The Edge Network AI Stack
scott belsky
@scottbelsky
+3.6%
since Sep 30
The idea
I believe the network layer, long seen as commodity plumbing, becomes the control point for enterprise AI as companies need to monitor and secure AI activity locally and as more compute moves to the edge. The basket owns the full stack: the equipment makers, the security and inspection software riding on the network, the edge AI silicon, and the traffic-visibility players that make it all observable. Weighting favors large-cap anchors with smaller mid-cap edge-compute bets, plus a 10% cash reserve, held with a five-plus year view rather than for any near-term catalyst.
What it owns
- Software and hardware that watches what flows through networks are natural checkpoints to govern and secure AI usage before it leaves or enters an organization. · Palo Alto NetworksIt builds tools companies use to see and govern what AI their employees actually use, providing the monitoring and security layer at the network level., FortinetIt ships security built into network hardware itself, the closest thing today to inspecting AI traffic at the point where it passes through the box., CloudflareIt runs security and compute at the network edge and is adding AI inspection and model hosting close to users, turning edge-level control of AI traffic into a service., ZscalerIt inspects company traffic, including AI usage, as data moves between users and the internet, providing coverage at the cloud-based inspection layer., Check Point SoftwareIt is the steady, profitable network security player providing inspection and control, and can extend its enterprise firewall base to govern AI traffic as that demand emerges.27%
- Chips that run AI models locally on devices and network gear reduce the need to send every inference back to a distant data center, shifting compute to the edge. · QualcommIt makes the chips that run AI models locally on devices, replacing the need to route every computation back to distant data centers., BroadcomIts chips sit inside most of the switches and routers in this basket, so if network gear becomes priced as the AI control point, it gets paid on nearly every device., Marvell TechnologyIt designs the networking and custom AI chips that move data through the equipment layer, benefiting from network upgrades driven by AI traffic growth., NXP SemiconductorsIt puts processing into embedded devices at the far edge of networks, from industrial gear to connected equipment, where local AI models would actually run., AmbarellaIts chips do AI vision processing on the device itself with no round trip to the cloud, a pure bet on compute staying at the edge instead of routing to distant servers., Lattice SemiconductorIt makes small, low-power programmable chips used for AI tasks right at the edge, a direct play on moving compute away from distant data centers.27%
- The routers and switches that carry AI traffic are becoming control points, not just pipes, so the companies that build and manage them stand to benefit from the shift. · Cisco SystemsIt installs and manages the most enterprise networks and is folding AI security and monitoring into its gear and software rather than ceding that control layer to cloud-only players., Hewlett Packard EnterpriseWith Juniper folded in, it sells the equipment that runs company networks and has the chance to position that hardware as where enterprises govern AI traffic, not just acquire boxes., Arista NetworksIt makes high-end switches that AI traffic runs through and is pushing from data centers into campus networks, exactly where local monitoring and control would live., Extreme NetworksIt sells cloud-managed campus networking that lets companies see and control AI usage on local networks and benefits if the repricing of network equipment for AI lifts the entire category., UbiquitiIt covers homes and small businesses with routers and access points, landing local AI compute and inspection where people actually manage their own networks.26%
- The systems that route, deliver, and observe traffic moving to and from edge compute make AI workloads observable and controllable at the application and network layers. · F5It inspects and manages application traffic, including the API calls that AI apps depend on, giving companies visibility into AI workloads at the application layer., NetScout SystemsIt does deep packet inspection that reads what moves across a network, the core of network-level visibility, and depends on enterprise demand for AI traffic awareness to grow., Akamai TechnologiesIt runs compute and security on servers close to users, a version of owning the edge, and is shifting revenue from content delivery to edge compute and security as AI workloads move outward.10%
- Cash10%