Plumbing the
information age

Skylar AI 2.5 addresses a practical constraint ops teams face: deploying advanced AI observability in security-sensitive and compliance-heavy environments. The update expands deployment flexibility for organizations requiring sovereignty, data residency, and strict audit trails—critical for government, financial services, and regulated infrastructure. Beyond deployment options, the release improves AI model accuracy for anomaly detection and correlation, platform performance for large-scale telemetry ingestion, and natural language interfaces so operators can query complex operational state without SQL or custom syntax. This matters for SRE and NetOps teams evaluating AIOps platforms because accuracy and query responsiveness directly determine whether correlation and root-cause analysis are actionable or noise. The native language interface also lowers the barrier for junior analysts to contribute to incident investigation without requiring deep domain expertise or coding.

Arista Networks' second-quarter 2026 earnings reflect the consolidation of Ethernet as the dominant protocol for AI data center networking. With Q2 revenue of $3.04 billion and raised Q3 guidance to $3.30 billion, management explicitly tied stronger outlook to accelerating AI-driven network deployments. This validates market data showing Ethernet switch sales more than doubled in Q1 2026, accounting for roughly two-thirds of AI cluster data center switch sales versus InfiniBand's declining share. Arista's competitive position rests on its 7060XE7 series, a portfolio of 1.6-terabits-per-second networking platforms designed for rack-scale AI infrastructure. The company has signed over 40 companies to an MSA (mutual support agreement) for its XPO (eXtra Dense Pluggable Optics) design, which claims to halve AI cluster footprints by integrating optical transceivers directly into switch form factors. Liquid-cooled variants ship in Q1 2027. For infrastructure teams, Arista's trajectory signals that high-radix, high-density Ethernet switching has become the baseline for new hyperscale AI deployments, with OpenConfig/SONiC software stacks and multi-vendor flexibility replacing proprietary scale-up interconnects as the preferred approach for both training and inference workloads at 100K+ GPU scale.

MAF 1.14.0 signals production maturity. Checkpoint resume is critical for production—enables workflow state persistence across restarts, essential for long-running agentic workloads. Mistral client addition reflects shift away from monolithic model choices; practitioners swap models without framework rewrites. Reliability fixes (approval, streaming) and new storage hooks indicate real-world deployments exposed edge cases. For MLOps and platform engineers, this means durable task semantics out-of-the-box via Durable Task extensions—no forced dependency on external orchestrators like Temporal. Convergence of AutoGen and Semantic Kernel into single MAF reduces fragmentation that plagued multiagent development through 2025. The framework now supports both .NET and Python with identical APIs, lowering polyglot friction.

Gemini 3.7 Flash incorporates developer feedback and algorithmic enhancements over 3.6 Flash. On Google's evaluations, DeepSWE v1.1 jumps from 49.0% to 65.3%, while FrontierCode 1.1 Main improves from 34.4% to 43.6%. For web development, the model achieves an Elo score of 1588 on Arena.ai's WebDev Arena versus 1538 for 3.6 Flash, with improvements in generating functional layouts and feature-complete applications in fewer prompts. Introductory pricing of $0.75 per million input tokens and $3.75 per million output tokens runs through end of 2026; starting January 1, 2027, prices double to $1.50 and $7.50 respectively. The model is positioned as Google's low-cost workhorse for agentic systems that plan tasks, use software tools, and execute multi-step workflows autonomously. Available through Google AI Studio, Android Studio, Google Antigravity, Gemini Enterprise Agent Platform, and rolling out in Gemini Spark for AI Pro/Ultra subscribers.

The Charter-Cox merger is now poised to close in mid-to-late August 2026, pending final administrative steps. Federal regulators (FCC and DOJ) approved the deal earlier in 2026, and 44 other states have already signed off. California was the final approval gate, with the Administrative Law Judge initially recommending stricter rural deployment and low-income service commitments than the settlement agreements the companies negotiated with consumer groups. However, the CPUC voted to approve on August 13. The combined entity will eventually operate under the Cox Communications corporate name but will rebrand former Cox territories under the Spectrum consumer brand over time. Charter CEO Chris Winfrey stated the company expects to close in mid-to-late August and shortly after launch Spectrum pricing and packaging in former Cox territory. From a network operations perspective, this consolidation is significant because it reduces operator fragmentation in North America's cable market and creates a single platform with enough scale to justify major AI-driven network optimization investments and unified NOC operations. The merger also accelerates investment in rural broadband and mobile convergence (Spectrum Mobile), both areas where unified capex planning drives efficiency gains.

Schulman's earnings call commentary reveals Verizon's strategic pivot toward AI-driven infrastructure monetization. The carrier currently has ~7.4 million legacy fiber connections across nine states plus D.C., with recent partnerships (including the Eaton Fiber deal with Tillman affiliates and the Starry fixed-wireless acquisition) designed to expand footprint outside core markets. The 32 million fiber passings target by year-end positions Verizon to compete directly with AT&T and Lumen for enterprise and hyperscaler connectivity contracts. Notably, Schulman signaled openness to "more partnerships, potential acquisitions, to speed the number of homes passed," indicating M&A appetite in a consolidating fiber market. For network operations teams, this matters because Verizon's fiber-first strategy combined with AI cost savings creates pressure on competitors to demonstrate equivalent opex reductions and service velocity. The multibillion-dollar AI revenue narrative—supported by the Google dark-fiber deal and similar anchor contracts—signals that telcos are transitioning from connectivity-as-commodity to infrastructure-as-a-platform business models. This requires unified, AI-driven network orchestration across multi-region fiber fabrics, not isolated optimization in legacy silos.

A few giant deals dominate the numbers, with most capital flowing to U.S. AI labs, compute, infrastructure, healthcare, legal tech, and regulated business tools. Investors are demanding evidence like paying users, narrow workflows, protected data rights, and repeatable sales paths, with the strongest opportunities sitting where AI helps people do costly work faster, with human review built in for legal, safety, or money-related tasks. Infrastructure rounds are now competing with frontier models for capital: Fireworks and Together are betting on inference efficiency and cost reduction, signaling that the market has shifted from bigger models at any cost to right-sized models that run profitably. For SRE and AIOps teams, this validates the business case for cost optimization and multi-model architectures; for enterprise builders, it means inference infrastructure startups are well-capitalized and moving fast.