To illustrate, let's consider a company that buys a $100 data centre every 10 years. (To keep things simple, let's assume the data centre is worthless after 10 years.) The data centre generates $20 in revenues for each of those 10 years.
Instead of showing an $80 loss in year 1 and then a $20 profit in years 2 through 10 (with the expectation of another $80 loss in year 11), accountants smooth the numbers based on expectations. The $100 data centre cost is "depreciated" over the expected lifetime of the asset. So one might account $10 of the data centre's cost to each of its ten years, thereby producing $10 of profit each year. This better reflects economic reality.
Ebitda does not include depreciation. The aforementioned company's Ebitda would be $20 for years 2 through 10. This is a small problem in year 2. But if you're an investor in year 10, ignoring that depreciation is the flip side of capital expenditure, you're in for a nasty shock when year 11's predictable capital expenditure comes down the line.
https://techcrunch.com/2016/06/14/dropbox-says-it-is-cash-fl...
Therefore, free cash flow in any single year does not say much, only when considered over many years.
The key question is, then, those large capital expenditures that are accounting for the large depreciation charges, were they truly one-off, or can we expect them to be recurring?
Having said that, I have made mucho dinero off of investing in companies that have negative earnings due to high depreciation charges from truly one-off expenses (infrastructure investment, foolish purchases by previous management, etc.). Eventually the depreciation tapers off and earnings fall in line with fcf, and I make money.
So if AWS cost $10/year, would EBITDA be $10?
EBITDA is Earnings Before ITDA. Quick example, with super simple math:
Scenario A (AWS)
- Over 5 years, you bring in $1M per year in revenue
- Every year you pay $200k in AWS fees
- Your earnings (Net Income) are $800k per year, your EBITDA is $800k per year
Scenario B (Buy your own hardware)
- Over 5 years, you bring in $1M per year in revenue
- You pay $1M in year one for hardware and depreciate it over the next 10 years at $100k per year
- Your earnings (Net Income) are $800k per year, your EBITDA is $900k per year
There are a ton of nuances to this, but this is generally how it works. As you can see the depreciation schedule affects this a lot, as well as when you choose to make the investment. Hence why it's feasible that Dropbox took out it's largest operating expense, i.e. AWS storage, which literally overnight increased EBITDA. It just so happens to be that a year ago they switched over to their own data center, and now they are touted "EBITDA profitable".
Long story - these stories are kinda stupid unless you get to see the full income statement, balance sheet and cash flow statement.
Maybe you can further explain the difference between cashflow positive and EBITDA?
The aggressive depreciation schedules might be due to the more compute-intensive use cases. Most servers are replaced due to new technology, mainly power consumption improvements. Maybe for low power scenarios the ROI bar for replacement is a lot higher, meaning low power servers get used longer than average. I would expect that Dropbox mostly uses these (I would assume they are more IO bound than compute bound).
See here for more info: http://www.investopedia.com/walkthrough/corporate-finance/2/...
Weird tangent, but there are methods to classify AWS spend as CapEx; while your typical startup isn't going to bother with it, companies spending in the 8 figure and up range annually on AWS annually do tend to go down that path.
A terabyte per second, am I reading that right? It sounds... Insane?
FYI - it's very likely that Dropbox is depreciating a HUGE ($100's of millions) amount of data center cost.[0] Translated into a more common "AWS opex cost", might yield a company that isn't really profitable^. EBITDA can be very misleading (not saying it definitely is in this case) so it's always worth analyzing it with a grain of salt until you have all of the full granular details.
[0] - http://www.datacenterdynamics.com/content-tracks/colo-cloud/...
^EDIT - I should have said, "might not be profitable yet".