Dropbox moved off the public cloud and doubled its gross margin
Between 2015 and 2017 Dropbox moved most of its storage off Amazon and onto its own hardware. Its IPO filing shows gross margin going from 32.5% to 66.6% over the same period.
One vote per reader. You can change it. No account needed.
The numbers from the IPO filing
Dropbox registered for its IPO in February 2018. The prospectus covers the three years of the migration.
| Year | Revenue | Gross profit | Gross margin |
|---|---|---|---|
| 2015 | $603.8M | $196.4M | 32.5% |
| 2016 | $844.8M | $454.2M | 53.7% |
| 2017 | $1,106.8M | $737.9M | 66.6% |
Revenue grew 83% over the two years. Gross profit grew 276%. That gap is the migration.
What actually moved
The migration was not "leaving the cloud" in the way the phrase is usually used. Dropbox moved the storage layer — the actual blocks of customer files — onto hardware it designed and operated. Metadata, parts of the serving path, and regions where it had no facilities stayed on third-party infrastructure.
That distinction matters, because the economics only work for the part that was moved. File storage at Dropbox scale is enormous, extremely predictable and almost entirely undifferentiated. It is the ideal candidate. A workload that is spiky, small, or changing shape every quarter is the opposite.
What it cost
The margin improvement is not free money. Running your own storage means buying hardware years ahead of the demand for it, staffing a team that can operate it at three in the morning, and accepting that a capacity mistake is now a purchase order rather than a slider. Dropbox spent years building that capability before the savings appeared, and the company was already large enough that a dedicated infrastructure organisation was affordable.
There is also a straightforward risk that does not show up in a margin table: once the hardware is bought, it has to be filled. Renting capacity means paying for what you use. Owning it means paying for what you guessed.
When this applies to you, and when it does not
It might apply if one predictable resource dominates your cost of revenue, that resource is undifferentiated, your usage is large and growing steadily, and you can afford a team whose full-time job is running it.
It does not apply if your cloud bill is mostly compute for variable workloads, if your usage doubles or halves with the season, or if the engineers who would run the hardware are the same engineers who would otherwise be building your product. For most companies below serious scale, the cloud bill is not the problem — the unexamined instances in it are.
The reasonable first move is not a migration. It is one week of looking at what the bill is actually made of, which for most companies surfaces more savings than any architectural change.
Questions readers asked
How much did Dropbox save?
Coverage of the prospectus reports operating cost savings of $74.6M across 2016 and 2017 from the infrastructure programme. We were able to verify the revenue and gross profit figures directly in the filing; the $74.6M figure comes from press reporting of the management discussion section, so we attribute it rather than stating it as our own verified number.
Does this mean the cloud is too expensive?
It means renting storage was more expensive than owning it at Dropbox’s specific scale and usage shape. That is a narrow finding, not a general one.
What is the smallest company this could make sense for?
There is no clean threshold, but the practical test is whether you could staff a team to run the hardware without taking those people off product work. If not, the migration cost is hidden in delayed roadmap rather than in the infrastructure line.
What should we do instead?
Audit the existing bill first: unused capacity, oversized instances, storage tiers and egress. Most companies find a double-digit percentage there without changing any architecture.
Where sources disagree: The widely quoted $74.6M savings figure appears in the management discussion section of the S-1. We verified the revenue and gross profit figures directly against the filing but could not independently confirm the $74.6M line in the section of the document we retrieved, so it is attributed to secondary reporting in the text above rather than presented as verified. The gross margin improvement, which is the basis of this article, comes from the filing itself.
This analysis is based on publicly available filings and reporting, all linked above. It represents our editorial interpretation of those sources. We have no affiliation with, and receive no payment from, any company named in this piece. If you believe a figure here is wrong, tell us — corrections are made publicly and the article is marked as updated.
Put your budget and channels into the simulator and see the range published cases landed in.