,
17 tweets,
3 min read
Read on Twitter
So in @lyft's S-1 today they announced that they're paying $300 million a year to AWS between 2019 and 2021. (1/16)
"That's nuts!" people exclaim. "You could build so many datacenters for that!" Yes, you could, and no, you shouldn't. (2/16)
I have these conversations with clients constantly. "We would save money if we built our own datacenters." Yes, until you staff up to maintain them, and ensure that the communications patterns between teams worked, and... (3/16)
The cloud providers are better than you are at running infrastructure, full stop. Leave an application to rot in both your DC and in AWS, and only one gets cheaper / better over time. (4/16)
If your workload is JUST "store exabytes of data" or "run this one specific application on 2 million servers" then there's an economic story. For large scale varied workloads there generally isn't. (5/16)
It's not just the infra cost; it's the staff time cost. Worse, it's the loss of focus. Now you're focusing on building and running datacenters instead of your core business. (6/16)
Does anyone honestly believe that "losing focus" for a company like Lyft at $20-25 billion is worth whatever cost savings they could realize? It simply isn't. (7/16)
If you got started in this space as a hobbyist you default to thinking of your time as being free. It's not; it's a near-certainty that Lyft spends more a year on payroll than it does on its AWS bill. (8/16)
I say it again: the problem with cloud bills isn't that they're too big; it's that they're not governed well. Nobody at Lyft internally is likely to say "the bill is too high." They'll instead say it's hard to predict, and always growing. (9/16)
It's a very different problem--and one that "well shit, my buddy down at the computer store can sell you drives for half that cost!" completely misses. (10/16)
Lyft isn't dumb. They've run the numbers and made a choice that makes sense for them. It's not that nobody ever told them that there were other options available. (11/16)
But wait, there's more. Many companies pay about this on the cloud and it's a cost savings in actual dollars and cents. Hear me out. (12/16)
10 years of CapEx purchases on computers that somehow don't get turned off after 3-5 years is a heck of a lot easier to shuffle away into the murk of financial statements than a recurring single line-item contract commitment. (13/16)
Contrast that with 18 divisions each with their own P&L on different hardware acquisition cylces, and it starts to dawn on you exactly why Dell's 2012 revenue was over twice AWS's 2018 revenue. (14/16)
Nobody freaked out, because it just wasn't visible in the same way. (15/16)
...and you're really going to flip when you realize that there are a number of companies spending 5-10 times as much as Lyft is on AWS--none of whom are planning to spend less next year. (16/16)
CORRECTION: $300 million over three years. The management regrets the error.
