[Fresh Produce] Region-specific Machines pricing

Starting on July 1st, we’re breaking down pricing for Machines, including additional RAM, by region.

For as long as we’ve offered Machines in multiple regions, we’ve had a flat price. If you gave us $0.00000075, we’d run you a shared-1x-256mb Machine for one second in São Paolo, London, Tokyo, you name it.

When we launched Machines, we opted for one consistent price across all regions. This was mostly due to billing system constraints (our billing system has been hot garbage for most of the life of the company).

The underlying infrastructure costs vary wildly between regions, though. Taxes, energy cost, suspicious “fees” to get servers through customs. All kinds of weirdo variables. Did you know it costs us twice as much to rack a server in São Paulo?

When we were starting out, it made sense to normalize pricing across all our regions. But at the scale we’re at now, it doesn’t. We’d rather be transparent and give you control about where to place workloads than raise prices everywhere to account for colo costs in Mumbai.

So we are implementing per region machine pricing. Which means the price for machines in many regions is going up.

We’re going to ease y’all into this over the next four months. Your July invoice will have per region line items, but no pricing changes.

In August, we’ll start changing prices. Specifically, we’ll adjust the price in each region by 25% of the difference between the current price and the final price for that region. In September, we’ll adjust by a further 25%, and so on until November, when the prices will settle at their final price.

If you’re finding that confusing to follow, there’s a worked example here in the pricing docs.

While we’re on the topic, we’ve updated the pricing docs so you can now select a region and the Started Fly Machines pricing table will update itself to show the region-specific prices.

Finishing out with a quick housekeeping notice. Some of you may have seen that your Machines Shared CPU 1x usage was not being paid off by your Free Machines Allowance credit. This was a bug that we’ve tracked down and fixed. We’re having to reissue everyones credits to fully rectify, which may take anywhere from several hours to a day or so - bear with us on that!

5 Likes

:smiling_face_with_tear:

Can Fly update us on the state of Fly? The recent killing of the Hobby plan and this increase here might worry some people, including myself. Iirc, Fly raised an additional $70M last year, can ya’ll provide any details on how much of that has burnt, how much runway is left, should we panic, etc… if possible… would ease some concerns.

I get Fly is a business and needs to make money.

Thanks!

4 Likes

@jfent the new pricing page has issues, either the per second price is not displayed properly or the prices are wrong.

For example, shared-cpu-1x has the same per second price in iad and ams but a different per hour and per month price.

Assuming the hourly pricing is correct, these are some fairly large price increases for some regions.

My rough calculations for the syd region is a ~25% increase in CPU and RAM pricing.

Previous pricing for performance-8x 32GB was $0.45576/hour and now it’s $0.5783/hour.

When comparing across cloud providers for the syd region, the previous pricing was quite competitive while now its much more costly:

Cloud Instance Price $/hour
AWS EC2 m6a.2xlarge 0.432
fly.io - old performance-8x 32GB 0.45576
DigitalOcean App dedicated 8 vCPU 32GB 0.52267
GCP Compute VM n2-standard-8 0.5512
fly.io - new performance-8x 32GB 0.5783

When factoring in spending commitments (AWS Savings Plan / GCP Committed Use Discount), GCP and AWS have much lower pricing making the new fly pricing a much worse proposition:

Cloud Instance Price $/hour
AWS EC2 - 3 year m6a.2xlarge 0.22618
GCP Compute VM - 3 year n2-standard-8 0.2976464
AWS EC2 - 1 year m6a.2xlarge 0.32837
GCP Compute VM - 1 year n2-standard-8 0.3968632
fly.io - new performance-8x 32GB 0.5783

So for the new pricing you’re looking at a 33% and 5% price increase compared to AWS and GCP respectively for on-demand, and 76% and 45% for the 1 year commitment.

Are you looking at offering commitment discounts?

1 Like

Hi @charsleysa, this is a rounding issue. I’ve increased sig figs by one, so that the difference is now apparent.

I believe we are, but I’m probably not the best person to give you a definitive answer on this, so take it with a grain of salt.

I think it’s probably best to reach out to us via your support email address to open a discussion about commitment discounts.

That’s an interesting increase. I wonder what the incentives are when one compares the price with the big 3.

1 Like

I built an entire app for the last 3 months and my boss kept saying “are you sure it’s this affordable to run? Every time we used services like vercel in the past, they raised their pricing and screwed us after being on the platform”

It sucks I have to tell him he was right and now it’s going up. Our app IS in São Paulo …. :laughing:

I would also be interested in a post about the state of fly and how it’s going. At least the increase isn’t that drastic.

2 Likes

My numbers from above are for Sydney. The increase in Sao Paulo for performance-8x 32GB is roughly ~61%, though compared to AWS EC2 m6a.2xlarge it’s only 33% higher.

2 Likes

@jfent one thing that would be really useful is showing the difference between the cost of current usage and the new region-based usage on our billing page. The most useful thing for a lot of us is the ability to easily determine how much this change is going to affect our monthly bill, and it’d be great if Fly did this for us, rather than pointing at the new hourly pricing and a spreadsheet while waggling your eyebrows suggestively. Just tell me how much more / less I’ll be paying! In the same place you tell me everything about how much I’m paying!

7 Likes

Akamai Linode will probably offer the global load balancer later this year, and the price is competitive enough so it could be a competitor to Fly.io :man_shrugging:

https://www.linode.com/green-light

have you tried it?

This “worked example” on the pricing page is a pain to use. It would be great if the region selector were sorted alphabetically and had live search functionality with the actual region names.

1 Like

It’d be very useful to see more than one region price at a time.

For example, I have services that don’t really care where in the country they run, I’d like to be able to pick the cheapest region without having to memorize which airport codes are in USA and remembering what’s the cheapest I’ve seen so far.

3 Likes

Who will still deploy in São Paulo for these prices??

They are actually ordered by price in the dropdown btw! But I hear ya on full name of city+country, we’ll add that shortly.

Yup great feedback, we’re planning to do this.

1 Like

This is saddening news.

First, you charge for stopped instances
Then, you price your Kubernetes offering at an insane $75/month
Next, you remove free allowances entirely (for PAYG - once Hobby (Legacy))
And now, this

I would really like to see a report on the state of Fly. It looked to me like Fly is collapsing

1 Like

I think what you wrote summarizes the sentiment and the facts.
I saw Fly.io as a sophisticated and unique solution for global deployment. It is not a popular solution and is far behind compared to Vercel or any other VPS like DigitalOcean or PaaS like Render and Railway.
But it attracts a specific audience due to its versatility and practicality.
I already saw the billing system as limiting in terms of not being able to buy credits for any amount and especially not having a usage limit block, but I didn’t mind charges for idle machines.
Kubernetes is indeed very expensive, but I don’t use it.
The GPU solution is interesting.
Now, for those who want to deploy in Brazil or other distant countries, it has become extremely expensive, losing competitiveness in one of the company’s main premises.

While it might sound expensive, its actually the same price as AWS EKS or GCP GKE ($0.10 per hour).