Plumbing the
information age

The Infrastructure Automation Report surveyed IT decision-makers and platform engineering leaders on AI governance maturity. The findings expose a significant confidence-competence gap that becomes critical as AIOps and autonomous IT operations scale. While infrastructure teams claim governance capability, the absence of formal policies means most organizations lack documented controls, audit trails, and escalation procedures — precisely what's needed when agents perform high-stakes operations like incident remediation or security response. For network and SRE practitioners deploying agentic AIOps platforms (LogicMonitor, ServiceNow, Dynatrace), this underscores the need to establish governance frameworks before expanding agent autonomy. The gap also reflects industry-wide immaturity: organizations are moving faster on agent deployment than on the operational controls required to run them safely. Teams should use this data to trigger formal AI governance policy development and map existing identity, secrets, and observability tools to support task-level auditability before expanding automated remediation workflows.

DuckDB will remain open source under its independent foundation and MIT license. Over time AWS plans to combine its speed at everyday queries with the enterprise scale of services like Amazon S3, Amazon Redshift, and Amazon Athena. The Agent Toolkit for AWS in the AWS CLI equips AI coding agents like Kiro, Claude Code, Codex, and Cursor with curated, up-to-date AWS knowledge and a secure connection to thousands of AWS APIs through the AWS MCP Server. For AI builders using these agents, this helps them choose the right services, use modern APIs, and follow security best practices to get AWS code right more often the first time. This represents the practical integration of MCP into cloud infrastructure operations, enabling agent-driven infrastructure automation with access to proven analytical capabilities.

ServiceNow's August 28 announcement represents a significant consolidation of its autonomous security platform, moving beyond point solutions toward integrated workflows that span the entire security operations lifecycle. The six unified solutions target multiple operational domains: unified exposure management for continuous visibility, continuous vulnerability detection with AI-assisted prioritization, cyber-physical security integration, identity and access security, agentic incident response automation, and cyber risk and compliance management. This reflects the industry shift toward integrated detection-to-response platforms where AI agents operate within predefined boundaries set by security teams. The agentic incident response component is particularly notable for organizations struggling with alert fatigue—enabling AI to autonomously investigate, correlate, and recommend responses while humans retain override capability. The bundling approach simplifies procurement and reduces integration overhead compared to best-of-breed point solutions, though practitioners should evaluate whether consolidated tooling sacrifices depth in any single functional area.

SK Telecom's decision to spin off SK Broadband's data center assets into SK Horizon, alongside its SK Hyper AI unit, signals that hyperscale AI infrastructure is no longer a side business but a core operator revenue stream, directly influencing how operators plan 5G/6G core, edge, and transport investments over the next decade. Network architects must design for spine-leaf fabrics extending beyond data halls into metro rings with deterministic latency and optical-first traffic patterns that assume AI-heavy workloads, moving away from traditional enterprise-first network design. Lumen's extension of its cloud networking fabric to roughly 10 million US business locations demonstrates that enterprise appetite for managed, programmable connectivity into cloud and AI services is growing, positioning operators to capture spend through Network-as-a-Service models with APIs for QoS, routing, and security. This represents a structural reorientation of telecom business and engineering strategy away from consumer-centric network design toward infrastructure-as-a-service models. For AIOps and infrastructure teams, this signals that the traditional separation between telco networks and cloud infrastructure is collapsing—operators are now building AI factories as primary revenue generators, which reshapes the competitive landscape and deployment models for network management, observability, and automation platforms.

The survey was commissioned between April 29 to May 25, 2026 and included responses from 650 respondents currently using AI agents; after removing low-quality responses, 554 respondents remained. This is the practitioner benchmark that matters: daily agent use is now the norm for 4 in 5 engineering teams, but the data confirms that AI agents have moved from experiments to daily tools for most surveyed engineering organizations, which raises expectations around reliability, observability, and governance. One group of engineering teams are using AI much more effectively; successful teams show a 70.8% leap in AI agent use with 80.8% using agents daily. For SREs and platform teams, the takeaway is stark: the infrastructure for reliable agent orchestration—state management, cost controls, observability—is now table-stakes, not optional. Teams that skip this layer face the classic reliability tax.

The survey was conducted between April 29 to May 25, 2026 and included responses from 650 respondents; after removing low-quality responses, 554 respondents remained. The most actionable finding: successful teams outperform by systematically addressing three infrastructure layers—state management, cost optimization, and reliability instrumentation. This isn't a model question (which LLM is best); it's an orchestration question. Successful teams are less worried and less stressed even while facing the same rate of errors; their success compounds as they use more tools, search more widely for help troubleshooting, and build more of their own developer applications, suggesting a 'great separation' between avid adopters and the rest. For SREs and platform leads, the implication is clear: the reliability gap is now the moat. Teams that invest in durable agent execution (via tools like Temporal's framework) early pull ahead; teams that try to retrofit reliability later face compounding technical debt.

The Model Hardware Standard stores device physical characteristics—weight, safety limits, adjustable parameters—previously locked in paper manuals or specialist tacit knowledge, and works alongside the existing Model Context Protocol (MCP) released in 2024. The standard is model-agnostic and can interface with any device exposing a programmable control surface. Lab benches and factory cells assembled from non-interoperating vendors typically require weeks to months of bespoke integration work; MHS reduces this to hours or minutes. Initially available through a gated research preview to select organizations in science, robotics, and manufacturing, with Anthropic intending to open-source it post-preview. A significant caveat: AI struggles with understanding physical cause and effect, requiring human expert oversight for safety-critical physical reasoning. Early partners span multiple industries, and Anthropic is actively building a silicon team to design custom chips, recently hiring hardware executive Caitlin Kalinowski from OpenAI, Meta, and Apple.

In August 2026, the venture capital market decisively reoriented from software-as-a-service toward physical infrastructure backing AI deployment. Andreessen Horowitz closed its Machine Age Fund at $1.1 billion, specifically targeting companies building chips, memory systems, networking hardware, storage, data centers, robotics, and connected home appliances. This represents a strategic shift in where capital sees durable returns: not in another layer of generic software, but in the physical backbone enabling AI at scale.

Major funding rounds bore out this thesis. Gatik, an autonomous freight company, raised $200 million as it scaled from pilots to commercial operations. Emerald AI raised $150 million at $1.05B valuation to make power-hungry AI data centers more flexible grid participants—addressing the electricity constraint on AI deployment. Stellaria, a Dubai-based startup combining satellite imagery and AI for geospatial intelligence (defense, maritime, infrastructure monitoring), raised $6.8M in seed funding, demonstrating investor appetite for AI applied to physical-world observation and decision-making.

For infrastructure and operations leaders: this capital rotation signals that the constraints on AI scaling have moved from compute availability to energy, grid flexibility, hardware supply chains, and logistics. Organizations building AI infrastructure strategies should expect venture and corporate capital to concentrate on solving these bottlenecks rather than incremental software features. The physical economy is becoming the limiting factor.

Omdia's semiconductor forecast reveals a structural shift in chip economics driven by AI infrastructure demand. Memory chips are projected to exceed 50% of total semiconductor sales in 2026, with AI data center ecosystems alone accounting for 36.5% of all semiconductor sales. By 2030, AI data center semiconductor demand is expected to surpass 53% of the market—a near-total concentration of chip production on AI workloads.

This demand concentration has cascading effects beyond data centers. Consumer electronics prices are rising due to memory chip scarcity and cost inflation driven by AI demand. This creates political vulnerability: the narrative that AI investment is inflating consumer prices will likely invite regulatory and tax policy discussions as 2027 approaches. Policymakers may increasingly view AI infrastructure spending as a public cost driver warranting intervention.

For capital planning and vendor strategy teams: expect continued memory chip price pressure and constrained supply through 2026-2027. The concentration of semiconductor capacity on AI creates both opportunity (if you are selling AI infrastructure) and risk (if you depend on commodity semiconductors for other products). Anticipate policy backlash around AI investment, computing cost, and energy consumption—these will likely become regulatory and tax considerations heading into the next election cycle.