Arm Holdings is betting that the next wave of artificial intelligence will move beyond chatbots and data centers into phones, robots and cars. The Cambridge-based chip design giant on Monday introduced a second-generation mobile compute platform with a dedicated AI accelerator inside its graphics processor, alongside a new server design that doubles per-socket performance and, for the first time, formally supports Samsung Electronics’ 2-nanometer foundry process.<\/p>\n
The announcements mark Arm’s most aggressive push yet into what it calls physical AI: systems that perceive their surroundings, make decisions and act in the real world. Company executives framed the launch as a response to AI workloads that are migrating from the cloud toward edge devices, where power efficiency and latency constraints demand fundamentally different silicon.<\/p>\n
The centerpiece of the mobile lineup is CSS for Mobile 2, an update to the Lumex CSS platform that underpinned Samsung’s Exynos 2600 chip used in recent Galaxy devices. The new design pairs Arm’s C2 CPU cluster with the Mali G2-Ultra NX GPU, which Arm describes as its first AI-native graphics processor.<\/p>\n
Inside that GPU sits the NX neural accelerator, a dedicated block for on-device AI computation. Rather than rendering every frame at full resolution from the start, the system can generate game visuals at 540p and upscale them to 1080p, or interpolate 30 frames per second into 60, using AI to fill in the gaps. Arm said the approach boosts AI graphics frame performance and efficiency by up to four times while cutting DRAM data movement by 70%.<\/p>\n
That matters in smartphones, where battery capacity is fixed and every milliwatt counts. The pitch is straightforward: deliver console-quality visuals without draining the battery or forcing users to rely on cloud rendering.<\/p>\n
On the data center side, Arm rolled out Neoverse CSS N4, codenamed Ranger, a semi-custom platform that lets customers configure core counts, cache sizes, I\/O and connectivity to match their workloads. The design supports between eight and 128 Neoverse N4 cores per die running at up to 3.8 GHz, with system-level scaling beyond that through multi-chiplet and multi-socket architectures.<\/p>\n
Arm said the platform delivers twice the socket performance of its N3 predecessor, a 25% improvement in performance per watt, and 75% more memory bandwidth. Each die can carry up to 256 MB of L3 cache, with 2 MB of L2 per core and 64 KB each of L1 instruction and data cache. The platform supports DDR5 or LPDDR6 memory and up to 128 lanes of PCIe 7\/6 and CXL 4.0.<\/p>\n
The jump from the previous N2 generation is substantial. That design topped out at 64 cores, 1 MB of L2 per core, 64 MB of L3 cache, and 64 PCIe 5.0\/CXL lanes.<\/p>\n
Perhaps the most strategically significant detail is the foundry list. Arm said Neoverse CSS N4 is validated on TSMC’s 3-nanometer and 2-nanometer nodes, as well as Samsung’s first- and second-generation 2-nanometer processes. The inclusion of Samsung’s SF2 node gives the Korean foundry a high-profile reference design as it competes with TSMC for leading-edge AI silicon customers.<\/p>\n
It is not the first time Arm and Samsung have collaborated on 2-nanometer technology. In 2024, the two companies announced plans with ADTechnology and Rebellions to implement a Neoverse CSS V3-based AI CPU chip on Samsung’s 2-nanometer process.<\/p>\n
Arm is also extending its reach into finished silicon. Alongside the CSS N4 announcement, the company said Oracle and ByteDance will deploy its AGI CPU, joining previously disclosed customers including Meta, Lenovo, SAP, OpenAI and Cloudflare. The AGI chip, built with Neoverse V3 cores, is Arm’s first production silicon and features up to 136 cores, 272 MB of L3 cache and support for up to 6TB of memory per chip.<\/p>\n
The company’s business model is shifting as a result. For years, Arm licensed IP and let customers such as Samsung, Qualcomm and Apple build their own chips. The CSS program already bundles CPU, GPU and system IP into validated platforms. The AGI CPU goes a step further, offering a complete data center processor to cloud providers that need to scale quickly.<\/p>\n
Arm has framed the move as expanding customer choice rather than competing with its own ecosystem. Companies that want to build custom SoCs can still license IP or use CSS platforms, while hyperscalers that need fast server expansion can buy the finished CPU.<\/p>\n
That strategy carries financial implications. Arm’s IP licensing business runs at roughly 92.5% gross margin. The AGI CPU, by contrast, is guided to the high 30% range, possibly low 40s in its first generation. The company’s stock trades at around 298 times earnings after a 130% year-to-date run, meaning investors are pricing in aggressive growth in the data center silicon business.<\/p>\n
The competitive landscape is unforgiving. NVIDIA posted $89 billion in data center revenue in a single quarter and its Grace CPU, built on Arm’s Neoverse V2, has already surpassed $5 billion in trailing twelve-month revenue. Qualcomm, Arm’s closest strategic peer, is targeting $15 billion in data center revenue by fiscal 2029 but faces a 1.5% to 2% margin drag from its custom silicon ramp and an ongoing license dispute with Arm scheduled for trial in the fourth quarter of 2026.<\/p>\n
Arm’s physical AI ambitions extend beyond silicon. The company announced Arm Total Design for Physical AI, an ecosystem initiative bringing together more than 80 partners including Samsung, TSMC, AWS, Hugging Face, NXP, Siemens, QNX, Unitree and ECARX. The goal is to connect semiconductor, cloud, sensor, software and robotics companies to accelerate development and validation of physical AI products.<\/p>\n
Arm also said it will propose a robotics capability framework next week, modeled on the Level 1 through Level 5 autonomy scale used for self-driving vehicles. The framework would create a common language for comparing robot capabilities across different vendors and use cases.<\/p>\n
The company projected that the physical AI computing market will exceed $200 billion annually by the 2030s. That is the prize Arm is positioning for: not just powering the next smartphone or cloud server, but becoming the default architecture for machines that interact with the physical world.<\/p>\n
Whether the strategy pays off depends on execution. The Neoverse N4 cores, codenamed Dionysus, will not appear in shipping products for some time. Arm has not yet named CSS N4 customers, and the AGI CPU has yet to produce real-world performance benchmarks. The company’s claims of more than twice the performance per rack compared to the latest x86 systems remain based on internal estimates.<\/p>\n
Still, the direction is clear. Arm already claims roughly 50% CPU compute share among top hyperscalers, with its technology inside NVIDIA’s Vera CPU, Google’s Axion, Microsoft’s Cobalt and Amazon’s Graviton 5. The company says Neoverse shipments have surpassed 1.5 billion cores, and data center royalty revenue more than doubled year over year in the most recent quarter.<\/p>\n","protected":false},"excerpt":{"rendered":"Arm Holdings is betting that the next wave of artificial intelligence will move beyond chatbots and data centers…\n","protected":false},"author":2,"featured_media":146806,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[18],"tags":[74631,4680,3628,37237,74629,49614,74630,15877,74628,2295,74632,6226,23122,127,276,3246],"class_list":["post-146805","post","type-post","status-publish","format-standard","has-post-thumbnail","category-samsung-electronics","tag-agi-cpu","tag-amazon-web-services","tag-apple","tag-arm-holdings","tag-css-for-mobile-2","tag-hugging-face","tag-mali-g2-ultra-nx","tag-meta","tag-neoverse-css-n4","tag-nvidia","tag-nxp-semiconductors","tag-openai","tag-qualcomm","tag-samsung","tag-samsung-electronics","tag-tsmc"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/korea\/wp-json\/wp\/v2\/posts\/146805","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/korea\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/korea\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/korea\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/korea\/wp-json\/wp\/v2\/comments?post=146805"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/korea\/wp-json\/wp\/v2\/posts\/146805\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/korea\/wp-json\/wp\/v2\/media\/146806"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/korea\/wp-json\/wp\/v2\/media?parent=146805"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/korea\/wp-json\/wp\/v2\/categories?post=146805"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/korea\/wp-json\/wp\/v2\/tags?post=146805"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}