Most developers I talk to assume cloud hosting is always cheaper than owning hardware. That was true in 2012. It's not true anymore, especially if you're running workloads 24/7 and your traffic is predictable.
The question isn't "cloud or self-hosted?" — it's "at what scale does the math flip?" Let's calculate the actual break-even point in 2026, including every hidden cost both sides conveniently ignore.
The cloud bill everyone forgets about
You're paying for compute, storage, and bandwidth. That part is obvious.
What sneaks up on most teams is egress. Transfer data out of AWS US-East and you'll pay nine cents per GB after the first 100 GB per month. A single WordPress site serving 50 GB of images and video a day will rack up $135/month in bandwidth alone. DigitalOcean and Linode cap or pool egress more generously, but the big three clouds treat outbound traffic like a profit center.
Then there's the reserved instance trap. Buying a one-year or three-year commitment gets you up to 72% off on-demand pricing, but now you're locked in. If your traffic shrinks or your architecture changes, that pre-paid capacity sits idle. I've seen companies pay for unused reservations longer than they paid for the original instances.
Managed services pile on faster than you expect. RDS, managed Redis, load balancers, NAT gateways — each one adds $15 to $50/month even at small scale. A typical production web app with a managed database, a cache layer, and a CDN will hit $200/month before you serve a single request.
What self-hosting actually costs
Buying a server means paying up front, then spreading that cost across its useful life.
A mid-range Dell PowerEdge R250 with a Xeon E-2434, 64 GB ECC RAM, and two 1 TB NVMe drives in RAID 1 costs around $2,100 in early 2026. Add another $600 for a second identical machine so you have redundancy, and you're at $2,700 before networking gear. Expect a four-year lifespan before hardware failures or CPU bottlenecks force a refresh.
Amortize that over 48 months and you're spending $56/month on hardware alone. Not terrible, but we're not done.
Power and cooling
Each server pulls 80-120 watts under typical load. At $0.13/kWh (rough US average for commercial power), that's $9-$11 per server per month. Two servers, $20/month. If you're in a region with expensive power or you need AC in the server closet, double it.
Colocation vs. home office
Running servers in a closet works for side projects. For anything customer-facing you need real uptime, which means colocation.
A quarter rack (10U) in a decent data center runs $200-$400/month depending on region and bandwidth commitment. That price usually includes 3-5 TB of monthly transfer, redundant power, climate control, and physical access. Some providers charge extra for remote hands, so budget another $50-$100/year for the two times you'll need someone to press a button.
IP addresses and network
You'll need at least one public IPv4 address. Some colos include one or two; others charge $1-$3/month per IP. If you're hosting multiple client sites or need separate IPs for SSL (less common with SNI, but still a thing), that adds up.
A managed switch and firewall are optional if your colo provides port-level access, but I recommend a small edge firewall anyway. Budget $300 one-time for something like a UniFi Dream Machine, amortized over the same four years.
Bandwidth overages
Most colo plans include 3-5 TB/month. Overages typically cost $0.01-$0.03/GB, which is still cheaper than AWS egress but not free. A WordPress multisite network serving 200 GB/day will blow through 6 TB/month and pay an extra $20-$40.
Labor is the real cost
Cloud marketing loves to claim "no ops overhead," but that's only half true. You still write deployment scripts, monitor logs, patch software, and troubleshoot outages. The difference is who patches the kernel and replaces failed drives.
With self-hosting, that's you.
Maintenance hours
Expect 2-4 hours per month for routine updates: OS patches, panel updates (if you run something like cPanel or Virtualmin), checking SMART stats, and verifying backups actually work. If you value your time at $50/hour, that's $100-$200/month.
Unplanned outages average 2-3 per year for a two-server setup with decent monitoring. Budget 2-6 hours per incident: diagnosing the failure, driving to the data center if it's a hardware issue, swapping parts, restoring from backup. That's another $200-$600/year, or $17-$50/month averaged out.
Opportunity cost
The honest question: could you spend those 4-6 hours per month on something that makes more money? If you're an independent consultant billing $150/hour, self-hosting costs you $600-$900/month in lost revenue even if the servers themselves are cheap.
For a small agency or a team already paying a sysadmin salary, the marginal cost is close to zero. Your admin is there anyway; adding two more servers to the monitoring dashboard doesn't double their workload.
The break-even calculation
Let's define two realistic scenarios and run the numbers.
Scenario A: Single server equivalent
You need the equivalent of a DigitalOcean droplet with 4 vCPUs, 8 GB RAM, 160 GB SSD, and 5 TB transfer. DigitalOcean charges $48/month for that.
Self-hosted, you're running one physical server (no redundancy, acceptable for dev or low-stakes production):
- Hardware amortized: $28/month (single $1,350 server over 4 years)
- Power: $10/month
- Colocation (shared quarter rack or 2U): $100/month
- IP address: $2/month
- Maintenance labor (2 hours/month at $50/hour): $100/month
Total: $240/month.
Cloud wins easily. The gap widens further if you value your time higher than $50/hour or if you hit a hardware failure in year two.
Scenario B: Production stack with redundancy
You need the equivalent of two AWS t3.xlarge instances (4 vCPUs, 16 GB each), an RDS Postgres instance (db.t3.large), 500 GB of EBS storage, and 8 TB/month egress. Running that 24/7 on-demand in US-East costs roughly $450/month for compute, $50/month for RDS, $40/month for storage, and $720/month for 8 TB egress after the free tier. Managed backup snapshots add another $30. Total: $1,290/month, and that's without reserved instances or savings plans.
Self-hosted with two physical servers for redundancy:
- Hardware amortized: $56/month
- Power (two servers): $20/month
- Colocation (quarter rack, 5 TB included): $300/month
- Bandwidth overage (3 TB extra at $0.02/GB): $60/month
- IP addresses (two): $4/month
- Maintenance labor (4 hours/month at $50/hour): $200/month
Total: $640/month.
Self-hosting saves $650/month, or $7,800/year. Over four years that's $31,200 in savings, minus the $2,700 initial hardware cost, for a net gain of $28,500.
Now let's say you value your time at $150/hour instead. Maintenance jumps to $600/month, and your total self-hosted cost becomes $1,040/month — still $250/month cheaper than cloud, but the margin is thin. One major hardware failure or architectural mistake could erase the savings.
When the cloud actually wins
Bursty or unpredictable workloads favor cloud. If your traffic spikes 10x during Black Friday then drops to nothing in January, paying on-demand for those spikes is cheaper than owning capacity that sits idle eleven months a year.
Rapid scaling matters too. Spinning up 50 new instances to handle a traffic surge takes five minutes in the cloud. With self-hosting, you're capacity-constrained by the hardware you already bought.
If you're a solo developer working nights and weekends, the labor cost of self-hosting eats your margin. Spending Saturday morning troubleshooting a dead power supply isn't worth saving $80/month.
When self-hosting wins
Predictable, always-on workloads tip the scale. SaaS apps, company intranets, WordPress multisite networks, and mail servers all run 24/7 with fairly stable resource usage. The longer your utilization stays high, the faster you recover the upfront hardware cost.
High egress workloads scream for self-hosting. Video streaming, large file downloads, image-heavy sites — anything pushing terabytes per month will bankrupt you on AWS egress fees. Colo bandwidth is often unmetered or priced 10x cheaper.
If you already employ someone who knows Linux, adding self-hosted infrastructure is nearly free labor. The sysadmin is on payroll whether they manage two servers or twenty.
The hybrid middle ground
Most teams I work with don't go all-cloud or all-self-hosted. They run databases and steady-state app servers on owned hardware, then burst to cloud spot instances for batch jobs and traffic spikes.
Another pattern: self-host your staging and dev environments where uptime doesn't matter, and keep production in the cloud for the SLA and support.
Or flip it: put your static assets and databases on a cheap dedicated server with unmetered bandwidth, and run your app layer in the cloud for easy horizontal scaling.
What about managed dedicated servers?
Renting a dedicated server from Hetzner, OVH, or Vultr splits the difference. You get a physical machine (no noisy neighbors, predictable I/O), but the provider handles hardware replacement and network uptime.
A Hetzner AX41 with a Ryzen 5 3600, 64 GB RAM, and two 512 GB NVMe drives costs €46/month (~$50 USD) with 20 TB of included traffic. That's cheaper than self-hosting after you factor in colo and labor, and way cheaper than equivalent cloud instances.
The tradeoff: you can't scale horizontally as fast, and you're still responsible for OS patching and app-level monitoring. It's a solid middle ground for small teams who want cost efficiency without the full ops burden of self-hosting.
The four-year TCO comparison
Let's summarize the total cost of ownership over a four-year hardware refresh cycle for a production stack (Scenario B above).
Cloud (AWS on-demand, no reservations):
$1,290/month × 48 months = $61,920
Cloud (with 3-year reserved instances, ~40% savings):
~$775/month × 48 months = $37,200
Managed dedicated (Hetzner-class, two servers):
$100/month × 48 months + $400 initial setup = $5,200
Self-hosted (colo, with labor at $50/hour):
$640/month × 48 months + $2,700 hardware = $33,420
Self-hosted (colo, with labor at $150/hour):
$1,040/month × 48 months + $2,700 hardware = $52,620
Managed dedicated wins on pure cost. Self-hosting with cheap labor beats reserved cloud instances. Cloud on-demand is the most expensive option by a wide margin unless your workload is truly bursty.
Making the call for your workload
Start by tracking your actual usage for 30-90 days. Log your peak CPU, average memory, disk I/O, and total egress. Most teams over-provision by 40% or more when they guess.
Calculate your monthly cloud bill with and without long-term commitments. If you're already on reservations, check your utilization — are you paying for capacity you don't use?
Price out equivalent self-hosted hardware, then add colo and labor at your real hourly rate (not what you wish it was). If the self-hosted total is more than 30% cheaper, the math probably works. Under 30%, the risk and opportunity cost aren't worth it.
Run the same calculation for managed dedicated. If Hetzner or OVH can deliver 80% of your requirements at 60% of your cloud bill, that's usually the best risk-adjusted option for small teams.
Start with your egress bill
If your monthly bandwidth costs more than $100, you have a strong economic case for self-hosting or switching to a provider with unmetered or pooled transfer. That single line item often outweighs every other factor in the break-even calculation.
Check your cloud bill right now. Filter by egress charges. If that number makes you wince, it's time to run the math.