The magic cloud. Good thing it is made of fairy dust rather than commodity servers and network equipment or Dell and Cisco might be around forever providing those things.
That's basically the point of the article: These companies that actually had to get up to that kind of massive scale quickly discovered that the way hawked by these big expensive "enterprise" companies didn't work. It was expensive and insufficient. So they've developed their own custom stuff (often built on open source; sometimes open sourced back out) that works a lot better while also being massively cheaper.
AWS isn't just a bunch of Dell servers hooked up to EMC SANs and Cisco switches. The only people who still use that stuff are people who a) have a ton of money and a desire to spend it freely, plus b) don't have really serious demands, which would mandate something better.
There's still going to be a large number of small companies that will be needing servers. There might be so many hardware providers as is, but you can't blame Dell for attempting to be the last man standing.
Similarly with Cisco, you can run the internet exchange of a small country on a single stock Cisco switch, and many do. Cisco isn't going anywhere in the near future. Amazon, Google and Facebook might not have a need for them, but thousand of other companies do, even large datacenters.
Yes. What does internet traffic have to do with it?
There are millions of SMBs and tens of thousands of large corporations around the world that own server hardware. A tiny percent of that is the major internet companies.
The designs for hardware are changing but the providers are still going to be the usual companies that are good at this and have all the manufacturing in place. For example a lot of banks have started to switch to Open-Compute designs [1] but they will still buy these from Dell or HP.
So the thing is, yes, that's true. Lots of companies can get by with Cisco/Dell/EMC stuff and don't need the ultra-high-end datacenters of Amazon/Google.
And if Cisco/Dell/EMC stuff was bargain-priced and the ultra-high-end stuff was expensive, well, there you go.
But the problem is, if you want your stuff hosted on Amazon-class hardware/network topologies, it's really cheap to do so with AWS (or Azure, or whatever). And the Cisco/Dell/EMC stuff is painfully, ruinously expensive.
If the high-end thing is also the cheap thing, what's the rationale for buying the low-end, expensive thing?
Not denying the cost of AWS (I priced out S3 the other day and it's amazingly cheap) however there legitimate business cases where companies won't put data in the cloud (and in fact I've found more clients are asking about those in light of the press around Snowden/NSA/GCHQ).
One interesting observation about commodity hardware is it increases in capability faster than the rate of many business demands. In the short term if there's a lack of infrastructure you can grow shipping at a very fast rate, until you catch up...
In the long run the "walking dead" from the article are doomed when due to technological growth an entire data center can be converted to a single rack in a closet without any expensive monthly upstream bandwidth charges and nickel and dimeing from a cloud. Those suppliers have expensive corporate infrastructure that depends on shipping $50M data centers in 2005 not a single $100K rack in 2015, or 2025 or 2035 or whatever, to do the same task. Maybe a great sales campaign can convince them to double their usage, that still leaves them $49.8M in the hole, non-inflation adjusted (ouch).
Computational use for non-computer businesses is kinda like electricity in the VERY long term. If you look at the early years of the growth of electricity vs GDP or whatever, you can rapidly get strange ideas like the average office in 2015 based on projections from 1915 should be radiating heat like the surface of the sun. However, here we are, there is a natural leveling off of how much electricity or municipal water or food a business can usefully consume per unit of GDP. Assuming GDP only goes up, inflation adjusted (LOL). When companies bounce off those limits, life will get very exciting for "eternal Moores Law growth forever" oriented suppliers for those businesses. Based on facebook use stat growth rates in 2007 we should all be spending 78626 hours per day on facebook... didn't happen... building a supplier company based on that assumed growth trend will lead to tears. Moores law says the transistor will shrink not that people will buy it.
Very large hosting/cloud providers will eventually build their own hardware. Even if Dell is involved, it's going to be in a supporting role and won't yield the kind of profits X-many server sales will. And besides, support contracts will also be of lesser value. This and Open Compute is not good for Dell-type of shops. Compound that with Microsoft building PCs in a shrinking PC market, the problems are numerous. Dell can't make a Windows PC better than Microsoft.[1]
All this also because these companies have never really tried to attract the best talent by paying above market compensation. It's fine.
[1] The kind of detachable GPU configuration we saw in Surface Book is harder to pull off outside of MS. Doable, but MS will have the advantage of getting inputs directly from the DX12 team.
http://www.wired.com/2014/11/facebooks-new-data-center-bad-n...
That's basically the point of the article: These companies that actually had to get up to that kind of massive scale quickly discovered that the way hawked by these big expensive "enterprise" companies didn't work. It was expensive and insufficient. So they've developed their own custom stuff (often built on open source; sometimes open sourced back out) that works a lot better while also being massively cheaper.
AWS isn't just a bunch of Dell servers hooked up to EMC SANs and Cisco switches. The only people who still use that stuff are people who a) have a ton of money and a desire to spend it freely, plus b) don't have really serious demands, which would mandate something better.
That's not a good place to be.