Bigger, faster, cheaper

As data traffic and the demand for data continue to grow exponentially, the hyperscaling trend is throwing up unique requirements and challenges.

With data traffic growing at 28 percent per annum, and data storage requirements at 40 percent*, it is not surprising that the big cloud operators have been building everything bigger, and wanting it ever-faster.  In 2016, this appetite led not just to big capex spends, but also to a major shift towards the lease space market.

Amazon, Microsoft, Google and IBM comprise over 50 percent of the cloud, and the first three between them invested US$26 billion in 2015*, largely in new hyperscale data centers – campuses with 100MW+ potential installed capacity.

This trend is set to continue as the demand increases which have been forecast, materialize, and the number of global hyperscale data centers is predicted to rise from 300 at present to around 485 by 2020.

So, seemingly all rosy in the garden for the industry, but fluctuations at this scale of growth lead to big headaches.  It takes a site-based workforce of around 1,000 people to build one of these projects, and its duration will span anything from two to ten years, depending on how the phases are timed.

That can mean mobilization and demobilization several times on one campus – not easy where skill shortages are already appearing as the economic tide rises around the world.  Workforces are quick to move to other locations and other sectors, dissipating hard-won experience in doing so.

For the owner / operator, the unpredictability leads to volatility in demand forecasted, and makes short-term investment planning very difficult.  More than one of the big operators has scaled back capex plans this year, awaiting a little bit of a catch-up.

Despite the push for bigger and faster, cost remains a significant factor, as each of the big players eye up their competitors’ cost base.

Traditionally, lease space has been the flex play in the cloud operators’ game plan, offering shorter time-to-market (TTM), which is attractive in this fluctuating demand scenario.   By developing sites to a pre-prepared stage as a grey box awaiting fit-out, or at least as a stoned and compacted site with utilities connected, the lease players have long been adept at rapid installation and power-up.

It also offers off-balance sheet options, sometimes attractive to investors.  Fewer internal resources are tied up in its acquisition (no internal engineering team required), and simpler, frills-free specifications can deliver genuinely lower infrastructure build costs.

However, the large take-up of lease space has attracted new players and stretched the resource base, questioning whether shorter TTMs will actually materialize.  Whilst in theory the delivery risk has been transferred, no amount of liquidated damages will bring a delayed schedule back on course, and if milestones are missed the leasee carries the can.

Modularization is back (if it ever went away) as a trend for 2017, the idea being to sequentially build out these hyperscale data centers in sensible, repeatable lumps, benefiting from procurement and design savings, whilst spreading the delivery timing.  As everything has gotten bigger, the data center in a container has been outgrown, but containerized and modularized switchgear, generators and UPS equipment have become very interesting.  

Lean construction strategies dovetail nicely with this off-site fabrication approach, so more of that will be seen, as well as a bigger role for the integrator of these different equipment elements.

All of this demand for more data ultimately leads to demand for more power, and several locations are feeling the pinch.  Sites with direct access to 400kV power circuits, the super grids of the host countries,  are the smart places to be; even the 220kV circuits in urban settings are no longer offering the security of supply at the required capacity that would have been perfectly acceptable five years ago.

Despite the push for bigger and faster, cost remains a significant factor, as each of the big players eye up their competitors’ cost base.  It is notoriously difficult to compare like with like, but whether the capex cost per MW is $10 million or $7 million, it will never be low enough.  However, data centers are like a box of chocolates – the fancier they are, the more they will cost.

*: Cisco Global Cloud Index

*: MarketRealist.com