[Bedon292]

The majority of Amazon profit (not revenue) is from AWS. And AWS started because they already had the servers for holiday surge capacity. Was extremely lucrative, so they invested more in it. Maybe in the first years there was a loss there. They didn't start separating the numbers until 2012, and in 2015 their margins were ~25% on AWS [1]. So sure they are spending billions, but they are not spending more than they make on AWS.

Perhaps Alexa / Echo are not directly profitable as a device, but they make money from people buying stuff on Amazon.com with it, and gain that brand loyalty from it. Which has definite value. But there are estimates it costs somewhere around ~$30 to produce one [2]. So at ~$60, its very profitable. And when they sell them for $30 for a special or something, its still in the break even area. A couple sales and you are in the black.

[1] https://www.wired.com/2015/10/get-used-to-amazon-being-a-pro... [2] https://voicebot.ai/2018/02/15/apple-homepod-costs-216-manuf...

[jsmith99]

I think they meant that AWS is hugely profitable according to accounting profit, but is cash flow negative due to the large amount of capital spend. When you buy assets, that uses cash but doesn't reduce accounting profit.

[njwi332]

>AWS started because they already had the servers for holiday surge capacity.

That's a myth. Source - worked at AWS for four years, but also common sense. If you rent out all your surge capacity to customers then what happens when the holiday season comes? You kick all your customers off and shut down their business for a while? Not a business model for happy customers and not what happened in real life.

[einpoklum]

> And AWS started because they already had the servers for holiday surge capacity.

That doesn't seem very likely. There's a limit to how much server work you need to handle one online shopping website.

[Bedon292]

That is what I had read or heard at some point. The majority of their business was in the holiday season, and needed several multiples of servers to handle it. So in say May, 80% of the machines were sitting basically idle. So why not rent them out.

Unfortunately I appear to have been mislead by a myth: https://www.networkworld.com/article/2891297/the-myth-about-...

So, thanks for calling that out. Would have gone on believing it otherwise.

[c22]

What is the limit? It seems reasonable to me that it might change based on number of users, frequency of transactions, amount of user engagement, etc. After all, non-shopping websites have to scale with these metrics.