Where are all the network engineers?

It's one of the quiet marvels of contemporary life that the internet—the infrastructure that shoulders the better part of human civilization—works at all. The protocols that carry modern medicine, global finance, and FaceTimes with grandparents were designed, through processes that more closely resembled long and occasionally acrimonious seminars than any kind of top-down engineering effort, for a few thousand machines.

That they now serve several billion isn't a testament to luck, but rather to the network engineers who've spent decades learning how to keep the internet working—who have mapped the eccentric topographies of systems that were never supposed to get this big, and who have developed, over years of patient attention, a sixth sense for when something is about to go wrong.

The problem is that those people are starting to leave, just as the internet is about to undergo its most dramatic transformation in history. So what does this mean for the network engineers who stay, and for the future of the internet?

Fermi's paradox for the internet age

When Enrico Fermi sat down for lunch with some colleagues at Los Alamos in 1950, he posed a question that's dogged physicists for decades. If the sheer scale of the universe makes intelligent life a near certainty, why don't we have any evidence of its existence?

Networking has stumbled into its own version of this paradox. Network usage has roughly doubled every few years, and it's slated to grow even faster as more devices come online and AI workloads start to scale. AI usage alone is expected to increase global data center capacity demand by 70%, and meeting that demand will require something on the order of $6.7T in new data center investment by 2030. Networking already accounts for roughly 10–14%¹ of the cost of building a data center, and every one of those facilities needs to be designed, configured, and maintained by someone who understands how the internet actually operates. In other words: everything is packets, and we're going to move a lot more of them over the next decade.

But roughly a quarter of the people who build the systems that move those packets are set to retire at the end of the decade, and we're seeing far fewer people studying to become network engineers.

That means that we won't have enough new network engineers to replace the ones that leave, let alone enough to keep pace with the massive growth in internet demand.

The reasons for this contraction reveal something important about how we've treated the craft of networking for the past twenty years.

• A well-run network is, by definition, one that nobody thinks about. It's only when something breaks—when a BGP misconfiguration at Facebook takes down WhatsApp, Messenger, and Instagram, or an empty DNS record topples AWS us-east-1 and sectors of the economy go dark—that everyone has something to say about it. It takes a toll to do a job that can feel so thankless.

• Though network engineers' salaries have grown over the past decade, they still earn 70% of what their software engineering counterparts do, and this disparity weighs on network engineers. (We know senior network engineers who spent years getting their CCNA certification who are paid the same as entry-level software engineers.) This gap is pushing ambitious technologists to become software engineers—leading the number of software engineers to outpace networking engineers nearly 4 to 1.²

• Networking as a field suffers from a big gender disparity, which effectively cuts the available talent pool in half. Only 12% of network engineers are women compared to the 20% of software engineers.

• Standalone network engineering programs are getting folded into other engineering programs as companies prioritize software engineering, cloud computing, and security hires.³ Networking is already a very hard skill to learn and there's no "Hello, World!" equivalent for it, which makes it even harder for students interested in networking to pursue it.

When you pencil it out, the math is blunt. If nothing changes by the end of the decade, 3 of the remaining 4 network engineers in a given cohort will need to support 3× the work, most of it the kind of time-consuming manual labor that doesn't require their judgment at all. This immediately places more pressure on network engineers, which in turn makes it much easier to make the kinds of errors that cascade into outages.

But there's also a deeper, more consequential risk. These engineers are about to be asked to do the most important networking in a generation, and we're on track to make sure they have no time to do it. Most protocols haven't been updated since the 80s and traditional networks are lossy by design, which means they're straining under the requirements of new zero-latency, zero-loss AI workloads. If we want to realize the promise of AI, we need to give the engineers who can architect its networks the time and the tools to build it.

Tools for reclaiming craft

Now, there are a number of efforts the tech community can and should undertake to improve the supply side of this equation, like lobbying the government to support standalone networking programs, pushing large industry bodies like the IEEE and Ultra Ethernet Consortium to underscore how important it is to prepare engineers for the networking demands of AI workloads, allowing existing networking students to opt into apprenticeship programs in lieu of writing a thesis, and more generally celebrating networking for the craft it is and not the commodity it's made out to be. As essential as these efforts are to building out a pipeline of network engineers, they'll also take time to bear fruit.

This is why we, at Meter, are so focused on equipping network engineers with the best tools to do their best work today. Fast, secure, and reliable networks should be table stakes as we build toward this new era of computing, so we designed our full stack networks to grow with you and deliver outcomes instead of more issues to manage. If your business doubles in size, or if you add a new office, or if your throughput suddenly needs to scale, we'll give you a predictable cost structure with no surprises. If your bandwidth doubles, we'll give you a bigger firewall. If your headcount grows by 100, we'll give you the switchports and access points to match.

It should also be as simple as possible to interact with your network. Command allows you to query it in natural language, and if you want it to take action on your behalf, it can do that, too. For a given network, you could ask Command to create an SSID with name guest with VLAN external, enable WPA3 authentication, auto-rotate passwords daily at 2am, and enable for all the APs on the network—and it'll do it for you.

Eventually, these networks will anticipate and solve problems before you realize they've happened. This is the real opportunity hidden inside what looks like, from a distance, a story about decline. The network engineers we have in the field are the ones best positioned to reinvent it for the AI era. Your expertise is available to us now if we can ease the burden on your shoulders. Think: what would you build if you knew your network wasn't going to fail?

That said, none of this is a fait accompli. The internet has always been a labor of love, sustained by people whose great work has long gone unnoticed. It's a rough bargain, but it held because the work was interesting and important. The work ahead is more interesting still—genuinely novel and the kind of engineering challenge that comes along once in a generation. It would be a shame to miss it because we were busy typing configuration commands.