Leaving the Cloud for the Rest of Us

When Basecamp announced they were leaving the cloud, everyone paid attention. And then everyone dismissed it as irrelevant to them.

After all, Basecamp was spending over $3 million a year on AWS. They bought $600,000 worth of Dell servers. They have a dedicated ops team. They run an email service handling millions of messages.

Most founders I talk to read those numbers and think: that's not me. I'm spending a few thousand a month on my cloud provider. I don't have $600,000 lying around to go buy new hardware so I can leave the cloud. Cloud exit must be for bigger companies with big budgets and infrastructure teams.

What if I told you that you can buy your own servers for a lot less than $600,000? What if I told you that one of those Basecamp servers costs around $30,000 and you can get enterprise grade Dell servers for under $1,000? What if I told you that smaller companies can exit the cloud, too, and own their hardware?

That's what we did at DevStats.

What Basecamp Actually Did

Before I explain what we did at DevStats, let me refresh your memory on how Basecamp left the cloud. It's worth understanding what they did and why before I go over the poor man's cloud exit.

They were paying AWS roughly $267,000 per month. Over $900,000 of that went to S3 storage alone for their 8 petabytes of data. Another $760,000 went to compute. When DHH broke down the numbers publicly, people were shocked. And when they announced they were buying their own hardware instead of renting from Amazon, it became a whole thing on tech Twitter.

Their solution? Twenty Dell R7625 servers, each with dual AMD EPYC processors running 96 threads per CPU. That's 192 threads per machine. Serious enterprise hardware. They split them between two data centers for redundancy. The total hardware investment was around $600,000, which works out to roughly $30,000 per server. They expect to save over $7 million in five years.

That's impressive. But here's the thing: you don't need any of that to leave the cloud.

Switching Landlords vs. Buying the House

After Basecamp announced their cloud exit, a lot of smaller SaaS founders got inspired. Many wrote their own blog posts about leaving AWS. But if you read carefully, most of them didn't actually leave the cloud. They switched from AWS to Hetzner or OVH or some other cheaper VPS provider.

That's not leaving the cloud. That's switching landlords.

Don't get me wrong. Moving from AWS to Hetzner can save you 50% to 80% on your infrastructure costs. That's real money. But you're still renting. You're still paying monthly for compute you don't own. You're still subject to someone else's pricing decisions.

What Basecamp did was different. They bought the servers. They own the hardware. They colocate it in data centers and pay only for the rack space, power, and network connectivity.

That's what we did at DevStats too. We didn't switch to a cheaper cloud provider. We bought our servers. We own them. And the economics at small scale are just as compelling as they are at Basecamp scale.

What We Did at DevStats

At DevStats, we were spending about $1,337 per month on Digital Ocean. Not $267,000. Not even close. Just a normal cloud bill for a small B2B SaaS.

Here's what that got us: seven droplets totaling 14 vCPUs and 40GB of RAM, plus three managed databases eating up over $1,000 of the bill. The servers were fine. I actually love Digital Ocean's user experience. That being said, as we kept growing so did our bill. It seemed like every month as we added a few customers and needed to increase the size of a droplet. Because server capacity was so expensive, we didn't want to over provision.

So I went shopping for used enterprise servers. Not the brand new 192 thread monsters that Basecamp bought at $30,000 each. I found solid, used Dell PowerEdge R640s at newserverlife.com

We picked up four of them for $2,913 total. That's about $728 per server, compared to Basecamp's $30,000. Two servers with dual Intel Xeon Gold 6126 processors and 128GB RAM each. One with dual Xeon Silver 4108s and 16GB RAM. One more with dual Xeon Golds and 64GB RAM. All with SAS SSDs, iDRAC9 enterprise management, and redundant power supplies.

Then we found a colocation facility that would rack them for us. Monthly cost: $240.

Let me show you what that looks like side by side:

We went from 14 vCPUs to 176 threads. That's over 12x more compute. We went from 40GB of RAM to 336GB. That's over 8x more memory. And we're paying 82% less every month.

The hardware paid for itself in roughly four months. Everything after that is pure savings.

You Don't Need to Be Perfectly Sized

One thing I love about owning the hardware is that we over provisioned on purpose because it was so cheap to do so. That's actually the same thing Basecamp did. They over provisioned intentionally.

These used enterprise servers are so powerful relative to what a small SaaS actually needs that we have ridiculous headroom. If DevStats grew 10x tomorrow, we still wouldn't need to worry about capacity. That's not an exaggeration. The compute we now own is just absurdly more than what we were renting.

And here's the thing about B2B SaaS: growth is predictable. You're not Twitter on Super Bowl Sunday. You're not going to wake up to 10x traffic because you went viral on TikTok. B2B customer acquisition is slow and steady. You have sales cycles. You can see your hardware needs coming months in advance.

That means autoscaling is overrated for most of us. The cloud vendors love to talk about elastic scaling like it's this magical capability you can't live without. But if your traffic is predictable, and in B2B SaaS it almost always is, you don't need to pay a premium for instant scalability you'll never use.

Server Capacity Becomes a Fixed Cost

This is the part that gets me excited as a founder.

When you own your hardware and it's cheap to over provision, your server capacity becomes a fixed cost for years. Not a variable cost that fluctuates with usage. Not a bill that creeps up as your business grows. Just a one time purchase that keeps working until your business grows a lot or your servers get too old. The servers we got have at least five years of lifespan left on them.

Our compute costs are now essentially locked in. We paid $2,913 once. We pay $240 per month for colocation. That's it. If we double our customer base next year, our infrastructure costs stay the same. If we 10x, our costs stay the same. If we grow more than 10x, we'll need to buy more servers, but at that point that will be a very good problem to have.

Compare that to the cloud where every customer you add means more compute, more storage, more bandwidth, more money leaving your bank account. The cloud's business model is designed to grow with you, which sounds nice until you realize that means your costs grow monthly too.

What About Disaster Recovery?

I know what you're thinking. What happens if your hardware fails? What if there's a disaster at the data center?

We thought about this too. Our approach is simple: we keep a warm failover at Digital Ocean in a data center far from our colocation facility. If something goes terribly wrong with our colocation or our hardware, we can increase the size of that failover droplet, change the DNS, and we're back online.

Would we be offline for a few hours in a worst case scenario? Yes. Did we decide that's a risk we're willing to take? Also yes.

Here's some perspective: half the internet went offline multiple times this year because AWS, Cloudflare, and Azure had issues.

In the rare case we're offline for a few hours, I'm confident our clients would understand. The big cloud providers aren't immune to outages either.

The point isn't that our setup is bulletproof. The point is that we've made a deliberate decision about the level of redundancy we need, and we're not paying a massive premium for fault tolerance we don't require.

Running Your Own Hardware is Not Scary

Managing servers sounds hard. What about security patches? What about hardware failures? What if something breaks at 3am?

Here's the reality: once the servers are set up and racked, you manage them like any other Linux machine. SSH in, run your updates, deploy your code via zero downtime tools. The same stuff you'd do with a cloud VPS. There's no special magic required.

Yes, if a hard drive fails, you'll need to deal with it. But enterprise servers have redundant drives for a reason. Hardware issues are rare, and when they happen, you handle them.

The mental shift is real. The first time something goes wrong and you can't just spin up a new instance, it feels different. But it's not actually harder. It's just different. And after a few months, you stop thinking about it entirely.

This is Not Only For Big Companies

The whole point of this post is that leaving the cloud is not a Basecamp scale thing. It's a "spending $1,000 or more on cloud bills" thing.

If you're a founder with a stable SaaS, predictable growth, and a monthly cloud bill that's become a real line item in your budget, you can do this. You don't need a massive ops team. You don't need $600,000 in capital. You need a few thousand dollars for used servers and a colocation provider who will rack them for you.

The economics work at small scale. They work even better at medium scale. The only places they might not work is if you are not a B2B SaaS or you're a brand new startup with unpredictable traffic who genuinely needs the flexibility to scale up and down on demand. But once you have product market fit and you're growing at a predictable B2B SaaS growth rate, it's absolutely worth thinking about.

The Bottom Line

Leaving the cloud and actually owning your own hardware is possible. Not just for companies spending millions. For companies like ours.

At DevStats, we left Digital Ocean by spending $2,913 on hardware and $240 per month on colocation. The investment paid for itself in under four months. We now have 12x more compute and 8x more RAM than we were renting. And our infrastructure costs dropped by 82% and now is "fixed" until we 10x our business, while our competitors keep paying cloud premiums.

Many founders who got inspired by Basecamp just switched to cheaper cloud providers. That's fine. But that's not what Basecamp actually did. Basecamp bought their servers. So did we. The difference is they spent $30,000 per server and we spent $728.

If you're a founder spending $1,000 or more on cloud bills every month, do the math on your own situation. You might be surprised how simple the alternative is.

You don't need Basecamp scale. You just need to stop assuming the cloud is the only option.