Table of contents of the article:
This is an article that probably will not appeal to many colleagues who will feel challenged by this post, however it is not a sterile criticism but a pure observation that this week more than ever has allowed us to understand the difficulties that some entrepreneurial choices extremely courageous and noteworthy undertaken a decade ago are proving to be problematic nowadays.
The specific case examined
We contact a customer who asks for a projectallocation of 6 machines with 32 cores and 64 CPUs each. The choice falls on the AMD EPYC 7502P 32core model of the well-known AMD house, now a worthy competitor of the more famous Intel.
We will not dwell too much on diatribe between AMD and Intel on the server side on which we have already written a post, but suffice it to dwell on the price of this one e single CPU which is about 3000 €.
And let's just talk about the CPU, the processor in short. Adding RAM, motherboard, case, and nVME disks in RAID 1 will reach over 4000 € per single machine.
Now focus on the fact that the aforementioned customer asked you for six, not a car but six. We should calculate a cost of 4000 euros for 6 machines to be charged, albeit at least 20% on the live cost incurred.
In particular we should spend 24 thousand euros for the purchase of the machine and invoice the customer 29 thousand euros including our earnings.
It might also seem like a bargain, but the customer needs those machines for 2 months and not forever, so if we wanted to distribute the cost for the necessary two months, the customer would still have to spend 15 thousand euros a month or decide to rent them to 2000 euros per month for a total of 4000 euros and hope to be able to rent them to someone else immediately afterwards.
In short, the problem is clear, some machines are very expensive and you may need to have them immediately and for a short time. Which datacenter would immediately buy 6 machines for a cost of 24 thousand euros to amortize 4000 immediately and leave to chance the possibility of renting them shortly thereafter?
Someone maybe yes, many smaller datacenters obviously not. And here is the problem that is revealed very eloquently, the costs for the management of a Hosting Provider today are beyond all logic and well beyond the entrepreneurial risk that every entrepreneur with capital is willing to run.
Liquid business, the basic concepts.
To better understand the concept behind the crisis of hosting providers with proprietary datacenters, it is necessary to introduce what a liquid business is.
A liquid business has the goal of create high-income businesses with minimal costs. The first thing a liquid entrepreneur must do is, as we have seen, eliminate the fixed costs it incurs every month. The liquid company will only face variable costs, that is, those money spent on getting more money back.
To create a successful liquid business you need to follow these steps:
- Minimize, or eliminate altogether, fixed costs;
- Organize a team of professionals located in different points, even geographically distant, but in a way that their needs and their times are compatible with the needs of the company;
- Make use of collaborators o freelance workers;
- Do without the company headquarters and its related costs;
- Focus the entire liquid enterprise on a model that provides exclusively variable costs, or costs that change proportionally to the turnover. Variable costs are all those costs that vary according to the quantity you produce. This means that if zero is produced, the variable costs will be zero. The most typical variable costs are the costs for the purchase of raw materials, semi-finished products, finished products. Then there are other variable costs such as utilities, commercial costs and some administration costs.
It is clear that having a proprietary Datacenter which necessarily has to deal with fixed costs, datacenter, security, fire prevention systems, redundant connectivity, redundant electrical systems, surveillance, routing as well as a vast fleet of machines and models ready for delivery in less than 24 hours is the exact opposite of a liquid business and consequently all the problems of management and cost containment that cannot be managed.
Maybe in our specific case examined above, we could also find the small datacenter that has 6 machines of that type available and free, but what if instead of 6 they asked for 20? We are sure that the small datacenter that has a turnover of 1 million a year with a profit of 100 euros he would have been able to put his hand to the cash register and spend 80 thousand euros on the nail (about half of the profit) knowing that it would amortize the remaining 60 thousand MAYBE in the following months?
The answer is easy and you don't even need to write it down.
The solution to an illiquid business is an economy of scale.
If you have important fixed costs between datacenter, personnel, servers, and the whole list of costs that we have listed above, the only optimal solution you have is to have economies of scale. If a datacenter that costs in the management, a living expense of 30 thousand euros per month, plus a fleet of 500 servers (cost an average of 1 million euros assuming they cost 2000 euros each), we know perfectly well that in order to go positively we should manage these machines for a 2 year life cycle.
The account of the servant says that the management of the datacenter costs 720 thousand euros in two years, and that the machines have cost me 1 million euros for a total of 1 million and 720 thousand euros.
How much should a single server's monthly cost to get even in two years?
143 Euros. To go there exactly.
Don't we want to add a profit of 25%? Then the car will have to cost 178 Euros.
How is it possible that that machine that costs 178 euros (the iron alone means without any system assistance above), we manage to sell it for 50 €?
Is the problem clearer now? Those machines that are sold for € 178 per month or:
- 135 € / month if you decide to amortize them over 3 years
- 124 € / month if you decide to amortize them over 4 years
- 113 € / month if you decide to amortize them over 5 years
- then they are old and to be thrown away
For example, we sell them for € 50 / month.
So let's simulate an economy of scale, in which the same datacenter with the exact same management costs, is able to allocate not 500 but 5000 servers.
We will have a monthly management cost of 30 thousand euros and a cost of the servers of 1500 euros (because we bought 5000 and we negotiated with the European importer, saving 25% of costs), spending not 10 million but only 7,5 million .
Now with these new parameters, let's see what is the best price that can be made to the end customer considering a profit margin of 25%.
How much should a single server's monthly cost to get even in two years?
68,5 Euros. To go there exactly.
Don't we want to add a 25% profit? Then the car will have to cost € 83,57
How is it possible that that machine that costs 83,57 euros (the iron alone means without any system assistance above), we manage to sell it for 50 €?
Is the problem clearer now? Those machines that are sold for € 83,57 per month or:
- 58 € / month if you decide to amortize them over 3 years
- 45 € / month if you decide to amortize them over 4 years
- 37 € / month if you decide to amortize them over 5 years
- then they are old and to be thrown away
In short, you understood with this extremely generous example (in which management costs are fixed for purely illustrative purposes) but where managing 5000 machines requires a real team of at least 10 systems engineers, that the road to success lies in the economy of scale, according to more basic rules of operations research.
They were once pioneers.
It is inevitable that in order to host someone at the time around 1996 decided to set up warehouses that would over time become datacenters. Recognition and glory should be given to those who as pioneers decided to start with the first server farms to set up a few hundred machines and then grow, grow and grow until reaching their maximum capacity.
We cannot blame the choices of those who decided to invest profitably in a business that at the time had to be managed and carried out in that way, the problem arises in the middle age when it was already too late to build datacenters for the vast public that they would not have been sufficiently scalable to guarantee the best price conditions for the average customer.
It is a bit like the history of video libraries and video cassettes, in short, those who started in the 80s and pushed until the 2000s certainly had an extremely flourishing and entrepreneurially profitable period, no one would have imagined that shortly thereafter with the 'by using the internet, video stores would disappear.
For those who started in the 80s, 2000 was probably the year to retire and enjoy the well-deserved earnings, for those who started in 1995, the year 2000 was probably the year in which they filed for bankruptcy and did not recover not even the investment.
What is happening is exactly that. The world of pioneers that collides with the world of liquid business introduced above.
Liquid business VS non-liquid business.
It is clear that an economy of scale that grows in a healthy way necessarily leads to profits as high as growth is high. However, growth cannot necessarily be infinite, and above all the economic resources of companies that do not have the strength to grow and are limited to server rooms with a thousand servers and very high management costs.
These companies, certainly with extremely competent technicians able to deal with the management of an Autonomous System and routing at the BGP level, however they clash with new school sysadmin or rather devops or Full Stack Developer who simply rent or rent servers or instances from datacenter companies. that make economies of scale not with 10 servers, but with MILLIONS OF SERVERS.
In short, the “liquid” entrepreneur discharges the business risks on the company that does economies of scale and agrees to pay 10% more on a theoretical cost that he could not have afforded anyway.
It is useless to reason in the hypothetical world of "if". If I had 10 million to invest I could have sold or bought a server for 31 euros a month. You do not have the capital, you agree to pay it 40 and top up on what is the real value-added service.
A $ 40 per month server is just a $ 40 per month server. A 40 euro per month server capable of providing and supporting 200 million page views per month with advertising revenue of about half a million per month, is no longer a 40 euro per month server, but is a tool that allows you to earn half a month. million to the publisher (end customer) and that by virtue of this and its function can easily be sold for 1000 euros per month.
Those who sell services do not need and do not care to have their own datacenter, is interested in having a datacenter that follows economies of scale and has better technical and technological resources for him and for the datacenter itself, with the awareness of finding in the datacenter itself everything needed for the hardware management of the machines (spare parts, failures, repairs, new installations), everything needed for networking management (redundant routing, BGP management, switch management, routers and network equipment), and everything needed from the point of view of energy management, perimeter security, access control, management fires and so on,
This is why in Italy from the Italian datacenters you will see increasingly higher prices compared to Hosting companies that are supplied from the "usual" German or Dutch datacenters, because economies of scale are to the advantage of datacenters with millions of servers.
And even in Italy, the best prices from Italian datacenters are found in datacenters that apply economies of scale and manage hundreds of thousands of servers.
We have seen how some pioneering business models have become extremely difficult to conduct in an increasingly competitive environment in which the economy of scale necessarily takes over and puts small and medium hosting companies in difficulty as they find themselves having to do the accounts with fixed management costs and lack of competitiveness with large data centers with 100 employees and 500 million in annual turnover.
Is it really necessary to continue selling "iron" rather than selling services and focusing on services and purely systemic aspects?
Do we still need to build proprietary datacenters which, when viewed from Google Earth, are essentially small warehouses with a two-meter high gate and not even barbed wire?
Is it really necessary to build proprietary datacenters and make millionaire investments to guarantee the bare minimum where a systems consulting company like ours is able to provide more, better and low-cost services as we do not have significant costs in the management of our virtual datacenter?
At the time of Amazon AWS, of the Cloud, of the disposable instance, of 1000Gbit of bandwidth used for only 24 hours, and contracts that are born and die with the same speed of the flapping of a butterfly's wings, it makes sense to think again with the server "rooms" in order to to say proud and satisfied "we have proprietary datacenters"?
Who cares ?