The internet virtually exploded when Lyft released their S-1 in preparation to go public.
It contains a lot of information, but the thing most directly relevant to my area of expertise is the gem that they’ve committed to spend $300 million on AWS between 2019 and 2021.
Some corners of the internet have lost their collective minds over this with some fairly horrendous takes. Allow me to address several of them.
#1. They could save a lot of money by building out data centers!
Running data centers is hard. You will take downtime, and you’ll do a worse job of it than any major public cloud provider who isn’t Orac—wow, a process server at my door already?! That was fast.
My point is that there are more people working for AWS’s data center teams than likely work at most people’s entire companies (yes, there are notable exceptions). They’re frankly better at handling failures, they’re more practiced and polished at carrying out their procedures, and they’re astonishingly capable to the point where we forget that what they do isn’t easy.
“Disk failures” manifest as latency spikes; networking failures are usually invisible. On the rare occasions where there’s a broad systemic outage, that’s the day that nothing on the internet seems to be working quite right, anyway. You own your own availability (and there’s no getting around that), but for many use cases taking an outage when half the internet is broken is understandable–and minimizes your “headline risk.”
Counterintuitively, if you build an application and let it sit and rot, it’ll slowly get better and cheaper over time in AWS. “Price cuts, improved disk and network performance, and upgraded equipment” tend to beat “the last spare drive just failed in your RAID array and now your site is down.” Try that in a datacenter, and the raccoons will carry off your servers within five years.
It’s not just that data centers are cheaper (they’re very much not), it’s that they’re not as good as the public cloud.
#2. We never used to spend that much money on data centers. Cloud is a rip off!
Take AWS’s annual revenue from 2018. Good. Now double it. Keep going. Okay, now we’re at Dell’s annual revenue in 2012.
IT costs have been skyrocketing for decades. There are just a lot more tricks you can use to hide it from view when you’ve got 40 different divisions, you can treat the spend as CapEx (well, easily anyway; you can still do some of this with cloud spend but that’s a longer conversation), and you can spread it across multiple years rather than one gargantuan payment to one vendor. That stands out on the forms, and it empowers lazy journalism.
Folks who are freaking out about a $100 million annual AWS bill apparently don’t realize that a single fabrication plant for a semiconductor manufacturer costs upwards of 30 times as much. Lyft isn’t just using a bunch of EC2 instances and S3 storage; they’ve told us as much.
#3. But they could save so much money if they just…
It doesn’t matter how that sentence ends.
Look at Lyft’s financial position. They’re not profitable. They’re focused on growth and have some ambitious milestones that they need to hit in order to get there.
“Saving money” is absolutely not aligned with those goals from a strategic perspective. Spending too much money is not Lyft’s primary problem. If and when it ever becomes their primary problem, they’re likely in decline. Regardless, there will be opportunities to address that then.
#4. Nobody ran the numbers on this! If they went with (insert company I work for here), they’d save a boatload!
I find it a bridge too far to presume that people savvy enough to grow Lyft from “an idea” to “a reportedly $26 billion company on the verge of going public” somehow never bothered to run a cost/benefit analysis on a nine-figure annual budget line item.
Data centers cost way more than hardware, power, bandwidth, cooling, etc. There’s the human time; engineers aren’t cheap. More insidiously, the cost inherent to the loss-of-focus is incalculable.
“We’re taking downtime in data center 4 on Thursday night to replace a generator so now we have to tell 200 service teams about it and make sure they’re all okay with it” is a logistical nightmare you simply don’t have with a public cloud provider.
I’m not a cloud apologist; there are budget issues with cloud computing. Some workloads flat out aren’t appropriate at scale; Dropbox’s multi-exabyte scale storage issues realized a reported $74 million annual savings when they moved off of AWS and into data centers.
So, there are definitely times when it makes sense. I’ve just seen no evidence whatsoever that Lyft’s environment is one of them.
There are certainly things potential Lyft investors should worry about, like pink mustaches. As for its AWS spend? Nothing to see here. Move along.
