Digital Ocean is a great company in a brutal, low-margin industry. Based on having run a similar (now mostly defunct) company in the past, I would guess that 80% of their customers are on the $5/mo plan.
That $5/mo has to cover the hardware costs: you're buying expensive physical servers to put the VPSes on, and lots of SSDs too. SSDs have a limited lifetime measured in writes, and some of your customers will leave broken programs running that chew through this precious resource for no reason. If you throttle them, they'll complain.
Then there's the support. Handling a support ticket costs you at least $3 in salary and benefits (remember, your typical customer pays $5/mo), and people will demand that you help them fix their broken MySQL server or whatever. They'll yell and threaten when you tell them that this is outside the scope of what you can do.
And don't forget security. Your customers will install broken-ass Wordpress sites and forget to upgrade them for 5 years. Then a worm sweeps through and now a whole bunch of them have been pwned and are mining cryptocurrency. Those pwned customers are complaining and demanding that you fix it, and the regular customers are also upset because of slowness due to the "noisy neighbor" problem inherent to all VPSes.
Speaking of which, preventing one VPS from hogging all the CPU or disk bandwidth is harder than it looks. The two dominant software platforms are Xen and KVM, and neither gives you great tools for dealing with disk bandwidth. Limiting CPU is much easier, but there's still the problem that you're overselling. Which is fine until half the VPSes on your machine are trying to mine Ethereum.
On the bright side: half your customers will buy the VPS, leave it running, and forget about it for years at a time. That's what makes the $5/mo business model work out.
Anyway, I do hope they can become profitable! They run a much better operation than the incumbents they replaced (slicehost, etc).
Person whose job title has "senior network engineer" in it here. I work for a mid sized regional ASN doing middle-mile and last-mile transport and transit. Quite intentionally we offer no services to customers related to VM hosting or managed hosting. The closest we come is selling rack space/cooling/power to people who want to colocate their own equipment and be fully responsible for it themselves.
We are considerably smaller than DigitalOcean in staffing head count. But we have a network that is spread out geographically across five states and 30+ cities and towns, built a combination of third party lit L2 transport, dark fiber IRUs, and fiber we built ourselves. We're a facilities based WAN provider.
There are so many different possible types of ISPs. For a small organization it only makes sense to decide whether you want to go after a huge number of $5 a month customers, or if you want to focus your time and effort on customers that spend anywhere from $250/month upwards for last mile broadband services, colocation/hosting services, etc. As a generalization, the higher the dollar value of the customer, the less of a headache they are, and the higher the clue level of the customer is.
I concur with 100% of what the above poster says about the hosting business.
Bulk hosting/VPS/VM hosting is an incredibly brutal race to the bottom in pricing. Extensive well crafted automation tools and massive economies of scale are the only thing that will save you. I truly feel sorry for the people who are working (mostly entry-level) jobs doing first tier technical support/customer service for 5 dollar a month VPS customers.
If somebody wanted to hire me to work for a consumer-facing hosting company I would run away screaming. It's my idea of a personal hell in the ISP business. Those who have found a way to make it work, not go bankrupt and not have mental breakdowns are a rare breed.
I've had a "droplet" going at $7/mo for about half a year and probably used it less than ten hours so far, and not doing anything too heavy when I use it. Every time I think about turning it off I say "yeah but it's only $7 and I like to ssh into it sometimes."
I've had a "shared hosting" setup on Dreamhost for at least 15 years, also using pennies per month in capacity, at most.
I get that it can be a brutal, low-margin business, but I also wonder how many "small" customers are extremely high-margin like me, and whether that can aggregate into a better overall margin than you might guess?
Or will you always have a few outliers running at capacity and calling the help desk and blowing out your margins?
One of the interesting metrics from my time working in global escalations for one of the worlds largest electronics manufacturers is that most customers don't call in for support, but on average we get 3.5 calls per user per account lifetime.
The 80/20 rule in in full effect here. The 'needy' customers are extremely resource intensive.
I've sat in on some really gross calls in which my CTO at the time berated DO support for issues that didn't _really_ exist in an attempt to get better rates, or practically yelled at them when minor hiccups occurred in our infrastructure. It was very painful to sit in on. They were remarkably patient and cool about it. I suspect there are many people like him out there sucking up their time and resources.
This is probably what I would do. Keep around the high margin customers like the OP, but do something with the Complainy Pants Inc; either force-migrate them to a higher cost service tier, throttle them until they leave, or maybe just slow down their service.
The key to remaining competitive in cut-throat industries is knowing where to spend your limited time and money.
"either force-migrate them to a higher cost service tier, throttle them until they leave, or maybe just slow down their service.
The key to remaining competitive in cut-throat industries is knowing where to spend your limited time and money."
Shit, you've just reminded me that I have to finish some ansible playbooks I started in september. The VPS they were supposed to configure only had some hardening an no traffic since then.
you, personally, may be a high margin customer, but some,
and perhaps a lot, of the $5/no customers are potential liabilities due to not patching software or libraries or choosing terrible passwords for their services, databases, etc.
one decent incident can cost a multiple of a year's revenue for the account.
in my experience, it's more about trying to keep your IP space clean for the rest of your customers and bandwidth. You either get your IP space blacklisted or some idiot starts burning 1Gbps on ssh attempts and you start getting abuse email from the .mil space which is awkward. Plus, you gotta pay for the bandwidth.
That 5/mo customer is much more likely to disappear than pay for the data they used.
no, more like someone's instance gets pwn-ed and is now part of a botnet and DO is getting calls and or isp-blocked and has to devote staff time to the incident.
Paid them for four years. If I had known they were going to delete the backups in just over one month, I wouldn’t have risked that.
Lost everything when they deleted the backups. They turned off the server on the same day, and it was running fine right up until then. Zero warning other than the emails I didn’t see.
A customer for 4 years and they delete everything after 5 weeks? Can you really defend that? It's pretty clearly a bad policy. You're being facetious about other contact methods, but I think that's a great idea. Allow users to add a phone number. Some people have faulty spam email blockers.
I had a similar situation, but, a completely different outcome.
I had my wallet stolen on one of my work trips, and had all of my credit cards deactivated and new numbers issued. Digital Ocean's emails went to my "bulk" folder, and I missed the notifications saying that they were going to disable, then deactivate my service.
What caught my attention was one of the support guys emailing me and saying basically, "Hey, I see that you've been a customer for a long time -- we're going to give you 30 more days to fix this (and btw, you still owe us for your services so far) before we delete everything."
That support interaction -- that wasn't flagged as spam -- is literally what kept my data, and it keeps me using Digital Ocean: amazing support.
How long to hold it for though? If say 90% of customers that haven’t paid for 5 weeks end up never paying that is resources they will never recover the cost for.
What they could do is work out some high restore cost which makes the numbers work on holding all that data for longer.
You keep it until x days/weeks after the servers are deactivated. Where the emails are going unnoticed, the servers getting cut off usually is noticed straight away. The story here appears to be that the backups were deleted at the same time as the servers.
For the relatively small cost of those who don't subsequently pay up, you get a lot of goodwill from the majority who just messed up.
When an account on DO is late with a payment, you go through three stages: hold, suspension, termination.
The hold happens right away and prevents you from creating new billable resources.
Then a suspension happens several weeks later, which powers off your servers.
You don’t get terminated until a couple weeks after that, at which point the data is irretrievable.
The time between each stage can vary depending on how long you’ve been a customer or what your monthly payments in the past have been, but that’s the gist of it.
It'd be nice if one could pay a bit more in advance, so one's data stayed some weeks or months longer than default, after the servers got powered off. — As an extra safety net, if the credit card expires when one is in bed for a month with a broken leg or some bad luck thing like that.
A lot of people seem willing to blame you without knowing the answer to this question.
On a different note, this kind of thing is probably one of the top threats to the survivability of a cloud-powered site or service. Alongside vandalism by disgruntled (ex-)employees, I imagine it happens far more often than outages in cloud providers' infrastructure.
The point of backups is that they still exist after something went wrong. Something went wrong, the backups didn't exist, DO didn't do the job it was paid for.
DO reserves say $20 (just an example) when one spins up a new droplet, and if one forgets to pay — then, DO shuts down the server. But keeps the backups, until the GB-month cost is $20.
One could choose how much money to reserve, depending on how important the data one stored on the droplet, was. If it's just for running test: $0. Customer data: Maybe $$$ instead.
There could be a default that made the data and backup last for 2 or 3 months?
And any remaining money would be refunded, if one closes one's account or sth like that.
* * *
Actually I'll look into implementing this, in my own SaaS (which, like DO, takes monthly recurring payments and stores customer data).
I believe a solution like this is an elegant way to ensure that the customer gets what they 'believe' they are buying which is a sure-fire way to keep the customer happy.
The other folks here are giving you a hard time, so I'll just say I actually see things differently here. Things get overlooked, card expiration dates being one of them. Deleting a backup after such a short time is really poor form. Especially because these backups only exist because people need contingencies for machine and human error. Sending a threatening email 'fix your CC in 60 days or everything goes bye-bye' is preferable for almost everyone in this situation, and the difference in cost to DO would be negligible.
DO sends an email when they can't charge. They send a second email 21 days later, when the account gets suspended (which doesn't deleted anything), with essentially your message, except it's 14 days instead of 60.
It's five weeks really poor form but eight acceptable? Seems a bit subjective.
DO reserves say $20 (just an example) when one spins up a new droplet, and if one forgets to pay — then, DO shuts down the server. But keeps the backups, until the GB-month cost is $20.
One could choose how much money to reserve, depending on how important the data one stored on the droplet, was. If it's just for running test: $0. Customer data: Maybe $$$ instead.
There could be a default that made the backups stay for 3 months?
Any remaining money could be refunded, whenever one wants.
6-week vacations aren't that uncommon, which is a scenario I would expect to encounter if I was DO. If that email doesn't get seen for 6 weeks it definitely shouldn't result in all backups being deleted. Beyond that, it's not unreasonable to imagine health issues or other external circumstances putting someone out of commission for 6 weeks (or even more, making it quite subjective).
All in all, from the moment of non-payment to the moment that everything you have on their servers is gone shouldn't be less than 90 days in my personal opinion. If that raised the cost of my backups by a few dollars, so be it. The backups are there to account for things I may not have been able to consider.
I may have said vacations, but people take time off work for any number of reasons. Paternity/Maternity leave, health, family needs, and so on. I had a 2-month lapse when I moved from Germany to the United States for example.
If there were, for any reason, an internal lapse in communication or processes (imagine small companies...) that could result in a pretty sour situation.
This discussion could be easily settled by comparing the grace period to the ones provided by Vultr and Linode. Then compare all of them to Amazon, which I would bet blows everyone else out.
This is a huge fear of mine. I tend to do month-long "research" stints where I cut myself off from the world. But I'm yet to set up any backups outside of DO. Also a fear with losing domain names because with namecheap you can't set up any "backup" payment method.
I haven't decided whether DO did the right thing or not, but to be fair, your analogy should be more in the line of "they evicted me and burned everything of mine in the apartment".
But landlords kind of do that... there’s even an entire TV show dedicated to it, called Storage Wars. People didn’t pay the bill for their storage locker so the owners of the locker sell off all the stuff inside.
Context really matters though. Is the storage locker being bidded off because the owner died 5 years ago? What if the last payment was 8 months ago, and you haven't heard from the owner since?
It's a completely different story if you were late by 1 day and they sold all your stuff the next day.
I guess the question here is if DO waited long enough before deleting their droplet. They were over a month late, and I'm assuming they sent notifications....
They also have a referral program that turned DigitalOcean to free hosting for me. A few years ago I got annoyed with something, wrote an article about how to fix it using a DO instance, linked to DO with my referral code and racked up thousands of $ in a payout. I used most when experimenting and feeling too lazy to shut down instances(because I didn't want to go through backup since maybe I will use it later etc).
For 6-7 years now I'm using DO for free but my only resource-intensive instance is a personal VPN server that I use from time to time(a few hours a day maybe?).
At first, DO use to give referral payouts in cash, then they limited the payouts to credits, then changed the structure and introduced expiration date to these credits.
I would guess that this referral program that probably helped them a lot with the growth at first is now a burden.
I don't quite understand the economics of referral programs. I'm dealing with levels of customer support to get signed up for a $120/paying signup referral program (cash not credits).
I don't understand
a) How does this possibly make them money?
b) If they want signups that much, why is the flow for joining the referral program so buggy?
My point was it seems insane to me that the lifetime value of a customer who has billed something to a credit card would be over $120, especially in the space of Email as a Service. I would guess (without any knowledge) that it'd actually be closer to $12 because of all the people who try it out and then never continue.
They take in less than the referral payout in revenue for a customer who sends tens of thousands of emails a year for five years based on their pricing page.
> a) You make money on people building large architecture on it, and running out of credits and relying on laziness to move it somewhere else.
My interpretation of their marketing pages is that they pay out referral payments in cash (actually SWIFT, but...) once a signup has something charged to their credit cards. So they can't rely on unused credits. And they lose money on every referred signup that makes them less than $120 in profit.
This seems insane to me, because someone sending tens of thousands of emails to hundreds of people a year gets them less than half that in revenue.
I did similar with Dreamhost back in the day, linking through comments and my signature. Great cheap advertising for a high karma to be recommending you, and I was using them.
> That $5/mo has to cover the hardware costs: you're buying expensive physical servers to put the VPSes on, and lots of SSDs too. SSDs have a limited lifetime measured in writes, and some of your customers will leave broken programs running that chew through this precious resource for no reason. If you throttle them, they'll complain.
This makes it sound like you could get a big win in the VPS-provider space by drawing an ROI line at ~$20, and making all instances below that size diskless, with their rootfs being either a tmpfs overlay of a shared SAN-mount of a base image (like a LiveCD environment), or a tmpfs into which was dumped a PXE initramfs image (as e.g. CoreOS does in its idiomatic deploy style.)
I feel like many customer use-cases would still be satisfied by such instances (especially if you also offer local object-storage for the diskless instances to interact with.) It'd sort of be a hybrid position between ephemeral PaaS containers, and actual persistent VMs.
Anyone know of a provider that provides low-cost long-running diskless VPSes like this?
I use Scaleway which does it this way. It was very nice to be able to detach my rootfs when my CPU blew up and just make it the rootfs of another instance.
You could. You’d just need some kind of file system driver that connects the application to a more stateful disk so as things pass through temporary memory they’re written to an SSD or HDD.
Of course that already exists, and DO offers it for $5.
> Digital Ocean is a great company in a brutal, low-margin industry. Based on having run a similar (now mostly defunct) company in the past, I would guess that 80% of their customers are on the $5/mo plan.
Well, to be honest, Vultr has 2.5$ (IPv6 only) and 3.5$ plans. So, if they're getting by, so could DO.
I actually migrated from DO to Vultr because at the time DO offered 512 MB RAM for $5, while Vultr offered it for $2.5. And Vultr gave me $50 bonus platform credit on sign up, valid for about 18 months (accounting for possible overage fees).
I'm still on Vultr, 3 years on. No problems at all, other than billing issues (accidentally was assigned Australian VAT despite living in Serbia), I had no support tickets. After some time I started using more instances, and more powerful instances, and more services (block storage, "portable" IPs, object storage, internal networks, etc).
I've had a lot of problems with DO's Object storage which was also one reason to move away from them. Problems were quite catastrophic in nature, i.e. the files were unavailable for a few hours every few weeks.
Recently I ave got into Upcloud.com, they have [1] flexible plan that you could mix and match resources, allow me to spin up 20x vCPU, 1GB RAM, 10GB SSD for $168 / month, or 4x vCPU, 128GB RAM, 10GB SSD for $550 / month.
> They'll yell and threaten when you tell them that this is outside the scope of what you can do.
They will, and you will tell them to leave. Unmanaged VPS' are unmanaged.
> Those pwned customers are complaining and demanding that you fix it
Tell them to leave, they are paying $5/mo after all.
> regular customers are also upset because of slowness due to the "noisy neighbor" problem inherent to all VPSes
Throttle abusive users, and hand-wave it via AUP, TOS, etc.
> neither gives you great tools for dealing with disk bandwidth
That could, should and will be fixed sometime.
If they ramped that $5/month to $20/month after X months, I would keep paying it. Digital Ocean is a fantastic deal, everything I've tried straightforwardly works, and the fact that I can't accidentally spend money as I experiment is a real boon.
Are you absolutely sure about that? There are several other services at $20/month that would offer a much more compelling value proposition compared to Digital Ocean's lowest-tier droplet.
I would definitely have signed up for a competitor if DO droplets cost 400% what they do now. As other commenters have said, this space is a brutal race to the bottom in pricing. DO is great, I might pay 50% more for their current services, but not 300% more.
Yes. Digital Ocean got me in the door for $5, but thus far has they have done everything I needed it to do, with clear documentation for how everything works, and no nasty surprises.
As someone doing proofs-of-concept and experimental work, the resource I'm trying to conserve is my own time, and Digital Ocean's value proposition on that is great. Everything works really straightforwardly and I can concentrate on my own code.
I wouldn't suggest Digital Ocean change their high-volume pricing, but the experience they provide to the small-scale user is worth a premium.
Yes, this. I was paying $24/mon after the backup service for something that basically just proved that I owned a domain. For years, using basically just 1 IP address.
In terms of similar competitor and ignoring those Cheap VPS, they started the whole $10 and later $5/month price plan. Linode has always maintained its $20 / Node starting price arguing for the exact reason you mentioned, Support Cost. And later DO / Linode became the price plan standards where everyone follows.
I remember at the time I suggested $5 plan should be limited to 1 per account or only for non- public internet facing usage. But the $5 plan made lots of headline and new customers during the growth at all cost stage.
So if $5 plan were really the problem that it was really their own making. Having said all of that I dont think $5 is really their concern. Hardware is cheap, and those plan with vCPU are shared and always over sold. The number of bad actor within the lowest plan are statistically quite small.
I actually think the future should be more like Render[1],
Linode is a sponsor to my YouTube channel. They have 5.00 plans and have for well over a year now. 10.00 plans too. Linode seems to be doing quite well. They tout themselves as the largest privately held cloud hosting company in the world and keep opening new locations.
DO makes a great product. I've been using them for many years and I hope that they succeed.
I've used other hosts in the past and had nothing but trouble. Joyent ended one of their hosting plans and I had to migrate EVERYTHING which took forever. Then Rimuhosting had an actual hardware failure that resulted in non-reproducible errors happening very frequently - that company nearly brought down my whole business. Then there was Serverpronto which had too much downtime.
by comparison, DO has been much much better, always up, always trouble free.
"Legacy Service End of Life" here are the details:
"We've been analyzing customer usage of Joyent’s systems and noticed that you are one of the few customers that are still on our early products and have not migrated to our new platform, the Joyent Cloud.
For many business reasons, including infrastructure performance, service quality and manageability, these early products are nearing their End of Life. We plan to sunset these services on October 31, 2012 and we'd like to walk you through a few options."
So, I had to migrate everything to a new host which was a huge pain at the time because i was running a live Strategy game which are very time consuming to move with a minimal amount of down time, disruption and risk. not to mention everything that goes to changing over the DNS servers to point to the new address. Facebook login also added another layer of complication. It's all doable, but it takes a lot of planning and risk mitigation to get it 100% right with an absolute minimum of downtime.
Other than that, joyent uptime and performance was fine. But, having to move hosts is a big deal breaker.
> Based on having run a similar (now mostly defunct) company in the past, I would guess that 80% of their customers are on the $5/mo plan.
I wonder if the Always Free tiers of GCP / OCI have helped DO in this regard. You can get a lot of free VMs these days, so maybe other cheapasses like myself have left the DO / Linode / etc platforms.
I sympathise with your customer-support point, but not the others. Customers are paying for cloud resources. If their demands are too much for your infrastructure, the issue lies in your infrastructure, or in your claims to customers.
A customer's need for additional resources should translate to a price-point question, rather than to uncertainty about what they've already paid for.
> SSDs have a limited lifetime measured in writes, and some of your customers will leave broken programs running that chew through this precious resource for no reason. If you throttle them, they'll complain.
High IO doesn't always mean an instance was compromised.
There should be clearly defined limits, and/or a clearly defined throttling policy, and the customer should have the option to buy their way out. Amazon gets this right. There should be no guessing game about reasonable use, or goodwill.
> Those pwned customers are complaining and demanding that you fix it, and the regular customers are also upset because of slowness due to the "noisy neighbor" problem inherent to all VPSes.
High CPU load doesn't always mean an instance was compromised. If you've sold CPU resources, the customer is entitled to use them. Obvious example: build servers.
If other customers experience unacceptable degradation, that means you overpromised, or else your isolation solution isn't fit for service.
Again, Amazon gets this right. They're criticised for their complex billing schemes, sometimes rightly, but it clearly makes sense to measure and be explicit about all resource-consumption. They even have an elaborate scheme to incentivise customers to tame down their CPU usage, in the form of 'burstable performance instances'.
> Anyway, I do hope they can become profitable!
Agreed. It's good to have smaller players, not just the big three of Amazon/Google/Microsoft. Competing on price-point without having the same scale, must be really tough.
> That $5/mo has to cover the hardware costs: you're buying expensive physical servers to put the VPSes on, and lots of SSDs too. SSDs have a limited lifetime measured in writes, and some of your customers will leave broken programs running that chew through this precious resource for no reason. If you throttle them, they'll complain.
they are heavily throttled on disk I/O, so much so that I had to switch to AWS and pay per I/O for one project.
also the network seems throttled to 100Mbps up/down, and a few TB/mo, something which Scaleway for $3/mo is unlimited TB/mo and at certain times 2.5Gb/s
Maybe instead of using expensive SSDs. A topology of many spanning disks in large ZFS clusters by using PCIe HDD controllers. Then link the machines via 10GBe could provide you the speed and performance you require require. I've set up a moderate size pool of 1Pb across 16 physical servers on 4 full size racks. This cost less than 50k. Electricity and cooling come from solar. Its the damn internet connection for people to access it that is the cost killer.
> Its the damn internet connection for people to access it that is the cost killer.
Nope. Any hosting provider will be located at a carrier neutral datacenter or the like. At any of these you will have access to low cost IP transit providers and Internet exchanges. You can buy 100G IP transit for $5k per month, so Internet cost isn't really an issue.
You know, there are some people that just like to do things themselves so that one day instead of buying from the big guy like everyone else, they become the big guy. You can build a distributed datacenter yourself for 500$ a month that can compete with the more expensive network until you grow your user base large enough to afford it. Sometimes VC magic is a bad thing because when you don't have that option, the only other one is to innovate.
Oh yea it's always easier with scale. The main point I was trying to make is that the hardware is actually pretty cheap compared to the network costs to link it. Especially when large companies dump 2 year old servers for pennies on the dollar because the electricity costs at their scale justify it.
Side note: I took that cluster and split it into 8. Then moved them to different geographical locations and where I could use friend or families residential connections and a cloudflare cluster to offset my cost in exchange for unlimited hosting. Very similar to Ceph actually
> Based on having run a similar (now mostly defunct) company in the past, I would guess that 80% of their customers are on the $5/mo plan.
The founders of NordVPN have recently invested in Hostinger[1], which has successfully adopted their extremely profitable pricing model: charging for 2-4 years in advance, by default. This way, even those who would have paid $5 / month and cancelled and a few months later, end up spending $100+ for 24-48 months at once, often without having a clear need for it, thus leaving a lot of resources underutilized – and available for overselling. The company has more than doubled in size in the last 3 years, more than a decade after its inception.
Just to be clear: I don't find such pricing models customer friendly, and this is just an example of how some companies manage to find their path in such a "brutal low-margin industry". Since shared hosting – just like VPN – is a commodity product, they primarily focus on marketing, sales, and support.
I like how simple their API is. I have infrequent cloud needs, so it's nice being able to set up a simple docker-machine script once in a while to spin up a ton of compute nodes for some scientific task.
Digital Ocean is a great company in a brutal, low-margin industry. Based on having run a similar (now mostly defunct) company in the past, I would guess that 80% of their customers are on the $5/mo plan.
And they are competing with AWS Lightsail that have similar prices and offers Windows instances for people who want it.
But even though I am very steeped in the AWS ecosystem and the price of Lightsail is competitive, if I just needed a VPS I would still go with Linode. I can’t imagine AWS’s support being good for anyone who doesn’t have a business support plan.
They charge you for everything. Data transfer in: charge, data transfer out: charge, DNS lookup: charge, storing a file: charge, etc.
It's one of those situations where it's a death by a thousand cuts. When you're cost provisioning, you can figure out the big stuff, i.e., we need 25 EC2s m4a.2xlarge instances with 40TB of S3 storage, and a 2TB Aurora instance. But once you get the bill, you start seeing the costs of ELBs, NAT gateways, inter-region transfers, etc. Individually, these costs aren't significant, but in aggregate, they can make up a health chunk of your monthly bill.
Plus, their managed solutions are fucking expensive. We moved a self-managed ELK cluster to an AWS-managed one and the costs went up by a factor of two.
This is why I don't use AWS for personal projects (S3 is the only one I'd consider using). EC2 for example is significantly more expensive than a Digital Ocean droplet.
Even S3 can hit random expenses when you least expect it. I let a process go crazy and create billions of objects in S3. Of course AWS was happy to charge me for all of this, but what I wasn't expecting was the charge just to list the objects to delete them. I spent more than $100 just to empty the stupid bucket out.
Perception that they are THE cloud player to use with great reliability and enterprise focus, so people support their protocols and develop their products for AWS and end up with a bit of lock-in. Mid to large enterprise continue to get on the cloud train hype and spend tens, to hundreds of thousands a month on AWS/S3 instance costs and cannot figure out how or why the bill is so expensive. Then they spend internal costs and personal resource time trying to figure out how they spent so much money on AWS. This happens where I work on both the test/dev and also production hosting.
If DO or Linode were an option to consider, we might be able to save so much money but our own customers use and believe in AWS so we develop and test on AWS. It's a bit of a vicious cycle.
AWS has few services that I imagine yield massive margins with not that much cost to them, which can mainly be justified by the services being pay-per-use and having a small overhead for the customers. And then, I believe most of the AWS's profits come from its enterprise customers which they have plenty of (who aren't in turn that price-sensitive as long as they see profits from their end).
DO on the other hand from my experience has been catered towards small businesses or hobbyists who simply just don't bring that much money vs even a one giant company will bring. And since DO doesn't have all the goodies of AWS, it can't really directly compete with it for those big customers.
It's a very good question. I would imagine more economies of scale? but, I would have thought that DO has enough economies of scale to reap all the benefits. but, I'm not sure.
My impression is that DO is doing just fine. I get the sense that the layoffs are just to add to the profitability, not necessarily a sign of weakness. The article did mention that they were still growing revenues very aggressively.
Speaking as someone with extensive background both in cloud and hosting:
Cloud hosting is low margin. You sell IaaS at about the cheapest possible price point you can. That's the very definition of low margin. Economies of scale don't enter in to whether or not it's a low margin business, they only define how competitive you can be in a low margin business.
IaaS is not where you make the money. The margins have to be tight to be competitive because that's the dollar value people see first when evaluating your cloud platform. It has an immediate effect from day one.
The profits are not made on IaaS, but on the PaaS and SaaS solutions that you, as a cloud provider, build on top of the IaaS. Things like your DBaaS, Streaming, Functions, Load Balancers, Data warehouse etc. products.
Once they're on your platform, that's when you try to get them to pivot. "Why spend engineering effort on running and maintaining database servers, when we can do it for you immediately?" Of course, then once they're using your value-added solutions, they start to get towards vendor lock in, every business's favourite situation. A customer that can't leave!
It's a difficult balancing point, you want to make it seem to the customer like they can realistically leave any time they want, but you don't want them to so you do just the absolute bare minimum you can get away with to make it seem like they're not locked in to your platform.
It seems like it took Digital Ocean a long time to realise that they need the SaaS and PaaS components if they're going to be in this for the long haul. When we launched Oracle Cloud Infrastructure some 3 1/2 years ago, we launched with features that Digital Ocean hadn't yet bothered with, and we were trying to launch with what was seen as the bare minimum to be a viable cloud product.
Prior to those services existing, it was relatively easy for any customer to just drop Digital Ocean for another cloud provider, but even those services aren't a big lock-in for customers.
I sincerely hope it's not too late for them. I like Digital Ocean. They really shook things up when they first hit the market, by bringing something a little different to the plate, but that was never enough to survive and it seems like they only relatively recently realised that.
That $5/mo has to cover the hardware costs: you're buying expensive physical servers to put the VPSes on, and lots of SSDs too. SSDs have a limited lifetime measured in writes, and some of your customers will leave broken programs running that chew through this precious resource for no reason. If you throttle them, they'll complain.
Then there's the support. Handling a support ticket costs you at least $3 in salary and benefits (remember, your typical customer pays $5/mo), and people will demand that you help them fix their broken MySQL server or whatever. They'll yell and threaten when you tell them that this is outside the scope of what you can do.
And don't forget security. Your customers will install broken-ass Wordpress sites and forget to upgrade them for 5 years. Then a worm sweeps through and now a whole bunch of them have been pwned and are mining cryptocurrency. Those pwned customers are complaining and demanding that you fix it, and the regular customers are also upset because of slowness due to the "noisy neighbor" problem inherent to all VPSes.
Speaking of which, preventing one VPS from hogging all the CPU or disk bandwidth is harder than it looks. The two dominant software platforms are Xen and KVM, and neither gives you great tools for dealing with disk bandwidth. Limiting CPU is much easier, but there's still the problem that you're overselling. Which is fine until half the VPSes on your machine are trying to mine Ethereum.
On the bright side: half your customers will buy the VPS, leave it running, and forget about it for years at a time. That's what makes the $5/mo business model work out.
Anyway, I do hope they can become profitable! They run a much better operation than the incumbents they replaced (slicehost, etc).