March 9, 2026

Hello World. A new AI networking startup just emerged from stealth, Meta and Microsoft are rewriting how GPUs talk to each other, and Google is closing one of the largest cybersecurity acquisitions in history. As always, thank you for reading - and hit us up with any feedback.

Today’s edition:

Let’s dive in.

Eridu exited stealth with $200M in Series A funding to build a clean-sheet AI networking switch, founded by Drew Perkins, who previously co-founded Infinera and Lightera. The core argument is that existing switch architectures were designed for cloud workloads, not AI, and incremental improvements won't close the gap — Eridu's high-radix switch claims to replace 30 lower-radix switches in a single unit, cutting networking tiers, power consumption, and latency for large GPU clusters.

Datalec launched a modular data center solution that cuts deployment timelines from 16 months to 10, primarily by compressing the design phase from six months to two. The approach moves more of the build offsite into a controlled manufacturing environment, which reduces onsite construction time and lets colocation providers and AI infrastructure teams add capacity in smaller increments rather than committing to full buildouts upfront.

Cisco released its first semiannual firewall update of 2026, covering 48 CVEs across 25 advisories for its Secure Firewall Management Center, ASA, and FTD product lines; the largest single firewall patching workload the company has issued. If you're running FMC in your environment, this one warrants prioritization before the next maintenance window.

Meta and Microsoft co-authored a new Ethernet specification through the Open Compute Project that redefines how GPUs communicate at scale inside AI data centers. The Ethernet for Scale-Up Networking (ESUN) 1.0 spec replaces the standard IP/UDP header stack, typically 28-48 bytes, with a compact 4-byte header designed specifically for scale-up GPU traffic, reducing protocol overhead, improving throughput efficiency, and lowering latency across large GPU domains.

Between the lines: The significance here is who is writing the spec and where. Meta and Microsoft publishing through OCP, with Broadcom, Cisco, HPE, NVIDIA, and OpenAI in the working group, is the industry's clearest signal yet that Ethernet is being purpose-built to displace InfiniBand in AI back-end networks. InfiniBand held roughly 80% of AI back-end networking share as recently as 2023. Dell'Oro now projects Ethernet-based offerings will drive nearly $80 billion in data center switch sales over the next five years. ESUN is the standards layer that makes that transition coherent rather than a collection of vendor-specific implementations. The 175-company working group is not a rubber stamp, it's a market consolidation in progress.

The broader read: ESUN is also an interesting sovereignty play. Replacing proprietary interconnects with an open, Ethernet-based standard means AI infrastructure becomes buildable by any vendor with Ethernet expertise, rather than dependent on a single supplier's ecosystem. That matters for sovereign cloud operators, neoclouds, and large enterprises that want to own their AI fabric without being locked into one vendor's hardware roadmap.

Read more about this initiative on the Open Compute Project

AT&T and Cisco have commercialized an IoT platform built directly into AT&T's 5G standalone core, moving IoT connectivity beyond the low-bandwidth use cases 5G was initially associated with. The integration connects Cisco's IoT Control Center and Converged Core to AT&T's nationwide 5G SA network, enabling network slicing, application-aware traffic handling, and stronger uplink performance for workloads like video surveillance and connected vehicles.

Why this matters now: 5G SA's architecture changes what's actually possible for IoT. Symmetric throughput and network slicing let operators match specific classes of devices to dedicated network resources, rather than sharing capacity across everything on the same pipe. For enterprise teams managing large IoT deployments, that's a meaningful operational shift from how connectivity has worked historically.

The open question: The real test is how widely this gets adopted outside of flagship use cases. The platform supports future RedCap and eRedCap capabilities aimed at lower-power devices, which suggests the architecture is designed to scale across device classes over time. Whether enterprise buyers actually deploy against it, or treat it as another carrier IoT pitch, depends on how the economics shake out at scale. Learn more about this here.

Learn more about this here.

💰 Google announced the completion of its acquisition of Wiz. Wiz will join Google Cloud and maintain its brand (Google)

💤 HPE CEO Antonio Neri said the vendor expects the current memory supply chain bottleneck will continue into next year, but that customers are getting creative in adapting to its implications. (SDxCentral)

🌐 Cato Networks releases Dynamic Protection, an AI engine designed to proactively stop stealthy, multi-stage attacks. (Channel Insider)

Catch up