NB578: Cisco Goes All-In on AI Ops with Cloud Control; China Floats Underwater Data Center
Network Break episode 578 covers Cisco Live US 2026 announcements including the ambitious Cloud Control AIOps platform, a Red Hat NPM supply chain compromise, Microsoft's Majorana 2 quantum chip, underwater data centers in China, and financial results from Broadcom and HPE showing massive AI-driven growth.
Summary
The episode opens with a security red alert for Android CVE-2026-0072, a critical 10.0 CVSS vulnerability allowing local privilege escalation to root with no user interaction required, urging listeners to patch immediately.
The bulk of the episode covers Cisco Live US 2026 announcements. The headlining product is Cloud Control, an AIOps platform designed to unify telemetry from across Cisco's entire portfolio—networking, security, compute, observability, and collaboration—into a single control plane where human engineers and AI agents collaborate. It incorporates Cisco's Deep Network Model and FoundationSEC, supports third-party frontier models, integrates via APIs and MCP with AWS, Microsoft, Google Cloud, ServiceNow, PagerDuty, and Slack, and features an 'AI Canvas' workspace for natural language queries. The hosts flag significant concerns about data quality normalization, attack surface expansion through agent integrations, and the complexity of governing permissions across a massive multi-vendor ecosystem. Cisco also highlighted LiveProtect, an eBPF-based feature for NXOS that can block known exploits before patches are available, which the hosts view favorably as a compensating control inspired by the Mythos vulnerability research. Cisco also pledged quantum-safe communications across most of its core portfolio by December 2026, with Chuck Robbins personally mentioning quantum in his keynote.
A Red Hat supply chain compromise is discussed, where a Red Hat employee's GitHub account was compromised, allowing malicious code to be injected into 96 versions across 32 official NPM packages. The worm steals credentials including GitHub Action secrets, NPM tokens, Kubernetes and Vault material, and cloud service credentials. The hosts note the irony of this occurring days after IBM announced its $5 billion Project Lightwell open source security initiative.
On the data center front, China has launched the first commercial underwater data center near Shanghai, housing nearly 2,000 servers at 35 meters depth using passive seawater cooling, achieving a PUE below 1.15 compared to the industry average of 1.5. The hosts reference Microsoft's Project Natick as prior art but note Microsoft has stepped away from underwater data centers. AWS's new Resilient Network Graphs (RNG) topology design is also covered, which uses pseudo-random routing inspired by random graph theory and a cable-shuffling robot, achieving a 69% reduction in network devices, 40% power reduction, and 33% throughput improvement.
Marvell unveiled its Teralink T100 switching silicon at 102.4 Tbps using a 3nm process, claiming 25% lower power consumption versus competitors, targeting large-scale AI cluster workloads.
On quantum computing, Microsoft announced Majorana 2, its next-generation topological quantum chip developed with agentic AI assistance, featuring a thousandfold reliability improvement over the prior generation, 20-second mean qubit lifetimes, and cutting its scalable quantum computer timeline to 2029—half the original estimate. Quantinuum also completed a $1.6 billion IPO implying a $14-15 billion valuation, which the hosts describe as a bellwether for the quantum industry given the company's full-stack approach and actual paying customers.
Financial results close out the episode. Broadcom reported record Q2 2026 revenue of $22 billion, up 48% YoY, with AI semiconductor revenue alone at $10.8 billion, up 143% YoY. HPE reported $10.7 billion in revenue, up 40% YoY, swinging from a $1 billion loss to $624 million net income, driven by cloud and AI infrastructure. The hosts note HPE remains significantly smaller than Dell in server and storage revenue. A brief mention of Muon Space's new Condor Ultra satellite platform for mass constellation deployment via SpaceX Starship closes the show.
Key Insights
- The hosts argue that Cisco Cloud Control introduces a massive new attack surface by combining AI agents, MCP integrations, third-party data sources, and multi-vendor telemetry into one platform, comparing it to a 'giant hairball of a security vulnerability' even if Cisco tests it rigorously.
- Jonna argues that data quality management is the foundational blocker for ambitious AIOps initiatives, claiming enterprises pursuing AI-driven network operations without a data quality center of excellence should 'put the brakes on' first.
- The hosts contend that Cisco LiveProtect's eBPF-based exploit blocking directly addresses the median total time to contain metric in cybersecurity, functioning as a 'ball jar over the roach'—containment without remediation—which they view positively despite concerns about stopgap measures becoming permanent.
- Jonna argues that the Mythos vulnerability research program is actively driving the industry to rethink cybersecurity operations, and that the volume of vulnerabilities and patches hitting organizations will increase dramatically in the near term.
- The hosts note the ironic timing of Red Hat's NPM supply chain compromise occurring days after IBM announced its $5 billion Project Lightwell open source security initiative, with Jonna suggesting it almost functions as inadvertent marketing for why such a service is needed.
- Jonna claims that quantum computing represents a more significant paradigm shift than AI, arguing that when the history of this era is written, LLMs will be seen primarily as a footnote that accelerated quantum computing development rather than a revolution in their own right.
- Microsoft's Majorana 2 announcement cuts its scalable quantum computer timeline from roughly six years to 2029—three years away—which the hosts argue makes post-quantum cryptography preparation urgent and non-deferrable for enterprises currently distracted by AI initiatives.
- AWS's RNG topology, which uses pseudo-random routing and a physical cable-shuffling robot to rewire connections dynamically, achieved a 69% reduction in network devices and 40% power reduction in production deployments, suggesting fixed hierarchical topologies may not be optimal for large-scale AI data centers.
Topics
Full transcript available for MurmurCast members
Sign Up to Access