud operator and a domestic model developer closes each of those gaps on paper.

Why does the consortium structure matter?

Each member covers a distinct layer:

• Hardware: FuriosaAI NPUs
• Cloud: Orchestro Cloud
• Models: Upstage
• The remaining two partners round out the stack

No timeline, budget figure or product milestone appeared in the initial announcement.

Five companies, one domestic stack, and no dates yet — the plan starts as architecture, not schedule.