Building end-to-end enterprise data center networks

Networks are the backbone of modern enterprise, yet we’ve been provisioning them the same way for the past 40 years. IT teams still need to stitch together networks from a fragmented market of hardware, software, installers, ISPs, and support providers, which makes it an ongoing struggle to keep those networks safe, performant, and reliable.

We built Meter to change that. By vertically integrating the entire networking stack—from hardware and software to deployment and maintenance—we deliver the network as a single managed service, making it simple for any organization to connect to, and work on, the internet.

With support from Sequoia, General Catalyst, Sam Altman, and Microsoft, we’ve spent the past decade deploying these end-to-end enterprise networks at scale. Now, we’re assembling a lean team of experts to extend that model to data centers.

Much like we provision networks for LAN customers who come to us and say, “I have X square feet, give me a network,” we want data center customers to come to us and say, “I have X GPUs or Y servers, give me a network.”

Why enterprise data centers?

Data center networks are some of the most operationally complex and capital intensive network deployments, and they’re the deployments that businesses can’t defer in today’s market. It’s very difficult to stand up a current-generation GPU cluster on a legacy fabric, for instance, and the longer you wait for your network to come online, the faster your GPUs depreciate.

Hyperscalers have hired massive teams to manage these deployments, and the networking incumbents are focused on serving this handful of companies. That leaves the small IT teams at thousands of organizations across the US without the technical and operational support they need to simultaneously maintain, upgrade, and stand up new fabrics. As fewer network engineers enter the field each year, those already-strained teams are stretched even further.

We’re building end-to-end data center networks for these customers: the enterprises who need the performance of a hyperscaler’s network without the hyperscaler overhead.

How will we serve the market?

Because these organizations chiefly need help upgrading their networks and deploying their GPUs, we’re building net-new, high-capacity hardware that we then deliver through Meter’s end-to-end operating model.

Specifically, we’re extending our hardware, firmware, and infrastructure platforms to include Ethernet top-of-rack and spine switching.

  • Our TOR switches will have 48x 50G downlink ports, 6x 400G uplink ports, and dual PSUs.
  • Our spine switches will have 32x 400G ports and dual PSUs.
  • Both switches will have buffer capacity designed for bursty data center traffic. Together, they’ll offer very high bandwidth across the entire data center.

Coupled with Meter’s end-to-end network delivery, customers get:

  • One accountable operator. There’s one call to make when something goes south, and it’s to the same team that designed the hardware, wrote the OS and monitoring software, and handled the deployment.
  • Faster, cheaper deployments. Our integrated supply chain and deployment operations reduce lead times, vendor coordination, and costs. Our pricing is tied to power consumption, which is typically how businesses buy data center capacity today.
  • Simpler lifecycle management. Meter’s deployment best practices are built into our fabric designs, hardware refreshes keep networks current, and proactive monitoring of the entire stack identifies and remediates issues before they escalate into outages. Learn more about our approach to automation at meter.ai.

Where do you fit in?

We know which market we want to serve and what hardware we want to build, but we need opinionated builders to shade in the specifics with us. This is your chance to reshape how a significant slice of American businesses connect to critical infrastructure.

Whereas the hyperscalers have built massive teams to work on a small subset of problems for an even smaller subset of customers, you’ll work with a wide range of businesses to build our approach to data centers from the ground up. How should we lay out the SerDes lanes? How should our control plane scale with the permutations of fabrics? Should we consider Ultra Ethernet?

We’re designing this team to be as lean and agile as possible. We work in our own OS, not SONiC; make decisions and iterate quickly; and trust you to identify and solve the problems you think will matter most to our customers. We intentionally have a small team of top contributors from Cisco, Anthropic, McKinsey, Stanford, and MIT. We think you’ll do the best work of your career here.

So, if you:

  • Have worked on hardware, firmware, product, data systems, QA, or support for a low-latency, high-uptime hardware platform.
  • Want to work directly with our CTO.
  • Decide what problems we solve, how we solve them, and see them deployed within a few quarters.
  • Expect a lot from your colleagues and want them to expect a lot from you.

Send us a note at dc@meter.com.

We’re also hosting an event at our Sunnyvale office on September 10th (details here). If you’ve read to this point, we’d love to see you there.