You're reading the Sunday, October 18, 2026 edition. Today's briefing →
Toronto
Sunday, October 18, 2026No. 126

Digital Plumber

Plumbing the information age

AI-curated intelligence for people who run networks. Daily coverage of AIOps, network automation, agentic operations, AI infrastructure, security and the vendors shaping them.

Today's 3 things that matter

Picked by the AI editor
  1. Automation·Primary source

    NetBox Labs links source of truth to NVIDIA and Grafana

    NetBox Labs announced partnerships with NVIDIA Config Manager and Grafana Labs, expanding the NetBox Infrastructure Intelligence Platform to support AI datacenter operations with enhanced observability and configuration management integrations.

    Why it matters Directly addresses network source-of-truth integration with AI infrastructure automation—critical for operators managing large-scale AI datacenter deployments.

  2. DC networking·Vendor release

    IP Infusion OcNOS adds Broadcom Tomahawk 6 at 102.4 terabits

    IP Infusion announced OcNOS-DC support for Broadcom Tomahawk 6, enabling 102.4 Tbps fabric with 512 ports of 200G, supporting scale-up clusters of 128,000 XPUs in a two-layer leaf-spine topology.

    Why it matters Eliminates super-spine layers and reduces hop count, cutting latency and optics cost while maintaining lossless RoCEv2 at full line rate for AI workloads.

  3. Startups·Industry news

    Pluto Security leaves stealth with $42M to govern AI tools

    Pluto Security emerged from stealth with $42M in total funding ($8.5M seed led by TLV Partners, $34M Series A led by Jump Capital) to secure AI applications and agentic workflows at the endpoint level.

    Why it matters Infrastructure and security ops teams managing AI agent deployment now have a baseline endpoint security control for agentic workflows expanding across enterprises.

Today's briefing

What happened, and why it matters

10 stories · 6 topics · Updated 4:03 PM ET

NetBox Labs links source of truth to NVIDIA and Grafana

NetBox Labs blog · Oct 13, 2026 · Primary source

NetBox Labs announced partnerships with NVIDIA Config Manager and Grafana Labs, expanding the NetBox Infrastructure Intelligence Platform to support AI datacenter operations with enhanced observability and configuration management integrations.

Why it matters Directly addresses network source-of-truth integration with AI infrastructure automation—critical for operators managing large-scale AI datacenter deployments.

NetBox Labs published three major announcements: (1) NVIDIA partnership for AI datacenter deployment and operations, integrating NVIDIA Config Manager; (2) Grafana Labs partnership for enhanced network and datacenter infrastructure observability; and (3) NetBox Designs public preview to reduce manual entry in network builds. These moves position NetBox as the system of record for AI hyperscale operations. For NetDevOps practitioners, this means tighter integration between your network source of truth (NetBox as IPAM/DCIM), AI infrastructure automation (NVIDIA), and observability (Grafana)—eliminating hand-offs between teams. The Designs feature directly reduces toil in day-1 network provisioning by automating the design-to-deploy workflow. Platform MCP (Model Context Protocol) is now generally available, enabling AI agents to act on intent through a structured interface.

Read the original at netboxlabs.com ↗

Itential maps shift from static runbooks to agent-driven network ops

Itential blog · Oct 15, 2026 · Primary source

Itential published a practitioner-focused guide on deploying AI across network and infrastructure, covering the shift from static runbooks to agent-driven operations with human-in-the-loop controls and intent-first network design.

Why it matters Shows the practical operationalization of AI agents in network automation—from experimental assistants to autonomous, governed operations—directly applicable to AIOps and NetDevOps practice.

Joksan Flores (Itential Principal Solutions Engineer, formerly Cisco networking architect) outlines the 2026 reality of AI in network operations. Three key themes: (1) Agent-Driven Operations—agents replace static runbooks by negotiating, validating, and escalating intelligently; engineers supervise, refine intent models, and approve high-impact changes. (2) Intent-First Networking—engineers define outcomes (performance, security, resilience), and agent-driven systems determine execution across vendors, domains, and layers. (3) The missing pieces are now available: MCP (Model Context Protocol), modern AI agents, A2A protocols, and orchestration platforms. For teams building NetDevOps workflows, this is the roadmap: start with high-fidelity intent models (Nautobot Jobs, Itential FlowAgent Builder), wire in human gates (especially for production), and let agents handle the cross-domain orchestration. The post specifically addresses the 2026 gap: AI agents work well in labs, but production requires fail-safe design—rollback capabilities, human checkpoints, and structured intent validation. Updated October 15, 2026.

Read the original at itential.com ↗

Oracle urges standard interfaces between AI data centres and grids

Data Center Frontier · Oct 15, 2026 · Industry news

At OCP Summit 2026, Oracle called for standardized interfaces between AI data centers and power grids; Microsoft, Oracle, HPE, Intel, and Google described an emerging need to coordinate computing, networking, power, cooling, and operations as unified architecture.

Why it matters Network infrastructure planning must now account for energy efficiency, grid integration, and system-level optimization—not just throughput and latency.

Oracle Cloud Infrastructure VP Pradeep Vincent called for industry-wide standards governing how large-scale AI data centers interact with electrical grids, emphasizing energy efficiency, cooling reliability, and responsible infrastructure development. Microsoft outlined its Maia 300 roadmap and full-stack co-design strategy to optimize AI inference performance per watt and per dollar. Intel and HPE emphasized heterogeneous, workload-specific computing architectures rather than one-size-fits-all design. OCI Supercluster now scales beyond 100,000 NVIDIA GB200 Superchips or 800,000 to 1 million GPUs for large-scale training and inference, combining bare-metal GPUs with high-speed RDMA networking. This reflects maturation of the field: after years of raw capacity expansion, operators now prioritize system-level efficiency metrics and grid-aware resource management.

Read the original at datacenterfrontier.com ↗

IP Infusion OcNOS adds Broadcom Tomahawk 6 at 102.4 terabits

Financial Content / IP Infusion · Oct 14, 2026 · Vendor release

IP Infusion announced OcNOS-DC support for Broadcom Tomahawk 6, enabling 102.4 Tbps fabric with 512 ports of 200G, supporting scale-up clusters of 128,000 XPUs in a two-layer leaf-spine topology.

Why it matters Eliminates super-spine layers and reduces hop count, cutting latency and optics cost while maintaining lossless RoCEv2 at full line rate for AI workloads.

OcNOS-DC on Tomahawk 6 delivers native 1.6T ports supporting scale-up networks connecting 512 accelerators in a single hop, while scale-out clusters reach 128,000 XPUs in a two-layer leaf-spine design. A flatter fabric topology reduces hops at every scale: for scale-up, twice the capacity in every switch means fewer devices in the traffic path; for scale-out, eliminating the super-spine layer dramatically lowers latency. OcNOS-DC provides Priority Flow Control (PFC) and Explicit Congestion Notification (ECN) to maintain lossless RoCEv2 transport at full line rate. With 512×200G ports, operators can connect more accelerators per leaf and build larger clusters with fewer switches. Scale-across topologies benefit from 1.6T and 800G uplinks feeding interconnect platforms running OcNOS, keeping data-center-to-data-center capacity in sync with growing intra-cluster demands.

Read the original at financialcontent.com ↗

Pluto Security leaves stealth with $42M to govern AI tools

TechCrunch · Oct 13, 2026 · Industry news

Pluto Security emerged from stealth with $42M in total funding ($8.5M seed led by TLV Partners, $34M Series A led by Jump Capital) to secure AI applications and agentic workflows at the endpoint level. The platform monitors AI tool interactions, detects exposed credentials in milliseconds using custom small language models, and can intervene in real time across browsers, desktop apps, and agent workflows.

Why it matters Infrastructure and security ops teams managing AI agent deployment now have a baseline endpoint security control for agentic workflows expanding across enterprises.

Pluto Security combines endpoint visibility, contextual AI interaction analysis, and real-time policy enforcement to monitor how employees and AI agents use AI tools. The platform runs proprietary small language models on endpoints to classify tasks and detect sensitive data exposure—distinguishing between legitimate schema references and credential leaks—returning decisions in milliseconds. Unlike traditional DLP systems, Pluto can inject context back into sessions to prompt agent confirmation before risky permissions escalations. The founding team includes engineers from Microsoft, Palo Alto Networks, and browser security backgrounds. The company reports $2M+ ARR and serves Fortune 500 organizations including Oscar Health, Workato, Sixt, and AlphaSense. Early adoption data shows 60% of detected AI security risks originate from non-developer-authored applications, with finance, marketing, and legal departments as the largest risk sources. The platform integrates with Cisco Agent Builder via MCP and Anthropic's Claude Compliance API, positioning it as an infrastructure control point for the emerging agentic enterprise stack.

Read the original at techcrunch.com ↗

Harmonic Security raises $50M to watch workforce AI data exposure

TechCrunch · Oct 14, 2026 · Industry news

Harmonic Security raised $50M Series B led by Crosspoint Capital Partners, bringing total funding to $76M. Its platform runs custom small language models to inspect employee interactions with AI tools in browsers, desktop apps and agentic workflows, flagging exposed credentials in milliseconds. The company protects more than 100 enterprises with millions of employees globally.

Why it matters Security and ops teams can now govern AI interactions in real time with sub-200ms latency and contextual risk classification, enabling safe AI adoption as autonomous agents proliferate in production.

Harmonic said it will use the new capital to expand its engineering and go-to-market teams in the United States and United Kingdom, growing to more than 90 people across both countries. The platform governs employee use of AI tools in real time by inspecting interactions as they happen and enforcing controls based on what it finds. Harmonic relies on custom small language models rather than giant general-purpose ones, classifying the task an employee is attempting and detecting sensitive data in the exchange. Harmonic puts its latency at about 200ms and claims 96% fewer false positives than traditional data loss prevention systems. The Usage Intelligence layer shows leaders which AI use cases create value, which employees have found faster ways to work, and where to invest next. The company integrates with Cisco Agent Builder via MCP and Anthropic's Claude Compliance API, addressing the gap between blocking risky behavior and enabling productivity in the agentic era.

Read the original at unite.ai ↗

Verizon and Ericsson make first outdoor 6G call at 7GHz

TelecomTV · Oct 16, 2026 · Industry news

Verizon completed first pre-standard 6G field call in 7GHz spectrum band using Ericsson 6G proof-of-concept technology at commercial network site in Concord, California, following earlier ISAC trials.

Why it matters Transition from lab to live-site 6G validation signals near-term spectrum readiness; operators should begin planning RAN modernization and agentic AI integration timelines for next-generation deployment.

Verizon's 6G field trial represents a concrete step beyond isolated lab proofs toward commercial-grade pre-standard validation. The deployment leveraged Ericsson 6G proof-of-concept technology at a live Verizon commercial network site in Concord, California, to establish an outdoor 6G field call in the 7GHz spectrum band. This follows earlier ISAC (integrated sensing and communication) trials conducted with Samsung and Qualcomm. For network engineering teams, the relevance is twofold: first, the move toward outdoor site validation signals that vendors and operators are transitioning from controlled indoor trials to real-world propagation and interference scenarios; second, Verizon's broader 60,000-site vRAN is now applying agentic AI to planned configuration changes, service assurance, and network optimization while calling for industry-wide interoperability standards, suggesting that RAN automation and 6G readiness are being pursued in parallel.

Read the original at telecomtv.com ↗

Spectrum deploys NVIDIA edge compute across more than 1,000 sites

TimothySykes / StocksToTrade · Oct 13, 2026 · Industry news

Charter's Spectrum brand is rolling out NVIDIA-powered edge compute across 1,000+ sites following Cox acquisition. Spectrum now serves 70+ million locations with edge hubs delivering <10ms latency to 500 million devices.

Why it matters 1,000+ edge compute sites at scale require operators to reassess edge QoS SLAs, backhaul capacity planning, and cloud-to-edge orchestration architecture for distributed GPU inference.

Charter's Spectrum brand is deploying NVIDIA-powered edge compute at over 1,000 sites across its merged footprint, enabling support for robotics, low-latency AI inference, and other latency-sensitive workloads at the edge. Spectrum positions hundreds of megawatts of power from more than 1,000 edge data centers and hubs that are less than 10 milliseconds removed from over 500 million devices in homes and businesses. The scale is notable: Charter's acquisition of Cox's cable assets brought over 11 million former Cox homes and businesses online, expanding Spectrum's total footprint beyond 70 million locations. Charter's virtualization of CMTS paves the way for this work, indicating that centralized packet processing and edge disaggregation are prerequisites for distributed edge GPU deployment. For operators, this deployment model suggests a shift toward treating edge hubs as distributed compute-on-demand platforms, requiring updates to fiber interconnect SLAs, backhaul capacity planning, and edge-to-cloud orchestration policies.

Read the original at timothysykes.com ↗

New RFC explains how to configure ECN on UDP sockets

RFC Editor / IETF · Oct 14, 2026 · Primary source

RFC 10055 provides guidance on configuring UDP sockets for Explicit Congestion Notification (ECN) across common platforms, published as an informational RFC in October 2026. The document addresses practical implementation considerations for ECN deployment in UDP-based applications.

Why it matters Practitioners deploying ECN for congestion management in real-time and delay-sensitive UDP applications need portable, cross-platform socket configuration patterns; this RFC codifies best practices across Linux, Windows, and BSD systems.

RFC 10055, authored by M. Duke, standardizes ECN socket configuration patterns across operating systems. ECN is increasingly critical for modern networking infrastructure, particularly in data center and WAN scenarios where congestion signals enable traffic management without packet loss. However, implementation varies significantly across platforms and network stacks. This RFC reduces deployment friction and improves interoperability. For network operators deploying congestion-aware applications and AIOps systems relying on fast feedback signals, consistent ECN implementation improves observability of network state and enables better demand-responsive traffic engineering.

Read the original at datatracker.ietf.org ↗

JPMorgan bars outside AI agents from customer accounts

FourWeekMBA · Oct 18, 2026 · Analysis

Anthropic reported third-quarter revenue of $16.8 billion, up 45% year-over-year. JPMorgan Chase CEO Jamie Dimon announced the bank will not allow outside AI agents to access customer accounts and operate on their behalf.

Why it matters Enterprise adoption is bifurcating—accelerating capex on AI infrastructure clashes with risk-averse policies on autonomous agent access to critical financial systems.

During the week of October 12–18, 2026, two trends emerged: companies selling into AI infrastructure reported strong growth, with Anthropic's Q3 revenue rising 45% to $16.8 billion, while TSMC posted Q3 revenue of $46.72 billion with a 66.4% gross margin. This reflects sustained enterprise demand for AI model capacity and semiconductor infrastructure. However, several major institutions drew boundaries around AI agent autonomy—JPMorgan Chase announced it will not permit outside AI agents to access customer accounts and operate independently on their behalf. This reveals a governance divergence: while AI spending accelerates at the infrastructure layer, financial services and critical operations are implementing explicit restrictions on agentic AI access, signaling regulatory caution and fiduciary risk concerns even as model providers report record revenue growth.

Read the original at fourweekmba.com ↗
Nothing in today's briefing matches that.

Listening

Podcasts and talks
  • Packet Pushers (Network Automation Nerds podcast)

    NAN133: Beyond the Hype: The Realities of AI Infrastructure

    Eric Chou interviews Leo Fleskes on infrastructure challenges surrounding AI resource consumption: massive water and power demands by data centers, local community pushback, regulatory constraints, and economic viability of AI provider expansion.

Vendor Radar

Last 7 days · arrows compare with the 7 before

What changed this week

Last 7 days vs the 7 before

Biggest moves

Trending topics

agent · observability · MCP · automation · OpenTelemetry · Agentic AI · Kubernetes · AIOps · RAG · inference

Get the daily briefing

Today's 3 things that matter and every story with why it matters, in your inbox each morning. Free, and you can unsubscribe at any time. Prefer a reader? Follow the RSS feed.