Help us build autonomous networks

The internet (unsurprisingly) runs the world.

Data centers and enterprises, colleges and manufacturing facilities – they spend billions annually on computer networking to stay productive.

What’s surprising is how messy and brittle these networks are. Decades have gone by without much innovation from large incumbents. Complex environments of all types are left struggling to keep their networks safe, performant, and reliable.

With support from Sequoia, General Catalyst, Sam Altman, and Microsoft (more on our Series C here), Meter has spent the last decade deploying networks at scale. Building on our traction, we’re assembling a small, focused research team to make networks shockingly great by making them autonomous end-to-end.

We’re achieving autonomy by leveraging our rare data advantage, control over the full networking stack, and tens of thousands of H100s. All while using our size to our advantage – eliminating bloat and aligning incentives to build extraordinary products.

As you continue reading, you may be surprised by the scale of the problem, how few people are working on it, and why Meter is uniquely positioned to solve it.

We believe you need a rare combination of a few ingredients to build great, model-driven products:

An important problem to solve
Control of the stack
Diverse, clean, user-relevant data
Scalable compute
A world-class team

Problem

Networks move every byte processed by a computer. Networks enable everything we do.

But, networking is a mess. Network engineers spend their days SSH-ing into routers, eyeballing logs, and tweaking configs by hand. When a link fails, they scramble to reroute traffic. When a firewall rule breaks something, they guess and check until it works again. Things grind to a halt because someone fat-fingered a VLAN setting or a fiber line got cut.

Network engineers spend years mastering this complexity. While much of Silicon Valley is focused on providing leverage to software engineers, network engineering is quietly one of the largest job categories in technology.

This issue becomes even more pernicious when considering that internet usage and data center traffic is growing exponentially.

Control

For the last few decades, networking companies only built a piece of the puzzle—either hardware, software, or operations—but never all three. They’ve grown their product offerings through acquisitions. That means they’re stuck trying to glue together legacy systems with no real control or unified data. Thus, they cannot make decisions based on data across the stack, and then also actuate those decisions.

At Meter, we own the full stack. We design and build the hardware for networks, including power, routing, switches, security, and wireless. We write the firmware, operating systems, and distributed software that run on them. We then deploy, operate, and monitor these networks at scale.

Control allows us to push networking forward, like when we released Command—a generative UI that lets users manage networks and build custom dashboards in plain English. Command writes functions against our backend with a custom model – the same model acting as the actuator into our networks to create changes, and retrieve more information.

Control will also let us do what no one else can: train a model that deeply understands real-world networks, predicts failures before they happen, and fixes problems before engineers even know there’s an issue.

Data

Models are only as good as their data. Large, clean, and diverse datasets from real-world customer use are essential. We operate the entire network stack across hundreds of networks, allowing us to see everything anonymized:

Network Designs – Floor plans, topology decisions, and initial configurations.
Configuration Changes – Every firewall rule, every firmware update, every policy tweak.
Operational Telemetry – Latency, packet loss, throughput, congestion patterns.
Failures & Fixes – Every single support ticket, correlated with network state before and after.

Other companies may have some logs or monitoring data. We have complete visibility into every network’s lifecycle, from deployment to daily operation to long-term evolution. This lets us train models that actually learn how networks work rather than just reacting to anomalies. It lets us continuously evaluate autonomous systems on actual customer pain points.


        Anomaly Detected:    [↑] Errors   [↑] Latency   [↑] Drops       
 ───────────────────────────────────┬───────────────────────────────────
                                    │
                                    ▽
                                    │
         ┌─────────────────┬────────┴────────┬─────────────────┐       
         │                 │                 │                 │       
         ▽                 ▽                 ▽                 ▽       
     Routing          Switching            Wi─Fi           Security    
     analysis          analysis          analysis          analysis    
         │                 │                 │                 │       
         │                 │                 │                 │       
         │                 │                 │                 │       
         ▽                 ▽                 ▽                 ▽       
 ┌────Review ────┐ ┌──Verify VLAN──┐ ┌──Examine RF───┐ ┌─ Audit ACLs,──┐
 │    routing    │ │    & trunk    │ │  parameters   │ │    firewall   │
 │   protocols   │ │configurations │ │               │ │  Logs, & IDS  │
 │               │ │   on switch   │ │               │ │               │
 │   OSPF, BGP,  │ │     ports     │ │               │ │               │
 │      etc      │ │               │ │               │ │               │
 │               │ │               │ │               │ │               │
 │     Route     │ │     Port      │ │Channels, power│ │               │
 └─────flaps?────┘ └────errors?────┘ └─interference──┘ └──Anomalies?───┘
         ▽                 ▽                 ▽                 ▽        
         │                 │                 │                 │        
         ▽                 ▽                 ▽                 ▽        
 ┌───[ ]─Yes─────┐ ┌──[ ]─Yes──────┐ ┌──[ ]─Yes──────┐ ┌──[ ] Yes──────┐
 │       │       │ │       │       │ │       │       │ │       │       │
 │       ▽       │ │       ▽       │ │       ▽       │ │       ▽       │
 └───Recompute───┘ └──Reconfigure──┘ └─────Tune──────┘ └─────Patch─────┘
         │                 │                 │                 │        
         ▽                 ▽                 ▽                 ▽
         │                 │                 │                 │
         └─────────────────┴─────────────────┴─────────────────┘
                                    │
                                    ▽
                                    │
 ┌──────Validate end─to─end QoS metrics, congestion, jitter, etc───────┐
 │                                  ▽                                  │
 └────Correlate with real─time telemetry and historical analytics──────┘

Compute

Today’s AI breakthroughs aren’t just about better architectures; they are about brute-force scale. You need serious compute to build serious models. Through our partnership with Microsoft, we have access to tens of thousands of H100s, translating to thousands of H100s per engineer.

We are training models with billions of parameters across text-based configurations, time-series telemetry, packet flows, and structured networking data. This is not generic AI; we’re building something purpose-built for real-world problems.

Team

At Meter, we believe the best products are built at the intersection of applied research and obsessive product development. We operate networks in the real world, where mistakes carry real costs, and the stakes are high.

We’re assembling a small team of talented researchers who want to apply themselves to a critical yet neglected industry. It’ll be ambitious, demanding work, but if you want to build models that create an outsized, differentiated impact, come speak with us.