Hacker News new | past | comments | ask | show | jobs | submit login
Amazon Web Services CEO: We’re a $30b revenue business in the ‘early stages’ (cnbc.com)
114 points by bovermyer 24 days ago | hide | past | web | favorite | 83 comments

I really wish they reduce their public bandwidth cost to half. It will make such a major impact on the internet as a whole that we will see lots of companies grow faster and newer video heavy apps launching.

Here's a few ideas to start a video heavy app on a budget:

You could store files in a Backblaze B2 bucket and serve them via Cloudflare with zero bandwidth fees. Or try DigitalOcean's Spaces offering with their built-in CDN at $0.01/GB for bandwidth.

Hardware video encoding could be done using cheap Hetzner servers that include GTX 1080 GPU's (lookup NVENC) or try Intel Quick Sync encoding (perhaps using OVH's overclocked i7-7700K CPU's).

>You could store files in a Backblaze B2 bucket and serve them via Cloudflare with zero bandwidth fees.

From Cloudflare's TOS [0]:

>2.8 Limitation on Non-HTML Caching

>The Service is offered primarily as a platform to cache and serve web pages and websites. Unless explicitly included as a part of a Paid Service purchased by you, you agree to use the Service solely for the purpose of serving web pages as viewed through a web browser or other application and the Hypertext Markup Language (HTML) protocol or other equivalent technology. Use of the Service for the storage or caching of video (unless purchased separately as a Paid Service) or a disproportionate percentage of pictures, audio files, or other non-HTML content, is prohibited.

Of course Cloudflare offers commercial plans that feature video CDN, but it will cost you and it will cost you similarly to other video CDN providers.

[0] https://www.cloudflare.com/terms/

Yeah, I meant zero egress fees from B2. It's a shame that Cloudflare isn't more transparent about their bandwidth limits but my guess is that traffic would still be cheaper than from Amazon. Maybe someone from their team could weigh in ..

All this means is that your main website, where these files are embedded/served from, also needs to be on Cloudflare (and proxied).

I'm not sure that would follow

> disproportionate percentage of pictures, audio files, or other non-HTML content, is prohibited.

If your video sharing app has any decent usage your ratios are gonna get in their red zone pretty quick...

I think you have a very different idea of what constitutes "mainly HTML."

Base64 encode the video and embed it in the HTML?

E: typo

Using Backblaze which is only in US and Hetzner which is only in EU might not be the best idea, you might have issues with networking between locations.

Good point. The last I heard, they were still working on bringing up an EU location (I believe it was Amsterdam).

You can have multiple cheap origins behind your CDN, and use DNS to load balance/failover between them.

Is YouTube still operating at a loss? If so, then don't expect video distribution to be cheap.

They probably weren't ever operating at a loss. There were some stories speculating about YouTube's costs a few years back, but they were based on market prices for hosting which would be much, much higher than Google's internal costs.

Depends on how you want to measure revenue. YT itself is a massive cash hole. It's the advertising that carries it. YT Red (and other types of paid service) are bumps in the economic redline.

"Google has never revealed YouTube's revenue" but you can quantify it (roughly) from known values and costs.

Ultimately profitability of a product within Alphabet depends on how you want to categorize technology within Alphabet. DFP and YT, at the very least, are separate products.

Why do you think they have the margins to afford this?

What do you mean? Their bandwidth charges currently would be considered egregious for even a medium sized company. Someone the size of Amazon has obviously negotiated far, far lower peering rates than that. The only reason their outbound prices are so high is to keep people from getting out in a cost effective manner. They want to make customers think long and hard before leaving.

> The only reason their outbound prices are so high is to keep people from getting out in a cost effective manner.

This doesn't make any sense. Getting out is a one time operation, vs paying a higher recurring fee.

Lots of data must be kept for e.g. 7 years yet is never used. Maybe that should never have been in AWS in the first place, but once it's there this is a factor.

Amazon will happily ship you a disk : https://aws.amazon.com/snowball/disk/details/

That's not free either.

You can get an 80 TB Snowball to use for 10 days for $250, and pay a minor shipping charge to get it to and from the AWS datacenter. That's not free but its cheap enough that it should be a tiny decimal point percentage of the value of the 80 TB of business data you can fit on the Snowball.

And yes Snowball goes both ways: you can get data into AWS or out of AWS.

And how many businesses have 80TB of valuable data that is valuable when it's offline for potentially weeks while it's moved onto the snowball, shipped, taken off of the snowball, and then converted into a usable format onsite?

Ignoring the fact you lose all metadata transferring via snowball and any ACLs depending on the type of data and where you pulled it from. You can't pull from anything BUT S3 either. So that database? Sorry, online. The Glacier archive? Going to have to rehydrate it first.

The $250 is a rounding error for sure, but not on the cost of the data, on the cost of restoring the data. Even S3 itself you're ignoring the fact that you've got $2,500 in transfer fees alone on that 80TB.

Thanks, that's great info. Hopefully the snowball is considered to be in the same "availability zone"?

In theory, but this would not be true in practice. The cost is meant to make Redshift the default data warehouse versus that “cool new service” over there as an example. Each one of these types of decisions creates more AWS team expertise, and a harder decision to migrate to another service.

Getting out is not necessarily a one-time operation. It can mean switching some services to a competing offer, even just for an evaluation period. So for example, say, you're using a bunch of hosted AWS services and Google's BigQuery.

Their gross margin is over 20% according to their public reports and their networking margin is about 90% according to the AWS network engineer I talked to last year so I think they could afford it.

That doesn't seem high enough to justify halving prices unless it significantly increases revenue, which is pretty unlikely at this point.

But OP's hypothesis is "It will make such a major impact on the internet as a whole that we will see lots of companies grow faster and newer video heavy apps launching." Disagree with that if you like, but if that's true then the revenue will take care of itself.

Losing money on every exabyte and making it up in volume?

90% margins -> cut prices in half -> lose money?!?

Well, yes, if you're paying $1.00 per unit of bandwidth and selling it for $1.90, that's what will happen. What else would you expect?

I suppose that could be interpreted as ~50% margins, rather than ~90%. Not sure how that calculation is actually supposed to be performed.

20% Doesn't help to just consider network margins.

That's all we're talking about; thread parent said "reduce their public bandwidth cost to half". Obviously, you can argue that this price change idea is not strategically in AWS interest (AWS management agrees right now), but the trivial nitpicking ITT is misguided.

Unlike the old web hosting platfors, Cloud computing is becoming a big monopoly. Which is really bad for everyone.

We are in early stages and the monopoly will get worse.

How are we even close to a monopoly? I could maybe see a oligopoly, but even though amazon is big, its not the whole market.

It will likely never be a true monopoly. But an oligopoly with two or three players? That's where we are and where we'll remain.

Its the only serious use case for enterprise.

What, you want to use azure?

I worked at Amazon for over 5 years and I currently work for a pretty big company (you've heard of them) that's all-in on Azure. It's got enough going for it that I wouldn't switch even if I were high enough in the hierarchy to make that decision.

GCP, Azure, Oracle, Alibaba ... just to name a few.

It certainly has the potential to be a huge market. What barriers prevent others from entering?

Software? Does Amazon have have a monopoly on strong technical workers or patents? Not sure about the ladder, but certainly not the former.

Capital? Cloud hosting is probably capital intensive but there is plenty of interested parties with capital.

Legislation? Probably a lot lower barrier to entry than other industries, although I'm not sure.

It could just be that they are executing better than the others.

I think you severely underestimate what it means to run a public cloud business. "Probably capital intensive" betrays the naiveté of this comment.

Even if you have 100 billion dollars in the bank, you're years behind. Go buy land, build datacenters (5 year lead time), sign power contracts with utility providers, figure out where to buy all your dram and disk and networking and shit and oh develop your cloud services too. Now keep them running.

Underrated comment.

To add on to your point: Amazon told Jassy in 2003/2004 to take 55 engineers and build out some cloud services they could sell to customers. ~6 years later, Microsoft decides to enter the market.

That is an epic first-to-market timeline.

Even apart from the large infrastructure cost for a geographically distributed and reliable network, a lot of the value of AWS/GC/Azure comes from the many built in services that completely change the ops workload.

Building out those services is a monumental task requiring loads of experts in their respective fields (databases, storage, networking, etc). I see this as at least as much of a hurdle as the capital expenditure.

Oracle, with OCI, is working to crack the market. It has the kind of capital to be able to do it, but it's just insanely expensive. We're running out of companies that can possibly fund such work.

Not only do you have to have the physical infrastructure to launch a region (expensive in and of itself), you pretty much have to have the infrastructure to launch at the minimum two regions geographically separated in some form. Plus with data locality laws increasingly enter the mix (something that AWS never had to worry about until they were already a success), you increasingly have to build regions everywhere or customers can't use you. Frankfurt, for example, is a virtual given.

That's before you even start to consider all the engineering work involved writing the code to run the cloud services.

Just launch a bunch of super-tiny, 1-AD regions so that you can have it on the map, and then rush to do region expansion within months of region launch. ;-)

You would probably start by co-locating in multiple data centers, if you rent large amounts of space you can get quite decent deals. You'd have to do this with the best DCs you could find, however.

Once you had enough demand to merit your own data centers, then would go ahead and build them. Thats what most VPSs do.

Agreed. Ask RAX how hard it is to compete in commodity cloud infrastructure.

I’m a RAX public cloud customer with a 6 figure bill every year and they are actively trying to get me to switch to AWS.

We could be. Edge computing though could be an interesting partial disruptor, or it could further entrench the leaders as AWS already has an edge computing package I believe.

You could run a small business with an RPi and residential internet. That, and turn key solutions for websites and stores are probably bigger competition to colos and vps services.

I love AWS.

Some things that I think can still happen on AWS, if not already happening:

VPC Images!

Basically a VPC launch button that has a series of terraform/*form rules that deploy a SaaS-in-a-box:

Meaning, VPC images of infra to support a business case.

Want to start up an infra that can receive video uploads and display them on a portal? Click here.


The first item in my wish list is a better and more intuitive Cloudformation.

So much busy work they’ve created around their platform.

This is the AWS Cloud Development Kit. It is still under very active development, so keep that in mind as a caveat, but it provides a much higher level, much more simple and intuitive way to define cloud infrastructure as code in your favorite languages: https://github.com/awslabs/aws-cdk

For context I've been able to take thousands of lines of CloudFormation YAML and replace it with around 500 lines of much clearer and easier to read JavaScript. And that 500 lines automatically builds five Lambda functions, builds, pushes, and launches two Docker images that run in AWS Fargate, configures an API Gateway and a CloudFront distribution.

CDK is really a game changer in terms of allowing developers to take an idea and turn it into infrastructure as code extremely quickly and properly (CDK automatically generates many best practices like minimal IAM roles, minimal security group settings, etc)

Damn, the cdk is really impressive. This example shows setting up a fargate cluster to be an order of magnitude less code than the raw cloudformation config:


Unfortunately it does still appear to be in preview.

I'd not call CDK an intuitive replacement for Cloudformation. It is like, hey guys! we don't have a better solution to give them a common tool/language to provision their stack. So lets make it an object-oriented abstraction, at least give them auto-complete.

CDK does a lot more than just be an object oriented abstraction. Often times one of the constructs in CDK actually makes many different underlying AWS resources.

In CloudFormation its always a 1:1 mapping: one CloudFormation resource typed out as YAML, one resource created on your AWS account. In CDK you can make something like "LoadBalancedFargateService" and just plug in the path to your local Dockerfile, and it builds and uploads the image to ECR, makes a Fargate service, makes a load balancer, connects the service to the load balancer, and returns the URL to you.

The other power of CDK is it sets things up right out of the box. Rather than having to explicitly created all your autoscaling rules it can create sensible default ones for you automatically. It also automatically creates the right IAM rules, and security group rules that you would otherwise have to define manually.

As a whole I'd say comparing CloudFormation and CDK is like comparing assembly language and C++. You could write your code as a bunch of hand rolled ASM for the most control, but realistically that's not feasible for large projects as its error prone, slow, and quickly gets unwieldly. CDK is like higher level C++ that synthesizes back down into the lower level assembly language that creates all those cloud resources

Check out aws-cdk ?

Have you tried codestar? It's the closest thing to what you're describing that they offer as far as I know. It would definitely be great if they were specific to various application use cases instead of the more general purpose technology choices.

You need the AWS Marketplace my dude!

Strongly Agree with Andy in that interview.

In U.S , one way or another everything is running in the cloud.

For Europe ? It’s more complicated , most large businesses still have their own datacenters or run using dedicated machine with European hosting (OVH).

I’ve worked for many differents banks and they barely use the cloud compared to startups like N26 or Revolut.

Same thing for Industries or Health Services they largely run « On Premise » compared to U.S equivalent.

There is still room for growth from Europe if they manage to convince leaders of large corporations that the cloud will not take the their sovereignty away from them .

I've worked as a consultant in Germany and I was shocked by how far behind your average company is regarding cloud adoption. Modern tools and processes are also usually a pipe dream. So I agree, the industry as a whole has way more room to grow than you would suspect.

Europe is the next big cloud customer since they're always behind US for tech adoption.

Always is a strong word.

France had usable networked computers everywhere long before the us did (it was called minitel)

Over on the UK my local Telco (not BT) was doing video on demand over copper ~2000, they where also one of the first to switch a fully packet switched internal backbone (so successfully other telcos came to look at how they did it).

We had chip and pin a decade before the US.

Our median internet connection is better.

Is the US the world leader sure, always ahead..not so much.

Internet in US is just terrible for 90% of the population. Even if you are in the Bay Area or Manhattan you are still getting bad speed and you are stuck with a provider like Spectrum or Comcast and you pay 70$ for 150-200Mbps. I just don’t see this changing anytime soon :(

I pay $70 a month for gigabit AT&T fiber in a southern city, 10 miles from the city center.

in nyc, i pay $100 for gigabit

Sweden.10Gbps for 30$ / month. It's kind of a long-running launch campaign, but still. https://www.bahnhof.se/villafiber/

Whats the typical bandwidth quota? Surely they wont let u saturate a 10gig pipe ..

I've got 400Mbs/35Mbps with no cap (literally unlimited) which would cap out at a round 4TB a day if I ever had a reason and could saturate it.

10Gbps domestically is insane though, that's pushing 100TB a day!.

Most of EU plans I have experience with are not capped; but 10gbps seems insanely high.

Belgium, 39€ for internet + digital TV + phone line 50 Mbps, but it is enough)

We are more on the expensive of Europe. My subscription is pretty cheap though.

200Mbps in the entire country would be around 55€ ( internet only)

Europe has 4 of top 5 of the fastest broadband countries in the works.

Romania (#5 in speed), goes as low as 7€/month.

So in Romania it's basically the same percentage of income as in the US.

Where does that reference "stop". Every utility still has a base cost.

Also, I'm pretty sure that the richest in Romania has a much higher income than the poorest in America. But they still pay the same price.

Ps. What if you reference it with social health care in Romania vs US. Not everybody works, you know.

But it does and it will take away an amount of sovereignty. "Excuse me AWS, Can you send me every hard drive used in the storage of Dataset X? Not copies, the original drives. I need to destroy all the drives per regulations." GDPR muddies the waters too. If AWS is the cause of a fuck up and a bucket is accessible on the internet, good luck proving your company shouldn't be saddled with the GDPR fine. AWS will have the technical means, monetary incentive and reputational impunity to claim it was your own fault.

Then there is the fact that a decade on the numbers are in; Cloud isn't any cheaper at the scale of most established businesses AWS wants to court. Often it's more expensive.

None of this will actually slow adoption. It probably should, but it won't.

Makes you wonder if one day AWS will overtake its parent.

AWS carried Amazon out of mediocrity in 2008. It accounts for something like 70% of operating revenue. What does 'overtake' mean?

I'm not sure how Amazon was mediocre by any measure in 2008 but your second statement is incorrect - AWS only accounts for ~10% of Amazon's revenue. https://venturebeat.com/2019/01/31/amazon-earnings-q4-2018/

> how Amazon was mediocre by any measure in 2008

History is there for analysis. The stock falling to April2007 levels in Sept2008 showed serious volatility and hadn't yet declared profits that carried it above Sun and Toy R Us yet. It was somewhere in the top 300 profitable companies. It was still carrying the HUGE debt load that they have been carrying over to enable yet another tax free year in 2018, despite another record year. Quite a bet to make that they would be successful where AWS was just exiting beta. It was completely orthogonal to their barely-on-track online retail business model. It was a meh company and the stock price reflected that.

5B out of total 60B for last Q = 8%, not 70%.

It is time for Kubernetes and other solutions to build a decentralized federated cloud offering.

Maybe they can get half of their services to work properly then?

If AWS gets Serverless right....and they are so far with Layers and Runtime API, they are in for a new golden era.

I know people have been on the edge about serverless in the past but it's now reached a point where I'm planning to actively employ them. 2019, will be the Year of Serverless methinks.

It will be great when serverless aurora's data api can run sql queries at 5ms-ish latency instead of the current 200ms. Then lambda functions will be able to use sql in a performant way. For applications that need relational databases, fargate seems to still be the best option.

Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact