Arm Neoverse CSS N4 debuts with 128 cores for AI infrastructure

The introduction of the Arm Neoverse CSS N4 provides cloud providers and streaming infrastructure teams a faster path to custom silicon optimized for autonomous agentic systems. By offering 1.25 times the performance per watt compared to previous generations, Arm is directly challenging x86 dominance in the power-constrained data center environment. For the streaming ecosystem, this efficiency gain is critical as platforms integrate more generative AI for content discovery and real-time metadata processing. The broad support from OpenAI, Cloudflare, and Nvidia suggests a consolidated industry shift toward Arm-based architectures for high-density AI compute. Watch for performance benchmarks from Oracle and SAP as they deploy these subsystems for enterprise-scale agentic workloads.

Arm's Neoverse platform has become the foundation for multiple hyperscaler custom silicon programs, creating a competitive ecosystem that extends well beyond the new CSS N4 subsystem. Google Cloud's Axion processor, built on Arm Neoverse V2, reached general availability in early 2025 and has been deployed for internal workloads including BigQuery and Spanner, demonstrating that Arm-based custom silicon can handle production-scale cloud services. Amazon's Graviton4, also Arm-based, expanded to 96 cores per socket and became available across more than 15 instance families by mid-2025, giving AWS the broadest Arm instance portfolio among hyperscalers. Microsoft's Cobalt 200, announced at Ignite 2024, targets cloud-native workloads with 128 Arm Neoverse cores and entered preview for Azure virtual machines in 2025, signaling that all three major clouds now operate production Arm silicon programs. Nvidia's Vera CPU, designed to pair with Rubin GPUs, uses Arm Neoverse V3 cores and is expected to ship in 2026, further cementing Arm's position as the default architecture for AI-adjacent compute. The business case for Arm in data centers has strengthened through licensing economics and power-efficiency mandates from cloud operators. Arm's royalty rates for Neoverse CSS designs increased with the V3 generation, reflecting a shift toward higher-value compute subsystem licensing rather than per-core fees, a model that aligns Arm's revenue with the complexity of partner implementations. Oracle's Ampere-based Arm instances delivered up to 40% better price-performance versus comparable x86 instances for Java and database workloads according to Oracle's published benchmarks, providing a concrete cost argument for enterprises evaluating Arm migration. Cloudflare, which has operated Arm-based servers since 2022, reported that its Ampere Altra deployments reduced power consumption by 30% compared to x86 equivalents at equivalent throughput, a data point that resonates with streaming infrastructure teams managing power budgets across edge and origin deployments. Independent benchmarking of Arm Neoverse platforms has focused on performance-per-watt advantages in cloud-native and AI inference workloads. SPECint_rate2017 results for Neoverse V2-based designs showed 1.4x throughput per watt versus comparable x86 server parts in standardized testing, though real-world gains vary by workload mix. For streaming-specific applications, Arm's SVE2 vector extensions have been validated for video transcoding pipelines, with FFmpeg and x264 ports showing 15-20% throughput improvements on Neoverse V2 versus prior-generation Arm cores. The AGI CPU announcement positions Arm to capture agentic AI inference at the edge of data centers, where latency-sensitive workloads like real-time content recommendation and metadata enrichment benefit from high core counts and lower thermal envelopes than GPU-accelerated alternatives.