As someone who has spent much of the past decade writing about cables, data centers, and the general physical stuff that makes the internet possible, I’m fascinated by how industry players, academics who study tech, and software-oriented foundations talk about technology as infrastructure. Infrastructure literally just means “the structure beneath (infra) another thing,” and technology as we know it operates on a hierarchical “stack,” so practically everything in that stack is infrastructural to some degree.
Hardware is infrastructure for running software protocols, protocols are infrastructure for accessing software applications. Admittedly, I think that the terminology sometimes gets a little out of hand: Mark Zuckerberg’s infamous 6,000-word “Building Global Community” letter from 2017 used the phrase “social infrastructure” 15 times to basically describe Facebook. But it’s not that I care whether or not Facebook counts as infrastructure. Debating what qualifies as tech infrastructure is a boring distraction from more interesting questions about what people intend when they declare something “infrastructure” — and there are other helpful words that we can use to illuminate hardware and software’s role in the world more clearly.
I’ve lately been trying an exercise where, when reading anything by or about tech companies, I replace uses of the word “infrastructure” with “means of production.” For example, from Facebook’s engineering web page:
“Our data centers are the cornerstones of the global means of production that brings Facebook apps and services to you every day.”
The sentence pretty much still makes sense — without data centers, Facebook can’t reach people, and therefore can’t make money. It also works pretty well with this copy from a Cloudflare tutorial on the concept of infrastructure-as-a-service (IaaS):
“In computing, the means of production refers to the computers and servers that run code and store data, and the wires and appliances that make connections between those machines. For example, servers, hard drives, and routers are all part of the means of production. Before cloud computing was an option, most businesses hosted their own means of production and ran all their applications on-premises.
“Means of Production-as-a-Service, or IaaS for short, is when a cloud computing vendor hosts the means of production on behalf of their customers.”
So… a landlord. It’s a little heavy-handed, I’ll admit, but precision in naming things for what they are matters. As use of the term “infrastructure” in tech has grown, it’s easy to lose sight of what actually gives Big Tech its power and what’s at stake when proposing alternatives to such centralization: capital, and who controls it.
Framing the platform policy discussion around the means of production also helps establish reasonable expectations. While it would be nice if Facebook as a “social infrastructure” provider had the vague sense of civic purpose that a term like “social infrastructure” implies, time and again we’ve seen that the company will not implement anything that serves the public but undermines Facebook’s profits. Instead of describing Facebook as providing social infrastructure, we could simply say that it utilizes social interactions for profit through the means of production (by owning lots of computers and cables) and dispel the illusion of kind civic intentions.
Maybe it sounds stodgy to refer to vanilla means-of-production capitalism without stapling a cool tech-inflected buzzword to it — the “means of computation,” say, or the “means of connection.” Whatever you want to call it, any vision of building a less centralized, more equitable internet has to ultimately account for ownership of all the physical stuff — the means — that makes the internet as we know it.
Property has always been a major determinant of power online, the same way it shapes power throughout society. Being a landlord is one of the oldest business models of the commercial internet — but through some bizarre sleight of hand, these companies are generally referred to as cloud or “hosting” services, as though they provide dinner parties instead of leased server space. Early coverage of the industry often emphasized its industrial, factory-like qualities. The term “landlord” was occasionally used, but it didn’t stick.
The business model of owning and operating terrestrial or subsea cable routes is also not a new idea — they weren’t ever really “the people’s cables.” What has changed since the early years of the internet is the concentration of ownership of these components and conduits into the hands of ostensible software companies.
Big Tech’s “big five” (Microsoft, Amazon, Apple, Facebook, and Google), of which three began as websites, today own and maintain multiple large-scale data centers, fiber lines, and, in some cases, submarine cables. Microsoft, Amazon, and Google dominate cloud services outside of the Asia-Pacific market, and they’re still top performers even within that particular market. While Facebook and Apple’s business models don’t include business-to-business cloud products, both are arguably in the business of consumer digital storage as the place where personal photos, videos, and group chats reside. In addition to establishing itself at the forefront of data center industry standards with the Open Compute Project, Facebook also expanded into infrastructure from a slightly different angle in 2019, launching a middle-mile fiber business to sell its excess network capacity to smaller local telecoms.
For better and worse, massive investments in the means of computation by those five companies set the stage for most of the internet as we know it today. Amazon Web Services arguably powered much of the tech bubble of the 2010s by making it easy to quickly spin up and scale a startup. The rapid adoption and spread of artificial intelligence technologies over the past half-decade could not have happened without the thousands of GPUs in the big five’s data centers, or the vast amounts of user data stored by these companies to build training models. The entire streaming business model as we know it could not exist without robust networks to transfer files. If you’ve spent the past year working from home dreading back-to-back Zoom meetings with your colleagues, you don’t just have Zoom and your ISP to thank; you can also thank the range of data center services that Zoom relies on.
The negative impacts of platform centralization on the public don’t directly result from holding the means of computation: having a lot of data centers has no correlation with spreading misinformation or enabling hate speech and online harassment beyond allowing such activities to scale. But a monopoly on the means of computation adds to the inertia that makes leaving said platforms challenging. Even if there’s an alternative platform with better safeguards against harassment, why go to a platform where I have to pay for storage, or where fewer of my friends are, or that’s occasionally offline because it’s actually just a single Mastodon instance run by one person?
In an era of precarious employment where survival increasingly depends on social media presence and/or crowdfunding, inertial platforms also hold tremendous power with the threat of deplatforming individuals based on capricious corporate whims. (Though in practice that tends to involve companies discriminating against sex workers while dithering about whether or not to ban obviously harmful actors who encourage violent insurrection.)
Though being in cloud services is very good business, it’s not always great for the businesses that depend on it. Startup rentiers on Amazon Web Services have accused the cloud landlord of essentially cloning their products. It’s also hard to quit the cloud: given the custom particularities of each platform, businesses can become locked into paying prices that they can’t afford to a landlord that they don’t like.
There are other, less obvious harms of computation concentration on both a hyper-local and a planetary level. Companies looking to build data centers make lots of promises to municipalities in exchange for cheap land, power, and water — promises that aren’t always fulfilled. And platform centralization happens in tandem with network centralization, as what began as a wide field of internet service providers has merged and metastasized into a handful of dominant vendors with strongholds on regional markets.
But shifting ownership of the means of computation is not as straightforward a process as workers taking over a factory or a mine. With internet infrastructure, we’re not talking about a discrete piece of property that can be autonomously taken over: it’s cables and antennae and spectrum and all sorts of very expensive stuff that requires specialized technical maintenance, not to mention coordination with other interdependent systems.
This is partly why many projects to recreate a distributed internet begin with software. Code is cheap; subsea fiber-optic cable is not. For companies and open-source projects working at the protocol level to shift away from the internet’s hub-and-spoke model, the means of computation tends to be redistributed to users (the basis of most peer-to-peer technologies). But consumer-grade laptops weren’t built to store the volume of data that a planetary-scale network can produce, and not all consumers especially want to invest in loads of external hard drives for all of the stuff that they’ve become accustomed to streaming or just pulling from iCloud.
Some distributed web projects, like IPFS, have come to depend in part on cloud services as hosts of “gateways” — which, while lowering the barrier for entry to users who might not want to install special software or give up storage space, also undermines the distributed-ness of the effort. This isn’t to fault IPFS in particular or distributed web projects that generally rely on cloud data centers. They aren’t really in a position to redistribute blade servers to users around the world, and a lot of them probably aren’t undertaking this work in pursuit of redistributing capital (especially if they’re simultaneously seeking venture capital funding). Developing the right piece of software to overtake existing centralized platforms on an increasingly cloud-centralized internet is a tall order, but redistributing the means of computation is ultimately more about changing property ownership models.
On the consumer technology side, there’s been some movement toward more equitable business models like platform cooperatives, which don’t redistribute the means of computation so much as the profits from them. These kinds of companies can in theory pool resources with other cooperatives to develop collectively managed data centers rather than outsourcing the task to yet another landlord. In the meantime, there are some interesting nascent efforts to redistribute consumer data storage without eating up laptop hard drives. Startups like Cubbit offer distributed, peer-to-peer cloud storage using technologies that can in theory run on most computers with an internet connection. Cubbit offers out-of-the-box custom hardware for consumers, and their argument for distributed storage is more practical and environmental than political, but as a proof of concept it’s a compelling model that companies and communities alike could provide as a service.
It’s a little ironic that some of the most inspiring models for distributing the means of production in computing and internet access operate at what industry treats as the end of the line — the “last mile” of network connections where ordinary people log on. But this is also where the impacts of limited or no internet access are most felt. Organizing alternatives emerges as a necessity.
Ownership models for ISPs serving poor or rural areas vary from utility cooperatives to municipal and tribal government-supported networks to community networks run by and for residents. Maybe part of imagining a different distribution of the means of computation requires a flipping of the script. Rather than assuming that the internet starts as massive nodes of platform data centers and internet exchanges, perhaps the last mile is actually the first step in working toward a different vision of who should own and govern the means of computation.
What’s at stake for both the tech industry and government regulators isn’t what is or isn’t infrastructure, but what the ownership and profit model for that infrastructure looks like and whom it benefits. Substituting “the means of computation” for “infrastructure” isn’t going to make it any easier to alter those ownership models, but it might make it easier for us to focus on building and maintaining an internet that serves the public’s needs.