It's a fair question.

Data centers are wildly expensive to operate if you want proper security, redundancy, reliability, recoverability, bandwidth, scale elasticity, etc.

And when I say security, I'm not just talking about software level security, but literal armed guards are needed at the scale of a company like Figma.

Bandwidth at that scale means literally negotiating to buy up enough direct fiber and verifying the routes that fiber takes between data centers.

At one of the companies I worked at, it was not uncommon to lose data center connectivity because a farmer's tractor cut a major fiber line we relied on.

Scalability might include tracking square footage available for new racks in physical buildings.

As long as your company is profitable, at anything but Facebook like scale, it may not be worth the trouble to try to run your own data center.

Even if the cloud doesn't save money, it saves mental energy and focus.

There’s a ton of middle ground between a fully managed cloud like AWS and building your own hyperscaler datacenter like Facebook.

Renting a few hundred cabinets from Equinix or Digital Realty is going to potentially be hugely cheaper than AWS, but you probably need a team of people to run it. That can be worthwhile if your growth is predictable and especially if your AWS bandwidth bill is expensive.

But then you’re building on bare metal. Gotta deploy your own databases, maybe kubernetes for running workloads, or something like VMware for VMs. And you don’t get any managed cloud services, so that’s another dozen employees you might need.

This is a 20-years-ago take. If your datacenter provider doesn't have multiple fiber entry into the building with multiple carriers, you chose the wrong provider at this point.

I work for a company making ~$9B in annual revenue and we use AWS for everything. I think a big aspect of that is just developer buy-in, as well as reliability guarantees, and being able to blame Amazon when things do go down

Also, you don't have to worry about half of your stack? The shared responsibility model really works.

No, you still do. You just replace those sysadmins with AWS dev ops people. But ultimately your concerns haven't gone down, they've changed. It's true you don't have to worry about hardware. But, then again, you can use coloco datacenters or even VPS.

They are preparing for next blog post in a year - „how we cut costs by xx% by moving to our own servers”.

A valuation is a just headline number which have no operational bearing.

Their ARR in 2022 was around $400M-450M. Say the infra budget at a typical 10% would be $50M. While it is a lot of money, it is not build your hardware money, also not all of it would be compute budget. They also would be spending on other SaaS apps like say Snowflake etc to special workloads like with GPUs, so not all workloads would be in-house ready. I would surprised if their commodity compute/k8s is more than half their overall budgets.

It is lot more likely to slow product growth to focus on this now, especially since they were/are still growing rapidly.

Larger SaaS companies than them in ARR still find using cloud exclusively is more productive/efficient.

Companies like Netflix with bigger market caps are still on AWS.

I can imagine the productivity of spinning up elastic cloud resources vs fixed data center resourcing being more important, especially considering how frequently a company like Figma ships new features.

Much bigger companies use AWS for very practical well thought out reasons.

Not managing procurement of hardware, upgrades, etc, and a defined standard operating model with accessible documentation and the ability to hire people with experience, and have to hire less people as you are doing less is enough to build a viable and demonstrable business case.

Scale beyond a certain point is hard without support and delegated responsibility.

> Why would they still be paying AWS rates?

They are almost certainly not paying sticker prices. Above a certain size, companies tend to have bespoke prices and SLAs that are negotiated in confidence.

There must be a prohibitively expensive upfront cost to buy enough servers to do this. Plus bringing in all the skill that doesn't exist that can stand up and run something like they would require.

I wonder if as time goes on that skill to use hardware is dissappearing. New engineers don't learn it, and the ones that slowly forget. I'm not that sharp on anything I haven't done in years, even if it's in a related domain.

There are a lot of ex-Dropbox people at Figma who might have learned firsthand that bringing your stuff on-prem under a theory of saving money is an intensely stupid idea.

> There are a lot of ex-Dropbox people at Figma who might have learned firsthand that bringing your stuff on-prem under a theory of saving money is an intensely stupid idea

Well, that's one hypothesis.

Another is that "Every maturing company with predictable products must be exploring ways to move workloads out of the cloud. AWS took your margin and isn't giving it back." ( https://news.ycombinator.com/item?id=35235775 )