Plumbing the
information age

NETSCOUT's position in the 2026 GigaOm Radar reflects a critical market consolidation theme: end-to-end visibility platforms with native AI are replacing point solutions in network observability. As agentic AI workloads introduce new traffic patterns and performance requirements, enterprises are actively consolidating around platforms that combine mature network telemetry collection with advanced AI analytics. NETSCOUT's emphasis on maturity and platform play functionality suggests the analyst firm is rewarding vendors who can service both traditional network operations (fault, performance, configuration) and emerging agentic NetOps requirements in a single integrated system. This is relevant for NetDevOps leaders evaluating whether to migrate from legacy NPM tools to modern AI-integrated platforms. The radar published September 23, 2026 comes as enterprises face decision fatigue between specialized tools like Selector.ai and unified platforms—NETSCOUT's radar position validates the unified approach for complex enterprise environments.

NetOps teams face rising performance expectations despite increasing complexity from applications, devices, cloud dependencies, and security controls—and AI workloads are adding new traffic patterns and operational complexity. The findings show organizations have moved past advisory AI; over half now run agentic AI systems that act in production today. Agentic AI for NetOps represents a paradigm shift from running operations to orchestrating intent, requiring trust built on visibility into every decision, explainable context, and guardrails that ensure deterministic outcomes. At current volumes, the average organization would need roughly 100 IT specialists to clear its daily network alert backlog by hand. For SRE and NetOps leads, this research provides quantitative validation that agentic operations are operational necessity, not experiment, with clear guidance on trust, observability, and control requirements.

The research, conducted independently by Omdia, surveyed 1,000 IT and network operations leaders at organizations with 500+ employees. Key finding: AI traffic is doubling every six months, and agentic AI tasks can generate 450% more network traffic than baseline. The market is moving from advisory AI to AgenticOps—agents that sense, reason, and act in production—but with unanimous requirements for trust infrastructure. Over two-thirds demand detailed explainability for agent-driven actions, and 86% say a single integrated platform (not point tools) is essential. Organizations are already allowing agents to make production changes including rerouting traffic, adjusting wireless parameters, isolating endpoints, and resolving incidents end-to-end without prior human approval. This represents the fastest adoption of autonomous infrastructure operations Cisco has measured.

Unit 42's new service represents a shift from point-in-time offensive testing to continuous, agentic vulnerability hunting. The platform treats frontier models as autonomous security agents capable of reconnaissance, exploitation planning, and exposure validation at machine speed. Threat actors are now compressing breach windows by ~97%—from weeks to hours—making static pentesting insufficient. Unit 42 Continuous Frontier AI Defense cycles through four phases: finding exposures, validating exploitability, pinpointing attack paths, and accelerating remediation workflows. This builds on Unit 42's April 2026 Frontier AI Defense offering, which provided benchmarking and security blueprints; the new continuous model enables persistent, automated exposure hunting. For security operations practitioners, the key implication is that agentic tools are moving from detection/response into proactive threat hunting, shifting the operational burden from reactive incident response to asset inventory validation and patch acceleration. The service requires enterprises to govern which models operators can deploy and validate before executing remediation actions—a significant change in how offensive tools are managed.

D&H Distributing, a leading technology distributor, has extended its U.S. distribution partnership with HPE, gaining access to the HPE Networking portfolio including HPE Juniper Networking routers and switches. Under the agreement, D&H and HPE will make the addition of HPE Juniper Networking routers and switches fully available to partners by the end of September 2026. This represents a significant inflection point for HPE's post-acquisition integration of Juniper Networks—converting it from a $14B acquisition into an operationalized go-to-market engine through established channel partners.

The integration adds HPE Networking to D&H's Advanced Solutions+ ecosystem, forging opportunities for U.S. channel partners to access HPE's networking, cloud, and AI portfolios along with D&H's signature training, enablement, and 24/7 personalized support, with HPE partners continuing standard operating procedures during the Juniper onboarding phase. For infrastructure practitioners, this signals that Juniper switching and routing—increasingly critical to AI cluster fabric—is now flowing through mainstream distribution channels rather than remaining siloed. This accelerates adoption timelines and enables smaller-scale operators to procure enterprise-grade AI networking fabric alongside HPE compute and storage.

Marvell's first live demonstration of 2nm 800G ZR/ZR+ with MACsec, powered by the Marvell Libra 2nm coherent DSP, expands the company's coherent DSP and COLORZ pluggable portfolios with media access control security, delivering efficient, high-performance and secure optical transmission for hyperscale AI and cloud data center networks worldwide. First 2nm 1.6T ZR and 1.6T coherent-lite O-band demonstrations, powered by next-generation Marvell DSPs, are optimized to deliver high-bandwidth connectivity within and between data center facilities and pave the way for 3.2T connectivity.

Marvell's advanced co-packaged optics (CPO) technology integrates optics directly with AI accelerators and switch silicon as per-lane speeds and multi-rack AI pods push past copper's reach limits. These demonstrations are critical for practitioners because they represent the optical DSP roadmap that will enable 1.6T and beyond deployments starting in 2027. The 2nm node translates to improved power efficiency and signal integrity—direct impact on operational costs in multi-exabyte-scale training clusters where thermal and power budgets are zero-sum games.

Snorkel AI's Series E, led by Insight Partners and S32, marks a dramatic valuation jump from $1.3B (May 2025) to $3.5B. CEO Alex Ratner frames this as the "Data 2.0" transition: frontier models no longer need only scale and labeled examples; they require carefully designed, difficult tasks, precise scoring systems, and practice environments. Snorkel's human-and-AI data factory combines subject-matter experts, specialized agents for routing and quality checks, and human feedback loops. The company reports that AI-assisted coding agents improved quality-control efficiency by >50% and review accuracy by 15+ points. Revenue crossed $375M annualized run rate in the announcement week, driven by partnerships with frontier labs, hyperscalers, and U.S. government agencies. Snorkel also announced $3M in commitments to its Open Benchmarks Grants program for independent AI benchmarks and evaluation artifacts. The shift from software to services fundamentally changes how investors value data infrastructure—from tooling to product supply.

Mirendil's founding team brings deep expertise from major labs: Neyshabur (ex-Anthropic, ex-Google DeepMind, led scientific AI reasoning teams) and Mehta (ex-Anthropic, senior research scientist). They're joined by Shayan Salehian (ex-xAI, worked on post-training and agent infrastructure) and Tara Rezaei (MIT, ex-OpenAI intern). The core thesis is recursive self-improvement: creating AI systems that autonomously experiment, evaluate, and improve themselves without constant human guidance. Frontier labs already run this loop internally—it's central to their training workflows—but terms of service restrict outsider access. Mirendil's goal is to democratize this as a platform for specialized domains: drug discovery, chemistry, biology, robotics. The company plans to launch a frontier model for engineering and research work by early 2027. With >20 staff now (up from founding team in June), Mirendil is targeting the gap where domain labs need AI research capabilities but can't afford frontier labs.

For builders, Opus 5.5 delivers a 40% reduction in typical workload costs compared to Opus 5 and output speeds increased by over 30%, supported by pricing of $4/$20 per million tokens and cache reads at $0.20 per million tokens. In practical applications, Opus 5.5 completed a legacy HAProxy port from C to Rust in 9.5 hours, a 51% improvement over the 12 hours required by Fable 5.1. On Anthropic's internal automated behavioral audit (its most extensive alignment test suite), Opus 5.5 is the best-scoring model released. The model's benchmarks largely exceed Fable 5.1 on agentic tasks despite costing 150% less over API. Opus 5.5 is available on AWS, Google Cloud, and Microsoft Azure, with Sonnet 5.5 and Haiku 5.5 arriving within weeks. The release, occurring just 10 days after CEO Dario Amodei's call to "pace the frontier," reflects competitive pressure as frontier labs shift focus from raw capability to efficiency and cost-per-task metrics.

This is a distribution architecture that establishes a price floor by seeding the consumer base with Luna, which scores 66.6% on DeepSWE at near-zero cost, forcing competitors to justify charging at all. The cost of entry for building on OpenAI's infrastructure dropped to zero for the lowest tier. On software engineering (DeepSWE), Sol scored 68.8% against Claude Fable 5's 69.9% at roughly 80% lower per-task cost. For builders evaluating model routing, the three-tier family enables single API integration covering lightweight extraction (Luna), production coding (Sol), and peak reasoning (Astra) without switching providers, reducing evaluation and migration overhead. The timing—hours after Anthropic's Opus 5.5 release—signals an industry shift toward efficiency-based competition measured in useful work per dollar rather than benchmark scores alone. OpenAI confirmed these are permanent prices, not promotional.

ShareGate's State of M365 report released September 23 found that 38% of organizations failed to remove former employee or guest access, 35% experienced audit or compliance gaps, and 26% had sensitive content misrouted, with 77% experiencing at least one governance incident overall, while two-thirds deployed multiple AI tools on the same content. Only 1% of organizations use purpose-built governance tools, leaving most to absorb compliance and security exposure they cannot measure or control. This data point captures the core 2026 challenge: AI deployment velocity—particularly multi-model stacks—has completely decoupled from governance maturity. The implications are material for any enterprise running generative AI in regulated environments or handling sensitive data, since control gaps now appear structural rather than transitional.