As the work-from-home revolution gained momentum, a new narrative began to emerge about the burgeoning costs of cloud computing. WSJ journalist Aaron Tilley, wrote that “The pandemic has made business more dependent on cloud computing than ever--and companies are now racing to rein in the soaring costs.” Months later Andreessen Horowitz wrote about the “Trillion Dollar Paradox” in “The Cost of the Cloud”. In this narrative, the cost explosion was primarily an affliction of multi-billion dollar companies. Tech behemoths, such as Dropbox or Crowdstrike, had reached a scale that had pushed them over a cost efficiency threshold. Once they crossed this threshold, operating in the cloud became actually more expensive than running their own infrastructure.
The open question though, is whether startups need to worry about this ‘luxury’ problem. As one Hacker News commentator put it “A startup worrying about long term cloud implications is like a hot dog vendor in New York worrying about how his branding will be received in India when his hot dog empire expands there”. Does that mean startups shouldn’t be concerned at all about overreliance on AWS, Azure, Google Cloud, or any other cloud vendor? In a startup’s early stages, this problem is rightfully the last thing on a CTO’s mind.
However, as CTO for Project A—it’s definitely a problem that I grapple with. Like any other venture capital firm, we’re betting on the hope that at least one of our portfolio companies does indeed become like “the hotdog vendor that conquers the world” and follow the trajectory of a Dropbox or an Uber. And as a16z rightfully points out, a unicorn can eventually reach a tipping point where they’re paying the big cloud vendors more than they have to. And this eats into their profitability. So, it’s also in our best interests to help our portfolio companies make the right decisions when scaling their infrastructure.
However, most VCs don’t get involved in these kinds of decisions on an operational level. That’s what makes us a little different. Project A is an “operational” VC which means we provide our portfolio companies with operational expertise as well as funding. I’ve seen many startups struggle with keeping IT costs down. But that’s not a problem that is inherent to the cloud. The challenge is having the in-house expertise to optimise the infrastructure so that you get the most performance for the least cost. Indeed, startups lack in-house expertise in a lot of things, which is why Project A was founded. As head of the IT department, I allocate our in-house engineers to different portfolio companies to help them build out infrastructure and products. Our team applies DevOps culture to help manage the complexity of running an application in the cloud. And with the right expertise, you can reduce costs dramatically. For example, we were able to help a specific portfolio company reduce their AWS costs by 65 percent by switching from Fargate to Lambda.
For the average startup, it takes a while to find the right talent and develop the in-house knowledge required to keep costs under control while navigating the ever-growing complexity of the cloud computing ecosystem
For the average startup, it takes a while to find the right talent and develop the in-house knowledge required to keep costs under control while navigating the ever-growing complexity of the cloud computing ecosystem.
But there is also a growing market for tools that help abstract away this complexity. The trick is to find out about them. In a nebulous labyrinth of today’s cloud tooling ecosystem, that’s easier said than done. It pays to have a network of CTOs that can help you figure out what works and what doesn’t. For example, through our network we found a tool that made it possible to replace cluster nodes with much cheaper spot instances while still maintaining a similar level of performance. We introduced it to one of our other portfolio companies and were able to reduce their AWS costs by 60 percent.
Another thing startups can do is incentivize their existing teams to be conscious of cloud costs early on. For instance, a16z recounted the tale of a CTO who offered bonuses to his developers if they made changes which helped to save cloud costs. Personally, I’m not sure if I’d endorse this approach since it may tempt some developers to exploit such a system for easy financial gain. However, I do think cost is a KPI that should be built into how you assess the overall performance of your engineering teams. This ties generally into wider themes such as measuring developer productivity and prioritising observability. And again, there is a variety of tools such as Datadog and Splunk that help you make cost more visible to your teams.
As the old adage goes, “knowledge is power”—and it also gives freedom to make better decisions. This applies whether you’re fully in the cloud, hosting your own infrastructure, or have some kind of hybrid architecture. So, depending on their current stage I advise startups to think strategically and build up the right knowledge and expertise required to navigate the complexity of the cloud technology landscape. This means hiring DevOps specialists and developers who are well-versed in DevOps culture. There’s a fierce battle in the market right now, to find the right talent which is why it’s important to start early. Salaries for experienced DevOps and CloudOps engineers can be high, but any decent engineer who can optimise your cloud infrastructure will quickly pay for themselves many times over.
So to go back to the original question—should startups worry about becoming too reliant on cloud computing? Generally speaking, no. Cloud computing is perfectly suited to the unpredictable growth trajectories and scaling requirements of tech startups. But within the cloud ecosystem itself, there are a hundred different, different ways to achieve the same goal—some of which will cost you more than others. If a cloud offering is highly “managed” and automated, the chances are, it’s also more expensive (think AWS Fargate vs pure EC2). That’s why it’s so valuable to have specialists who understand what’s going on underneath those layers of abstraction. This frees you from being over reliant on those “easy to use” but expensive services. And often there will be 3rd-party vendors who have found ways to optimise costs better than the cloud computing provider themselves. That’s why startups need to focus on getting the right people and tools, rather than worrying about getting “too big for the cloud.
