--- title: Why CPUs still matter in the age of AI agents | daily.dev description: As AI workloads shift from chatbots to autonomous agents, CPUs are regaining strategic importance. Google's Mo Farhat describes the CPU as an 'air traffic... canonical: https://daily.dev/posts/why-cpus-still-matter-in-the-age-of-ai-agents-mtvzvx9b6 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Why CPUs still matter in the age of AI agents | daily.dev og:description: As AI workloads shift from chatbots to autonomous agents, CPUs are regaining strategic importance. Google's Mo Farhat describes the CPU as an 'air traffic... og:url: https://daily.dev/posts/why-cpus-still-matter-in-the-age-of-ai-agents-mtvzvx9b6 og:image: https://api.daily.dev/og/posts/MtvzvX9B6.png og:image:alt: Why CPUs still matter in the age of AI agents og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Why CPUs still matter in the age of AI agents **[The New Stack](https://daily.dev/sources/newstack)** · 4 min read · 0 upvotes · 0 comments ## Summary As AI workloads shift from chatbots to autonomous agents, CPUs are regaining strategic importance. Google's Mo Farhat describes the CPU as an 'air traffic controller' for agentic systems, handling orchestration, control-flow logic, data preparation, semantic search, and vector databases — while GPUs/TPUs handle heavy model inference. Modern 6–8B parameter models can run on CPUs at ~25 tokens/second, sufficient for many agentic tasks. Secure code execution is addressed via gVisor-based sandboxing; Google's GKE Agent Sandbox can spin up 300 sandboxes per second per cluster using pod snapshots and warm pools. Google's Arm-based Axion processors (C4A for high performance, N4A for cost/density) offer up to 2x price-performance and 60%+ better energy efficiency versus comparable VMs. The overall recommendation is a fluid compute model combining CPU, GPU, and TPU resources. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How many tokens per second can a CPU deliver for agentic AI workloads? CPUs can deliver roughly 25 tokens per second for some specialized agentic workloads. This throughput is considered sufficient for many agent tasks, particularly with modern 6–8 billion parameter models, which perform significantly better than earlier generations. CPUs handle orchestration, control-flow logic, data preparation, semantic search, and vector databases, while accelerators handle heavier model inference. _Teams sizing CPU vs. GPU infrastructure for agent pipelines track benchmarks like these on daily.dev._ ### How fast can GKE Agent Sandbox spin up sandboxes for AI agent code execution? GKE Agent Sandbox can spin up 300 sandboxes per second per cluster. It uses pod snapshots and warm pools of suspended environments to scale quickly without keeping all environments fully provisioned. The sandboxing is built on gVisor, an open-source isolation layer between the application and the host OS, designed for zero-trust environments where LLM-generated code cannot be implicitly trusted. _Engineers securing AI agent code execution environments find the latest sandboxing approaches on daily.dev._ ### What is the price-performance advantage of Google Axion C4A vs comparable VMs? Google Axion C4A offers up to 2x the price-performance of comparable current-generation virtual machines, with over 60% better energy efficiency. C4A targets compute-bound, long-running jobs where completion time is critical, while N4A targets smaller code-execution tasks where density and cost matter more. Both are Arm-based and positioned for scaling large numbers of AI agents. _Cloud architects choosing between VM families for agent workloads follow comparisons like this on daily.dev._ ## Similar posts on daily.dev - [The CPU's growing role in agentic AI infrastructure](https://daily.dev/posts/the-cpu-s-growing-role-in-agentic-ai-infrastructure-a8rzlmngb) · DEVCLASS · 0 upvotes · 0 comments - [Agentic AI Crunch Creates CPU Comeback](https://daily.dev/posts/agentic-ai-crunch-creates-cpu-comeback-cnadb5uae) · IEEE Spectrum · 1 upvotes · 0 comments - [Why real-world AI performance depends on the control layer](https://daily.dev/posts/why-real-world-ai-performance-depends-on-the-control-layer-undig6ygq) · The Register · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#gcp](https://daily.dev/tags/gcp), [#computing](https://daily.dev/tags/computing), [#arm](https://daily.dev/tags/arm) [View this post on daily.dev](https://daily.dev/posts/why-cpus-still-matter-in-the-age-of-ai-agents-mtvzvx9b6) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Why CPUs still matter in the age of AI agents","url":"https://daily.dev/posts/why-cpus-still-matter-in-the-age-of-ai-agents-mtvzvx9b6","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/why-cpus-still-matter-in-the-age-of-ai-agents-mtvzvx9b6"},"datePublished":"2026-08-11T20:50:58.026Z","dateModified":"2026-08-11T20:51:30.177Z","description":"As AI workloads shift from chatbots to autonomous agents, CPUs are regaining strategic importance. Google's Mo Farhat describes the CPU as an 'air traffic...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/5169adcde074f3207a313826e3873f55?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/5169adcde074f3207a313826e3873f55?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The New Stack","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The New Stack","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/newstack","url":"https://daily.dev/sources/newstack"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/why-cpus-still-matter-in-the-age-of-ai-agents-mtvzvx9b6","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,gcp,computing,arm","timeRequired":"PT4M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The New Stack","item":"https://daily.dev/sources/newstack"},{"@type":"ListItem","position":3,"name":"Why CPUs still matter in the age of AI agents"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/why-cpus-still-matter-in-the-age-of-ai-agents-mtvzvx9b6#faq","mainEntity":[{"@type":"Question","name":"How many tokens per second can a CPU deliver for agentic AI workloads?","acceptedAnswer":{"@type":"Answer","text":"CPUs can deliver roughly 25 tokens per second for some specialized agentic workloads. This throughput is considered sufficient for many agent tasks, particularly with modern 6–8 billion parameter models, which perform significantly better than earlier generations. CPUs handle orchestration, control-flow logic, data preparation, semantic search, and vector databases, while accelerators handle heavier model inference. Teams sizing CPU vs. GPU infrastructure for agent pipelines track benchmarks like these on daily.dev."}},{"@type":"Question","name":"How fast can GKE Agent Sandbox spin up sandboxes for AI agent code execution?","acceptedAnswer":{"@type":"Answer","text":"GKE Agent Sandbox can spin up 300 sandboxes per second per cluster. It uses pod snapshots and warm pools of suspended environments to scale quickly without keeping all environments fully provisioned. The sandboxing is built on gVisor, an open-source isolation layer between the application and the host OS, designed for zero-trust environments where LLM-generated code cannot be implicitly trusted. Engineers securing AI agent code execution environments find the latest sandboxing approaches on daily.dev."}},{"@type":"Question","name":"What is the price-performance advantage of Google Axion C4A vs comparable VMs?","acceptedAnswer":{"@type":"Answer","text":"Google Axion C4A offers up to 2x the price-performance of comparable current-generation virtual machines, with over 60% better energy efficiency. C4A targets compute-bound, long-running jobs where completion time is critical, while N4A targets smaller code-execution tasks where density and cost matter more. Both are Arm-based and positioned for scaling large numbers of AI agents. Cloud architects choosing between VM families for agent workloads follow comparisons like this on daily.dev."}}]} ```