The cloud
The infrastructure that made modern software possible
Description
The word cloud has been doing a lot of work for almost two decades, and most people who use it would struggle to define it precisely. It evokes something diffuse and weightless, a place where data lives without any particular geography. The reality is the opposite. The cloud is a network of concrete buildings windowless warehouses in suburban Virginia, cooling systems that drink rivers of water, fiber-optic cables that follow nineteenth-century telegraph routes, and electrical substations that draw enough power to run small cities. The metaphor was chosen by marketers who wanted to abstract away the physical layer. They succeeded, perhaps too well.
What we call cloud computing is a way of selling computing as a metered utility rather than equipment a company buys and operates. Before the cloud, launching a web service meant ordering servers, waiting weeks for delivery, racking them in a data center, configuring the operating systems, and praying the hardware lasted. After the cloud, it meant typing a command and paying by the hour. Most of the modern internet streaming video, mobile apps, social platforms, the AI products of the last few years depends on this shift in a way that is easy to forget.
The story is partly about a single company making a strange bet, partly about economics that had been waiting for the right packaging, and partly about the consolidation of computing power into a small number of hands. We need to decode the three threads together to make sense of why a generation of software companies could exist at all, and why the same dependence now produces concerns no one was raising in 2008.
The question we're asking: what is the cloud, why did it arrive when it did, and what did building on it cost us?
What we'll see: the technical foundations, the economic logic, the market that emerged, and the dependencies we now live with.
Table of contents
01What the cloud actually is
The technical idea behind the cloud is older than most people assume. Time-sharing systems in the 1960s let multiple users at separate terminals draw computation from a single mainframe, paying for what they used. John McCarthy gave a talk at MIT in 1961 suggesting that computing might one day be organized as a public utility, sold like electricity or water. The vision was correct in outline but premature by about forty-five years. Time-sharing faded as personal computers got cheap enough that owning one was simpler than renting cycles from a remote machine.
The technology that made the modern cloud possible was virtualization the ability to run several isolated operating systems on a single physical server, each one believing it has the machine to itself. The technique existed on IBM mainframes in the 1970s but became practical for commodity x86 hardware only around 2001, when VMware figured out how to do it efficiently. A server that previously hosted one workload could now host ten or twenty. The same building, the same power draw, the same staff produced an order of magnitude more useful computing.
02The economics that changed everything
The simplest way to describe what the cloud changed is the shift from capital expenditure to operating expenditure. Before the cloud, a company that wanted to run software bought servers. Servers were assets on the balance sheet, depreciated over several years, and required staff to maintain. Buying more capacity meant a multi-month process of forecasting, ordering, installing, and configuring. After the cloud, the same capacity arrived as a line item on a monthly bill. A startup with a credit card could provision in an afternoon what once required a procurement department.
The economic argument that mattered most was elasticity. A traditional data center had to be sized for peak load the busiest hour of the busiest day of the year. The rest of the time, capacity sat idle. Cloud providers could pool demand across thousands of customers whose peaks rarely aligned. A retailer's holiday surge happened when an accounting firm's quarterly close was over. A social network's evening peak happened when a bank's batch processing was running at midnight. The pooling meant the cloud provider needed less total capacity than the sum of what its customers would have built individually. Some of that efficiency was passed on as lower prices, some kept as margin.
03How three companies divided the market
The cloud market is dominated by three providers. Amazon Web Services holds roughly a third of the global market, Microsoft Azure roughly a quarter, and Google Cloud Platform around a tenth. The rest is fragmented across regional providers, specialized players like Oracle and IBM, and Chinese hyperscalers like Alibaba and Tencent that are dominant at home and minor everywhere else. The concentration is unusual in technology a market this large normally has more competitors.
Each leader entered for different reasons. Amazon was first, and its lead came from years of operating without serious competition. Microsoft entered seriously around 2010 under Satya Nadella, recognizing that its enterprise software business needed cloud distribution. Microsoft's existing relationships with corporate IT departments turned out to be its most valuable asset — a Fortune 500 company already running Windows, Office, and Active Directory had a natural path to Azure. Google Cloud entered last and struggled longest. Google had unmatched internal infrastructure, but selling it required a different muscle, and the company spent years discovering that engineering excellence does not automatically translate into enterprise sales.
04The dependencies and the fragility
The convenience of building on the cloud has produced a level of concentration that occasionally becomes visible when something breaks. A 2017 AWS outage in northern Virginia took down a substantial fraction of the internet for several hours Slack, Quora, Trello, Coursera, Imgur, and thousands of smaller services were unreachable. A 2021 incident at Fastly broke major news sites simultaneously. Each incident is treated as a one-off, but the pattern is structural. When a small number of providers carry most of the traffic, their failures cascade in ways that single-company failures used to not.
Vendor lock-in is less spectacular than outages but probably more consequential. Providers compete partly by offering services that make development faster managed databases, machine learning APIs, queue systems, identity management. These services are specific to the provider that offers them. A company that builds extensively on AWS Lambda, DynamoDB, and SQS cannot easily move to Azure without rewriting substantial portions of its software. The cost of leaving grows with every new service adopted. Renegotiating contracts is harder when the alternative is a multi-year migration.
05Conclusion
The cloud is one of those rare technological shifts that arrived faster than its critics predicted and produced more dependence than its advocates expected. A generation of software companies exists because launching a service no longer required raising capital for hardware. Researchers can run experiments in hours that previously took months. Small teams can serve global audiences with infrastructure that would have required hundreds of staff in 2005. The productivity gains compound across every industry that uses software, which is now most of them.