Do people still use Pinterest? I used to hear/see females in my family talking about it but not anymore. It seems like all Pinterest does today is destroy google image search.
The hubris of HN posters has no bounds. I’m sure that tweet gave you all you need to know to contradict the insiders responsible for making multi-billion dollar decisions… It’s probably the case that they’re all dumb, and the Internet rando has the absolutely guaranteed answer.
WeWork is proof that people making multi-billion dollar decisions aren’t always smart, and that’s even in a somewhat less volatile sector. When it comes to tech and the cloud hype I bet it’s even worse.
More middle men means more expensive. Regardless of what you think, you can't cheat the laws of conservation, they're built into the physical structure of our universe.
AWS is a middle man in the hosting business. Sometimes it's okay, but you can't cheat the fact that Amazon has to take their cut.
(This might change in the future when FAANG privatizes the global Internet for their own use, but we're not there yet today.)
And Toyota is a middleman between you and the raw materials to build a car. I guess it’s obvious that you can build a car cheaper than Toyota? How do the economies of scale fit into this false equivalency of a physics analogy?
Amazon's added value is in bundling open source tools and in provisioning and autoscaling. They don't make hardware, they don't generate electricity, they're not an ISP. They outsource the actual hosting part to other people.
If you just want to host stuff you can cut out the AWS middleman and rent a data center location for much cheaper. (Like, 4-5 or even 10 times cheaper.)
And yes, as a programmer/devops/sysadmin thinking about provisioning, autoscaling and packaging is your job. And yes, you absolutely can do it more efficiently than Amazon, because economies of scale don't apply to IT human resources. (Software projects don't become cheaper to make when you hire more programmers, quite the opposite, in fact.)
Companies don’t usually publish their definitions, but they also have to make an effort to exclude fake users or risk being held accountable by their investors.
Sure, but if I accidentally click on one of their links in google image search I'm "using" the website and not "fake", but $0.80 would be quite a bad deal ;) Whereas its registered users actively doing something on the site, not as terrible.
Believe it or not cloud providers don’t have unlimited compute elasticity. That’s a fundamental deceit of cloud computing. While Pinterest by itself may not make a difference, the accumulated guaranteed compute requirements across all such deals give AWS a pretty solid resource provisioning baseline to target.
I sometimes wonder about these huge deals like this...
Is there a bunch of AWS sales people who just became instantly "Fuck You Money"-rich as part of their sales commission on this deal?
I don't know how it works in FAANGS for sales commission, but I've heard of 5-10% being "normal"... so sales comp on this deal @ 5% comp is 150 mil? If so, are there loads of sales people just kicking around waiting for their "one big sale" and then insta-quit and go live on their own private island?
And if commission is capped, then surely the flip side is now there are loads of disgruntled sales people sitting around spitting teeth and annoyed that they just pulled in 3 billion for the company but they "only" got 100K capped-commission or whatever and so rage-quit and go somewhere else?
It's not clear what was sold by the saleperson since Pinterest could have just self-served themselves infrastructure off any combination of platforms. Presumably the idea is they're getting a discount in exchange for having this incentive favoring exclusivity. Their commission might also be amortized over the length of the contract (2029) since that's the length of time the value is generated.
This is not a new contract - Pinterest has been using AWS since at least 2013 (I used to work in a company that Pinterest purchased a service on AWS from, and Pinterest was our largest customer by far on AWS even then.) So maybe they've gone from 200-odd or 300-odd to 400 million an year. I bet that would affect the commission dynamics too, but I'm sure it's a decent chunk of change nonetheless.
Pinterest voluntarily entered this agreement, which provides better pricing in exchange for committed use - an offer AWS provides to many of its customers. This is a highly sensationalized headline.
Regardless of the long time horizon, it is still a significant number. Cloud is known to not be the cheapest option at large scale, so it is a bit curious they decided not to build their own infra. I'm sure the folks at Pinterest are smart cookies, so this is an indicator that cloud is now cost competitive even at scale (with these kinds of deals, which presumably come with a discount).
> "Cloud is known to not be the cheapest option at large scale"
Oh really? Ever try building, staffing, powering, cooling and equipping a data center? And then build a DR? Let me tell you, my company has foolishly done that just in the past ten years and WE are screaming uncle! We're now transitioning to AWS because it's much, much, much cheaper!
The biggest eye-opener? Not only is the physical part of the data center a super expensive ongoing expense - staff, server life cycles, power, property tax - but the software licenses (we gotta have support!) will EAT YOU ALIVE! I've transitioned many projects to AWS and was able to justify it on the ongoing licensing costs alone - never mind all the other costs related to building a data center.
I bet Pinterest did their homework and realized the $3 billion they're going to pay to AWS is cheaper than building, staffing, and running their own infrastructure. That's the conclusion we came to - after building our own data centers.
I did exactly this project for a high traffic site about 10 years ago. We got datacenter rooms wholesale that included power and cooling. Did everything else ourselves and cut our burn rate in half. However, we were totally open source so no licensing costs.
Even after leasing all the hardware, hiring the staff, getting the transit brought in, we still cut our compute costs in half.
Even so, we ended up keeping cloudfront+S3 for serving static content because we couldn't beat the price they offered.
> However, we were totally open source so no licensing costs.
This is not a healthy opinion to have, not for the open source community and neither for the company using open source projects. All the bugs and missing features of the open source project you use are also a part of your technical debt now, you can either donate to project to keep it healthy or hire FTE working on contributing to these projects, which is just indirect form of payment and should absolutely be a part of your budget
You seem to be jumping to a lot of conclusions simply because we chose freely available open source.
To have a business depend on software you either need a support service or you support it yourself. We chose the latter and made sure the team had the skills to do so. Of course we contributed our bug fixes back and even released our own tools open source.
Good but those two things aren't nearly enough. Then people wonder how we end up with HeartBleed bug, Elastic & Grafana moving to stricter licenses, Mozilla dying in favor of FAANG provided software.
I'm surprised that static content is where AWS is competitive for you, considering the high egress pricing ($20/TB is the lowest advertised price for cloudfront).
It is highly negotiable when you get to large volumes. We got a term where CloudFront fetches from EC2 were free for objects <101K (or something like that) and free for EC2 from S3. We were able to take great advantage of that with the way we built our app.
It was a long while ago so I may be off on the exact details.
> Oh really? Ever try building, staffing, powering, cooling and equipping a data center? And then build a DR? Let me tell you, my company has foolishly done that just in the past ten years and WE are screaming uncle! We're now transitioning to AWS because it's much, much, much cheaper!
Since most of my career postdates and predates renting capacity on someone else's computer, yes, have done that many times.
Would be great to publish a case study on your experience, would love to understand the numbers.
While for some rare scenarios AWS can be cheaper, for just about any use case it's going to be much more expensive than running your own.
>Cloud is known to not be the cheapest option at large scale, so it is a bit curious they decided not to build their own infra.
The "cloud" provided by AWS is cheaper -- if you use the higher-level AWS portfolio of services . E.g. DynamoDB, Redshift, Kinesis, etc.
If the mental model of "cloud" is just "bare metal servers", aka EC2, then AWS is more expensive than self-owned datacenters.
So yes, if you're only looking at cpu and egress costs, AWS as a "dumb data center" is a terrible deal. But that's not how Pinterest uses AWS. They use AWS high-level services:
(Yes, those are AWS marketing articles but writing those case studies require cooperation and approval by the customer (Pinterest in this case.).)
So far, Pinterest concluded it would not be cheaper to build their own machine learning algorithms to process images. Paying for AWS Rekognition is part of the ~$400m/year cost.
Interesting points. Is AWS Rekognition just doing OCR as a service? I guess there isn't enough value in that to build a competitive solution. So good point. For more complex perceptual models, I'd be really surprised if the model wasn't considered proprietary/value-added.
These types of agreement are relatively common in some manufacturing environments, and are often called "take or pay." In essence you're reserving capacity.
What's interesting to me about this one is that I fundamentally believe that both:
- Cloud computing will be increasingly commoditized, with lower cost providers emerging in the next 3-5 years
- More companies of scale will go back to building their own infrastructure as a cost savings project.
For Amazon, I think the incentive is as much about locking in revenue as it is locking in "demand" - helps with their demand planning of how much more to build out.
What does cloud computing mean to you? It seems like you are thinking of traditional “virtual machines in the cloud”.
When I think of AWS for new development I mostly think of serverless technologies like Lambda, DynamoDB and even serverless RDS. It’s going to be difficult for any company to do it for themselves cheaper.
To me, AWS is about the exact resources needed for a task blinking in and out of existence for the exact number of milliseconds they’re needed. I don’t know how anyone will compete with a cheaper on-prem solution.
> for a task blinking in and out of existence for the exact number of milliseconds they’re needed
Right, in cases where the capacity needed is so low that "blinking out of existence" is a frequent scenario, it is indeed cheaper to rent only those CPU-milliseconds that are needed.
But as usage grows, it will fairly quickly exceed the cost of just having dedicated servers.
> Cloud computing will be increasingly commoditized, with lower cost providers emerging in the next 3-5 years
There are plenty of alternatives ranging from just IaaS ( you rent VMs) to more advanced PaaS/SaaS type "clouds" (e.g. Scaleway have managed databases, Kubernetes, FaaS, CaaS). There are some fairly decent ( in terms of coverage, cost, features) providers out there, outside of the big three.
However they are in an entirely different market similarly how Tata Motors, PSA Group and Lamborghini aren't competitors. Yes, both are used to move people around, but the scale and features aren't even close. There are so many managed services on AWS it's impossible to list them all coherently. Whatever you could need, it's probably there. Digital Ocean or Scaleway or similar are perfectly fine when you're smaller, more technically advanced, with more time available, higher failure tolerance. They can make sense to use, but aren't for every use case.
For example in 2017[0] Snapchat committed to spend $2B with Google Cloud in the next 5 years plus $1B with AWS over the next 5 years. I guess they wanted to rely on 2 public cloud providers not exclusively on one.
Btw it's crazy how these big cloud providers and especially AWS are making billion dollar deals left and right. It's good that we have multiple of them so they can compete meaning lower prices for buyers and increased quality for end users.
This is just a normal business-to-business pricing agreement contract for a long-term commitment. In this case, it's ~$400 million per year for 8 years.
Also, I guess lawyers copy&paste each other because the Snapchat text about its GCP contract looks very similar with the word "required":
>excerpt from Snapchat S-1: On January 30, 2017, we entered into the Google Cloud Platform License Agreement. Under the agreement, we were granted access and use certain cloud services. The agreement has an initial term of five years and we are required to purchase at least $400.0 million of cloud services in each year of the agreement, though for each of the first four years, up to 15% of this amount may be moved to a subsequent year. If we fail to meet the minimum purchase commitment during any year, we are required to pay the difference. Our agreement with Google permits us to use other third-party service providers for a portion of our cloud services.
This is standard practice. Most SaaS IT departments negotiate big rebates in exchange for long term commitments, and those who don't are leaving money on the table. I've seen datacenter deals over 10 years that looked far worse.
I'd really like to know where their costs are coming from, $400m/yr seems absolutely wild. I work for a large org who has 90-95% of stuff in cloud infra and we don't spend even close to this.
or you could pay $1m to someone like Corey to revisit the architecture a little and shave $100-200m off that bill by not doing silly things? https://www.lastweekinaws.com/about/
In short: it does not matter for a company that does not even have a profit and probably won‘t reach break-even soon. It‘s a waste of time to cut cloud-computing cost at that stage of the company.
Usually a large company has people as smart as Corey to avoid mistakes. AWS also tends to assign solution architects to contracts this large to help customers avoid waste. Sure, it loses them some spend, but the customer obsession earns them more trust.
Capacity planning challenges can't be ignored and it's interesting that Pinterest share their forecasts with AWS significantly in advance. Makes me wonder if AWS will offer more incentives to older/larger enterprises that are used to datacenter-era capacity planning, I suppose RIs and Savings Plans are a manifestation of that already.
For big orgs it’s quite common to have NDA’s with AWS and embed their staff for support and decision making. You get expertise and insights into the upcoming roadmap as well. It would be no surprise to know that Pinterest shared these details with AWS.
Is it still going to be ~$3B by 2029 if they use (a lot) more?
I mean I can't see it happening; I can't see Pinterest growing much larger than it currently is, and as hard- and software improves, AWS will lower prices and Pinterest will be able to use resources more efficiently.
I used to sell both cloud products and server/storage infrastructure.
There are too many factors, often legal requirements and the umbrella company policies not being flexible, but , if you need 20 or more servers and redundancy and are legally permitted to go cloud, do it.
You can run non business critical parts of the it on cloud and see if it makes sense to the cfo.
Buying physical servers and the premises is one thing, but there's so much more to it, it's far from plug and play.
You need hypervisor, licences, one or two admins, it adds up.
I used to use Pinterest for recipes a few years back, but their recommendations sucked (or they didn't have enough content). I would either keep getting the same suggestions again and again or recommendations for meat dishes though I'm vegetarian. Now I just follow a few well-known chefs on youtube instead.
Great. Drug dealing may bring in money to but what kind of value do they deliver? Except staling content all over the internet, spamming google picture search and then offering click bait?
But you couldn't, could you? 400 million is more than Hetzner total revenue.
I think 400 million is staggering amount of money to be used for computing, but if you are using that amount of money there are only handful amount of players that can absorb that kind of money and provide you services.
And deal with the physical stuff ( racking, cabling, cooling, electricity, DCs) and networking stuff ( high quality redundant networking routers, switches, firewalls, the engineers that can use them) ? It's quite possible it would cost more than $400 million / year to do all of that properly at a global scale, and a big part would be paid upfront.
Another useful comment from the thread:
> Makes sense for both parties.
> For AWS, it allows them to make hardware investments early on
> For Pinterest - it allows them to keep costs manageable and get a significant discount on MSRP