Custom Software Case Studies: 3 Companies That Built, And What They Still Bought
by GainHQ Editorial Team | Feb 8, 2026 | Software Development Insights | Editorial policy
Table of Contents
Most companies should buy most of their software. The interesting custom software case studies are the ones where a company built something specific because it was core to the business, and kept buying everything else. The three examples below come from the companies' own published accounts.
Example 1: Dropbox Builds Its Own Storage (Magic Pocket)
Decision: Dropbox had stored file content on Amazon S3 since its early years. It decided to build its own storage system, called Magic Pocket.
Why build: Storage is Dropbox's core product. Running it in-house let the team tune the whole stack for its workload and improve unit economics at its scale. The team also found no open-source option it judged reliable at that scale.
Result: Dropbox began serving user files exclusively in-house on 27 February 2015. By 7 October 2015 it was storing and serving more than 90% of users' data on its own infrastructure, nearly a month ahead of plan.
What it kept buying: Dropbox said it would keep working with AWS "where it makes sense," including for storing data in Germany for European business customers.
Source: Dropbox Tech blog, "Scaling to exabytes and beyond," 14 March 2016.
Example 2: Netflix Builds Video Delivery, Buys Cloud Computing
Decision: Netflix moved almost all of its systems to AWS between 2008 and early 2016. It built its own content delivery network, Open Connect, for streaming video.
Why buy the cloud: Netflix said it "could not have racked the servers fast enough" to keep up with growth. The cloud gave it capacity within minutes and better availability. Netflix noted that cost "was not the main reason" for the move.
Why build delivery: Video delivery is the core of the product. By knowing what members are likely to watch, Netflix could place files on servers near viewers before they were requested, which third-party CDNs could not do as efficiently.
Result: By 2016, Open Connect delivered 100% of Netflix video traffic from appliances in close to 1,000 locations. Per-server throughput rose from 8 Gbps in 2012 to over 90 Gbps.
Sources: Netflix, "Completing the Netflix Cloud Migration," 12 February 2016; Netflix, "How Netflix works with ISPs around the globe," 17 March 2016.
Example 3: 37signals Leaves The Cloud
Decision: 37signals, the company behind Basecamp and HEY, moved its applications from rented cloud infrastructure to its own servers.
The numbers (preliminary, as published):
- 2022 cloud bill: $3.2 million, of which just under $1 million was S3 storage.
- Hardware purchase: roughly $600,000.
- Expected yearly cost after the move: about $840,000.
- Expected savings: about $7 million over five years, without growing the operations team.
Caveats: The author called the numbers rough and preliminary. The S3 storage move was planned separately and was not included.
What These Custom Software Case Studies Have In Common
| Company | Built | Kept Buying | Main Reason To Build |
|---|---|---|---|
| Dropbox | File storage | AWS for some regions | Core product, scale economics |
| Netflix | Video delivery (CDN) | AWS for almost everything else | Core product, delivery efficiency |
| 37signals | Server infrastructure | Data-centre space, power, bandwidth | Predictable workload, cost |
Each company:
- Built only what was core to the product or a large, predictable cost.
- Had the scale and skills to run what it built.
- Kept buying everything that wasn't a differentiator.
These are large or specialised companies, so the numbers won't transfer directly to a smaller business. The reasoning does: build where it sets you apart or saves money at your scale, and buy the rest.
For a step-by-step framework, see our build vs buy software decision guide. For budgeting a build, see how much custom software development costs.