They could do, and it might save them a few hundred million dollars, but the risk of project failure (lateness, budget overrun, etc) would be a lot higher. For example, committing $1bn to AWS with with a 5% risk of failure is a lot more sensible than committing $750m with a 25% risk of failure doing it yourself. It's entirely possible Snap's board would have refused to let them do it themselves if the risk was high enough.
It's not like they don't have the option to build it themselves in the future. This provides a lot of flexibility with very little liabilty to get their platform massive.
Besides they aren't like Facebook where they will grow their hardware slowly as they ramp up, Snapchat is already incredibly popular. It would be a massive undertaking to move everything to the data center. The experts who are good at that kind of thing are all already gainfully employed.
> This provides a lot of flexibility with very little liabilty to get their platform massive.
This is completely untrue. They are committed to spending that money with at least Google (whatever they don't use in infra, they have to make up with a true up payment). I don't know the AWS terms, but they would have to be successful at a level that is completely unrealistic for this to be financially sound (considering their growth numbers show users have tapered off).
"Snap’s daily active user grown went from single digit to flat last year"
If user growth is flat, and you're not yet profitable, the only nob you can turn is more ads or making ads more expensive for ad buyers. Neither option will endear Snap with either party.
It's much easier to go belly up with chapter 7 and not have to sell millions and millions (billions?) in physical assets then to simply say "sorry, we're out of business" to the contracts for future obligations.
I don't think they could build their own infrastructure cheaper than Google or Amazon. Both companies have very large scale and have been doing this for quite a time, have custom hardware etc.
For deals of this size, it might make sense for both Google and Amazon to put their margins quite low - as they know the alternative is that Snapchat just builds their own infra. For Google there could be also the PR factor involved.
It would definitely cost less, and datacenters are readily available, already running (i.e. zero actual project risk).
My guess is that they hope to use Google's TensorFlow and/or a lot of the GPU instances from AWS - TensorFlow needs a special chip and GPUs become outdated in 1 year, so it makes more sense to lease them as-needed.
I doubt their concern is saving money in this case, it's more about time to market for new products/features. If they can get a head start by spending more up front and focus entirely on their product, they may be able to grow significantly faster than if they kicked off a multi-year effort to role their own infrastructure.