Arm Everywhere China 2026 — Agentic AI, Physical AI & Neoverse Key Claims
Longscribe citation dossier for Arm Everywhere China 2026, covering CSS for Mobile 2, Mali G2-Ultra NX, SME2, Total Design for Physical AI, the Robotics Capability Framework, Neoverse CSS N4 and AGI CPU. The page turns Arm's official 8-minute keynote highlights into timestamped, source-grounded claims for journalists, analysts and researchers.
This page is a Longscribe editorial citation dossier, not a verbatim transcript. It is built from the creator-provided English captions on Arm's official YouTube upload. Product names and event context were cross-checked against Arm's same-day newsroom pages. Presenter names are intentionally not assigned because the edited highlight video does not identify each speaker in the spoken track.
Transcribed and published by Longscribe. Editorial policy.
Arm used its 2026 China keynote highlights to connect three parts of its AI strategy: an AI-native mobile platform, a physical-AI ecosystem effort and new cloud infrastructure. On mobile, the company introduced CSS for Mobile 2 and Mali G2-Ultra NX while arguing that agentic workloads increasingly stress the full CPU/GPU subsystem rather than a single accelerator. Source · 00:10
The physical-AI section shifts from products to coordination. Arm announced Total Design for Physical AI as a way to help partners build and validate systems, then paired it with a Robotics Capability Framework manifesto intended to give the industry a shared way to describe the problem and collaborate on common foundations. Source · 03:39
For cloud infrastructure, Arm introduced Neoverse CSS N4 with a wide core-count range and stronger socket performance, then linked that platform with AGI CPU density, China-market ecosystem work and a common software stack. The closing message is that Arm wants the same architecture and software base to span mobile, physical systems and agentic-AI data centers. Source · 04:37
Key facts
- CSS for Mobile 2
- New Arm mobile compute subsystem positioned for AI-native graphics and agentic AI. Source · 00:10
- Mali G2-Ultra NX
- Arm's first AI-native Mali GPU, introduced for 2026 mobile devices. Source · 01:32
- Mali performance claims
- Arm cites 4x FPS, 4x efficiency, 24% higher benchmark performance and 14% higher non-AI gaming performance. Source · 01:56
- Second SME2 engine
- Up to 70% higher performance in some small-language-model CPU workloads, according to Arm. Source · 03:10
- Total Design for Physical AI
- A new ecosystem effort to build, validate and scale optimized physical-AI solutions on Arm. Source · 03:39
- Robotics Capability Framework
- Arm released a manifesto asking the robotics industry to collaborate around a common capability framework. Source · 04:06
- Neoverse CSS N4 scale
- Configurable from 8 to 128 CPU cores, with multi-chiplet designs able to scale beyond that range. Source · 04:37
- Neoverse CSS N4 performance
- Arm claims 2x socket performance versus the previous generation and continued gains in performance per watt. Source · 05:18
- AGI CPU density
- Arm says one air-cooled server can hold 30 high-compute nodes with more than 8,000 Neoverse V3 cores. Source · 06:12
- Common software stack
- Arm says Linux, cloud-native, enterprise and AI/ML software is already optimized across its platform to reduce developer friction. Source · 06:53
Mobile and edge AI: CSS for Mobile 2, Mali G2-Ultra NX and SME2
Arm presents mobile AI as a system problem rather than a single-accelerator problem. CSS for Mobile 2 is the configurable foundation: partners can adapt Arm IP to their own SoC requirements, while the graphics pipeline increasingly uses neural processing to create and reconstruct pixels.
Mali G2-Ultra NX is the showcase GPU in the highlights. Arm positions it as its first AI-native Mali design and couples the neural-graphics story with conventional rendering gains. On the CPU side, the presentation says agentic workloads exercise the full cluster and adds a second SME2 engine, with a claimed uplift of as much as 70% on some small-language-model workloads.
Physical AI: ecosystem coordination and a robotics capability language
The physical-AI announcement is deliberately ecosystem-led. Arm says Total Design for Physical AI is intended to help companies develop, validate and scale optimized systems while reducing the integration burden across the stack.
The Robotics Capability Framework is presented as an invitation, not a completed standard. Arm's manifesto asks the industry to align on the problem and develop a common language for describing robotic capability, with collaboration itself treated as part of the infrastructure needed for physical AI to scale.
Cloud and agentic AI: Neoverse CSS N4, AGI CPU and the software ecosystem
Arm's cloud section emphasizes configurability, density and software continuity. Neoverse CSS N4 spans designs from eight to 128 cores and can extend through multi-chiplet systems; Arm also claims double the socket performance of the prior generation while continuing to improve efficiency.
AGI CPU is used to make the density argument. Arm says the platform is being adapted for the China market and can pack 30 high-compute nodes with more than 8,000 Neoverse V3 cores into one air-cooled server. The hardware story closes with software: Linux distributions, cloud-native tools, enterprise software and AI frameworks are described as a common layer that lets developers move across Arm-based systems with less friction.
Key timestamped claims
Each claim is an original Longscribe summary anchored to the moment(s) in the recording where it is supported; follow a timestamp to verify it in the source video.
00:10Arm announced CSS for Mobile 2 as its mobile platform for the AI era, pairing AI-native graphics with agentic-AI workloads.
00:39Arm says CSS for Mobile 2 lets silicon partners configure its latest IP around their own requirements and use it in custom SoCs.
00:56The keynote describes moving AI into the graphics pipeline itself so neural engines can create pixels, with the GPU evolving beyond traditional rendering toward an AI engine.
01:35Arm introduced Mali G2-Ultra NX as its first AI-native Mali GPU and said it expects the part to be its most advanced mobile GPU offering this year.
01:56For Mali G2-Ultra NX, Arm claims 4x FPS and 4x efficiency improvements for AI-native graphics, plus 24% higher benchmark performance and 14% higher non-AI gaming performance.
02:38Arm says agentic workflows can occupy the whole CPU cluster and that its new cluster brings double-digit gains, including stronger multithreaded performance for multi-agent workloads.
03:10Arm says a second SME2 engine can deliver up to 70% higher performance in some small-language-model workloads running on the CPU.
03:39Arm announced Total Design for Physical AI, describing it as a program to help the ecosystem build, validate and scale optimized physical-AI solutions while developing shared foundations.
04:08Arm also released a manifesto calling for industry collaboration on a Robotics Capability Framework, framing it as a shared problem statement rather than a finished solution.
04:37Arm introduced Neoverse CSS N4 as its most configurable Neoverse CSS so far, supporting SoCs from 8 to 128 cores and multi-chiplet designs that can scale further.
05:22For Neoverse CSS N4, Arm claims twice the socket performance of the prior generation while continuing to improve performance per watt.
05:41Arm says the AGI CPU launched in March and that, six months later, ecosystem work is adapting the platform for the needs of the China market.
06:12Arm says an AGI CPU configuration can place 30 high-compute server nodes and more than 8,000 Neoverse V3 cores in a single air-cooled server for AI workloads.
06:37Arm announced two additional China-market partners that are evaluating deployment of AGI CPU and related compute services across China; the highlight clip does not name them in the spoken track.
06:53Arm emphasizes a common software layer across the platform, citing Linux, cloud-native software, enterprise applications and AI/ML frameworks already optimized for Arm as a way to reduce developer friction.
07:43Arm says its major AI-hyperscaler partners are committed to long-term roadmaps based on Neoverse technology as the company positions the ecosystem for agentic-AI data centers.
Source & methodology
Source: the official Arm YouTube video HJ2opWgOY3U, titled Arm Everywhere China 2026 | Keynote highlights, published September 8, 2026 with an 08:22 runtime. Longscribe captured the creator-provided en-US caption track; the product's YouTube ingestion path prefers creator-made captions over generated captions when they are available. This page is an original paraphrased citation dossier and timestamped claim index, not a verbatim transcript. Every public factual block is grounded to committed source segments in docs/ops/curated-transcript-sources/HJ2opWgOY3U.json, and the companion coverage manifest accounts for every caption segment as covered or explicitly excluded. Arm's September 8 newsroom material was used to cross-check product naming and event context, but the facts displayed in the dossier are grounded in the recording itself. The highlight edit does not verbally identify each presenter by name, so Longscribe does not infer speaker identities from faces or role assumptions.
Paste an accessible public YouTube video or Shorts URL: you do not need to download it to your device. LongScribe uses captions when available or transcribes reachable speech with AI. Three hours is 180 transcription minutes; signed-in Free includes 60 minutes across 20 recordings per month, subject to remaining allowance, per-job limits and source availability. Completed eligible account jobs can export TXT, SRT, DOCX or JSON; the anonymous demo is a separate flow with its own limits.