The essay is a bit messy.

Big era of domestic IDC development

Around 2012, the strategic focus of IDC of domestic operators shifted from the first and second tier cities to the third and fourth tier cities. At that time, the telecom operator was a design institute, and the IDC technology and construction department within the operator, so the starting point for IDC site selection was largely based on lower power and land resource costs. At that time, the Inner Mongolia bases of the three major operators were clustered together. Who do you want to sell it to? This is a complicated proposition. Technical and construction personnel only care about design and delivery. Whether they can sell is not their concern. It can also be said that the sales level is not high. Even now, the large-scale data center is often 20MW~50MW. Compared with the base type computer room made by operators in those years, it is not worth giving up. There are many reasons why they can't sell. As an IDC operator, it is useless to calculate how low the TCO is. Those fancy technical gimmicks are also gimmicks. How many customers will feel advanced and reliable? In those days, few customers were willing to rent computer rooms so far away. The rental period of those rented computers is approaching. Are these computer rooms safe now?

Look at the market of the first tier cities. The scenery is unique. Since the end of 2011, 4G has just started, and the era of mobile Internet has come too late, which is much slower than that in the United States. At one time, the global Internet market value was among the top ten, but China should not be on the list. Today, it has occupied 3-4 places. Once a small third-party IDC company, relying on operators to move from the transfer center to the third and fourth tier cities. With the rise of Internet companies, strong demand also drives the demand for IDC in first tier cities. As a result, the rise of many third-party IDC depends on this wave of development. Mobile Internet has only been two to three years. In about 2014, domestic cloud companies also started to relay, and IDC demand is still strong, even now. Operators retreat behind the scenes in the first tier market, have the brand of network resources, and can also get a share of profits by cooperating with third-party IDC companies. This era has led to the development of many companies, as well as domestic and overseas listings. Some did well and sold to other leading companies to make a profit. Time also dictates... In fact, there are only two kinds of IDC locations in China, one is called first tier cities, and the other is called other cities.

Go to stock game

The housing price, land price, GDP assessment and other factors in the first tier cities are mixed together, and the government is slowly finding out how much the IDC industry contributes to the local government? Energy assessment and power approval documents have become the key scarce resources of first tier cities. The government mandated that the PUE of new IDC projects should be from 1.5 to 1.3, and then to 1.25... The resources should be so limited that enterprises should make good use of them, do more value-added services, and improve the return rate of resources

Different Internet services, different network structures, reliability levels, and delays all affect the user experience. Although the cost of third and fourth tier IDC is cheap, how much difference does it have in the proportion of business cost compared with that of first tier cities? In many cases, IDC in first tier cities is still in demand. The supply and demand of resources in first tier cities have a great impact on site selection. The site selection of the computer room in Shanghai has been to Baoshan several years ago, which is close... What about those in Baosteel and Taicang? Not to mention Beijing. In southern China, there has been a slight oversupply in Guangzhou in the past two or three years, and many IDC has not found tenants. On the contrary, Shenzhen has a tight card rating system. In the era of stock game, everyone first played the card of resources, and the card of technology... nothing would happen.

The stock game is not a bad thing. Similar to Shenzhen, the results of the stock game will spill over to the surrounding cities if they can't meet the demand. Dongguan is a typical example. Both Huawei and Shenzhen Stock Connect have built their own IDC. However, it is difficult for Dongguan to get a rating. Since Shenzhen launched the big move of PUE 1.25, and then recently the country issued a document to position the future development direction of Shenzhen. I think it's hard for Shenzhen to get big IDC resources again. According to the future national positioning of Shenzhen, the center of the Greater Bay Area, the Asian center and even the international center, as well as technology, port and finance, will be the center The land price in Shenzhen must be very expensive ... It can be compared that there are no big IDCs in Silicon Valley, and IDCs are located far away from Silicon Valley. If Dongguan can't get any comments, will Huizhou be the next stop for Shenzhen's various thresholds? As for Shanghai, the demand has spilled over to Nanjing and Nantong. There is Huailai in Beijing. The stock game seems to be an important label of this era. The spillover effect has created the possibility of the third kind of IDC location, the first tier city ring. As for the fate of the third and fourth tier cities as IDC, the game still needs users to form positive feedback. If there is no final customer, please pray for Buddha's blessing to make the receiver more fierce

Overseas IDC small era

The author is not clear whether the term "small age" is appropriate. At least 10 years ago, many IDCs were 2-5MW in size. In this era, there are not so many third-party IDCs, nor so many large customers. The scale of IDCs is very small, and many operators will consider the local distribution of PoPs. Because many operators had access in the early years, many companies later went to the local area to develop, and they had to go to the same IDC to establish a PoP site. The positioning of these IDCs is neither retail nor carrier hotel. In any case, there must be few racks, small capacity, low power density, high retail prices, and no love. Just in front of us, even Tier2 still needs to be used, even if the machine room is so bad. Who let the family survive in their childhood and accumulate family resources? This is a long-term meal ticket. The typical computer room in Hong Kong is Mega I, which is located on Hong Kong Island. The unit price is very high. Not only the rent is high, but also the toll for the interior line is paid every month. Another one is Singapore's Global Switch computer room. The author doesn't want to comment on it any more. He quotes others' original words, "This is still a computer room.

It's hard to learn this model. The small age has passed. To accumulate slowly, it is better to have PoP points for various operators to access. When the supply was short of demand, the operators were all clustered, and it was difficult to break the situation after they formed a pattern. It doesn't matter if the small era has passed. It's not easy to copy the Equinix model. The only way to pursue it is through acquisition. The big era is wholesale, which can still be copied for fun. We can PK resource capital and supply chain, and the key market value can soar rapidly. If there is more wholesale, we can also play retail to improve profit margins, which can be said that different development directions have evolved on the wholesale road.

Users should strengthen themselves

For many users, which strategy is appropriate for IDC? Large scale users feel that the technical level and design of IDC outside are average, and they begin to develop their own technical specifications. From simple leasing to customized leasing. Of course, this process also allows IDC suppliers and users to share some best practices with each other, which may also optimize costs. But it is hard to tell how much the technology has reduced the cost. For a simple example, if a computer room is designed with micro module or traditional computer room layout, is it cheaper to use micro module? Maybe, maybe not. The author has personally experienced a case that if Hong Kong is designed according to similar micro modules, many places will be wasted due to layout, columns and other reasons, which will increase the cost estimate by 15-20%. In short, a building can rent 12MW originally, but only 9MW with micro modules. Can the rent be expensive? Besides, the rent of the building is fixed. The key to understanding customization from a certain perspective, regardless of how different the technology is, is to change the playing method of a business model, that is, customization by customers. Users began to order their meals according to their own taste rather than according to the brand. In addition, the customized machine room is not in stock yet, so they have to place an order before doing so. The author has read the customized IDC specifications of several Internet companies in Laos and America, as well as the domestic customized IDC technical specifications. Several major customers in China can basically publish a book, and customers in Laos and America generally do not exceed 10 pages.

Over the years, customized IDC has become the mode of mass customers. If Party A raises these people, it must do something and have more control. If the customized lease fails, it will play some other modes, such as agent construction and self construction. The United States has all these models, and American companies are ahead in this regard. Similar to Amazon, it tends to build its own infrastructure and support many people, but this is not the only one. In many cases, from a pragmatic perspective, if the cost of self construction is not cheaper than leasing, why self construction? A typical example is the Hong Kong market. Which American or Chinese company wants to build its own computer room in Hong Kong? Google took a piece of land from the government in Tseung Kwan O in those years, but later refused to do so, and the land was returned to the government. Later, Global Switch received the offer. Even though the land was much cheaper, it was delayed to build the computer room until the deadline for various reasons, otherwise it would be taken back by the Hong Kong government. The IDC market in Hong Kong has long been spoiled by real estate developers. In the early years, the rent was cheap, but now they go to build their own houses. Seeing the rent price is a headache. No matter how optimized the technology is, we can't beat the rent difference... It's not easy for Hong Kong to engage in IDC. With various external restrictions and business conditions, a good computer room is like a difficult surgery. American companies and Chinese companies found that the self built computer rooms in Hong Kong are not as cheap as those rented on the ground, and they are quite complicated. Why do they build their own computer rooms? Since it is expensive, why build it yourself? This kind of thing is only known by insiders. The experience of the author is that this kind of thing is not calculated, but photographed. How can we know if it is cheaper or better without doing it? If it is done, the investment person will do it again... Who is the cost and quality benchmarking of the self built machine room?

Users do not always build their own computers. In the United States, there are also signs that many large users are slowly moving to rent computer rooms. IDC is not a fuel efficient company. After all, the user company has limited understanding of how to do this business. If you stay inside the user company and work as a quiet engineering student, how can any outside businessman play the routine? If you know the routine, will you go out to create a career and realize your ideal? Far from it, let's talk about what I saw in the United States. American companies began to accept some designs from IDC. The design, operation and construction capabilities of these IDC companies are really no worse than those of users themselves. Let's talk about several major wholesale companies in the United States, as well as rising stars such as CloudHQ playing wholesale. Many of them are users and have a deep understanding of technology, delivery and operation. Since the other party has a mature design and operation system, why force the other party to use its own design? Finally, the other side also has advantages in business, why not use it? Microsoft has adopted this model in large numbers overseas. The technical requirements are flexible, and it is not mandatory to use certain technologies.

Deep customized pit

When it comes to user customization in the development of IDC, the larger the customer, sometimes the more they like in-depth customization. Deep customization technology is unique, which makes IDC technicians talk about with relish. PR students also have something to boast about. Even if it is customized, every company will be different, and there will occasionally be a chain of disdain between various technical factions... When FB engaged in OCP, most companies focused on participating, in fact, many companies mainly played their own game. Apple doesn't use FB, Microsoft doesn't use Google, and Google pushes its own... Is this scene a bit similar to the domestic ODCC? According to the hardware company, the product roadmap and product version are eight systems in seven countries. Even though this technology despises the chain, OCP advocates a 21 inch rack, while LinkedIn's chief hardware architect advocates a standard 19 inch rack for compatibility. So this guy set up the Open 19 organization, made use of LinkedIn's platform, and also made a speech in domestic DCD. The author had a few acquaintances with this fellow man and had a slight conversation. Later, the story was that Microsoft acquired LinkedIn, and the Open 19 architecture that the brother worked on in LinkedIn was different from Microsoft's Olympics architecture... First, the integration was business departments, then the company's operating departments, finance, administration, etc., and then the IT system, technology operations, etc., which became popular. Originally, the technology of both sides could be different, playing one country, two systems? Later, after the integration of internal departments, even if the chairman and director of Open19 were successful, Open19 also became Over 19. Users engage in technology and rely on their father's purchasing power to influence the supply chain and market structure. After such a long time, they can't compete in technology... Interestingly, this guy was acquired from FB to LinkedIn and then entered Microsoft. Hehe, paste the profile

Deep customization is not always a good thing. First, scale advantage, transformation difficulty and construction difficulty are all major problems. In those days, Yahoo's data center in Singapore was deeply customized. But what about the IDC? No one wanted it in the market for a long time! Because it is too difficult to use, who would like to be the Pan Receiver? Receiving Pan Xia is an IDC company. After receiving the order, we have to find another customer. As you can imagine, the outcome is low price. So the problem with heavy assets is how to measure recovery when a large amount of money is invested when the sustainability of subsequent business is not clear? Because this is not a production line, it cannot directly generate revenue. If self construction is only 5~10% cheaper than external leasing, it is also necessary to consider various factors of capital costs... If the service life is not enough, even if the external IDC costs are compared, it is not worth every minute of accounting. When self built, it will count as a variable of cost reduction. The same variable also exists in the IDC unit price. This factor should be taken into account. From the perspective of business, the risk of deep customization+self built is not small.

There is also a routine of depth compatibility without playing depth customization. In fact, Amazon and Microsoft have both done well. The author does not comment on the domestic high-voltage DC architecture. In order to unify the supply chain, Amazon and Microsoft insist on the delivery of AC UPS, AC servers and global racks, and launch the so-called Universal rack rPDU version compatible with different server plugs. In this way, at least self built and leased machine rooms can be used. The domestic market is relatively happy, because the supply chain is perfect, but once overseas expansion, compatibility must rank first. The time to open the service is slow, and the satisfaction of that technology is likely to be taught in the business team at any time... FB, which plays BBU and PSU very well, is busy with OCP, and then most of them have to build their own.

epilogue

Like the recent popular American TV series American Factory, the author here provides some documentary and perspective, so that you can have different views.