openKylin wants Linux to do more. The open-source operating system community originating in China used October to outline an AI-integrated operating system for global users, while industry leaders launched an open framework for managing AI data center infrastructure.
At Open Source Summit Europe 2026 in Prague, Czechia, openKylin presented its vision from October 7 to 9. Min Liu, Executive Director of openKylin, delivered the main-stage keynote on October 8, titled “From AI Runtime to System Agents: Reimagining Linux for the AI Era with openKylin.”
The proposal starts with an unfashionable but important observation: model capability is not the only obstacle. openKylin identifies the deeper problem as system infrastructure that fails to connect AI models with desktop tasks, leaving users with fragmented AI stacks and plenty of buttons that do not know what to do with one another.
From resource management to intent management
openKylin proposes shifting operating systems from resource management to intent management. Its architecture has three layers: a runtime layer, an agent execution layer, and a capability layer.
The runtime layer supports KART, a unified AI runtime. The agent execution layer provides a system-level environment for running agents, while the capability layer connects those agents to system functions. Together, the three layers enable the operating system to understand user goals and translate them into system actions.
That approach appears in openKylin 3.0 through AI Assistant, Global Intelligent Search, and Intelligent Space. The names are straightforward, which is refreshing; the difficult part is making these features work across hardware, software, and desktop tasks without turning the operating system into a collection of disconnected demos.
Jianfeng Li, a community Technical Committee member, delivered a lightning talk titled “Building an AI-Ready Linux Desktop on RISC-V.” The hardware focus matters because openKylin’s plan includes hardware enablement, RISC-V Vector extension optimization, AI runtimes, and heterogeneous computing.
RVV kernel optimizations deliver a 2–5× speedup in FP32 inference with ONNX Runtime. The AI stack also supports CPU inference with vLLM on SpacemiT K3, while openKylin showcased AI PCs and devices at the summit, including a VR experience, a K3 laptop, and AI edge computing products.
AI infrastructure needs a control layer
On October 13, industry leaders announced the Secure Control and Management Protocol, or SCMP, OCP workstream during a joint keynote at the 2026 OCP Global Summit. Lattice Semiconductor and AMI announced the workstream, which launches in Q4 2026 and remains open to all OCP participants.
SCMP is an open, vendor-neutral architecture for secure control, boot, and management of data center infrastructure. Its goal is to give operators confidence that complex AI systems will work as expected across vendors and platforms — a reasonable demand when data center AI capital expenditure is projected to reach $1.2 trillion in 2027.
The workstream defines how a control authority establishes and maintains trust from a silicon root of trust through the rack and pod. It addresses boot-up procedures, control and telemetry interface standardization, component identity trust chains, and procedures for recovery and ownership transfer.
Founding contributors include AMD, AMI, Arm, ASPEED, Intel, Jabil, Lattice, Microsoft, and NXP. The work will sit within the OCP community, which has more than 400 member companies, giving SCMP a broad field of potential adopters rather than a single vendor’s private rulebook.
Ford Tamer, CEO of Lattice Semiconductor, said, “We’re pleased to launch this work with OCP and our founding contributors, and we invite the industry to help define it.” Lattice Semiconductor is a low power programmable and platform firmware leader, while Sanjoy Maity serves as SVP of AMI, a Lattice company.
Maity said, “SCMP turns that experience into an open specification that any vendor can implement and any operator or auditor can verify.” That distinction matters: interoperability is useful, but verifiability is what keeps infrastructure promises from becoming another slide deck with a server rack in the background.
Arm is also contributing to the effort. Eddie Ramirez, Vice President of Go-to-Market, Cloud AI, Arm, said, “Through SCMP, Arm is bringing our compute, security and platform expertise to the OCP community to help define an open, interoperable foundation for trusted AI infrastructure at scale.”
George Tchaparian, CEO of the Open Compute Project Foundation, called SCMP “an important milestone in advancing open, trusted infrastructure management for next-generation AI data centers.” Taken together, the two October developments point to the same pressure: AI needs operating systems that understand intent and infrastructure that can prove what it is doing.
Based on
