Server Vendors, RIP?

In recent months, I have come across a new piece of received wisdom when in conversation about the cloud computing market. It goes a little like this.

“In the future people will no longer buy servers, they will simply run all compute workloads on the cloud.

Cloud is commodity, so scale will win out.

In the future, only a small handful of large scale global hosters will exist, killing smaller sub-scale players, and eliminating the need for server vendors as we know them today.“

I’ll save some of you reading this through to the end by saying I don’t agree.

For the rest, let’s walk this through.

It is rational to say that the scale, security, and ease of use of cloud platforms for almost any compute workload supports the first part of this thesis. It makes complete sense to move to cloud services, whether at the infrastructure level or up to delivery of platforms and software applications. Most people are used to installing an app and just consuming the compute, storage and services delivered seamlessly by the cloud platforms that sit behind the lovely IOS or Android interfaces we already take for granted.

It is also true that IT is increasingly appearing to be commoditised, in which case scale is the only way to survive such an onslaught. The argument made by most is to look at the electricity markets, with standardisation around AC despite the energy efficiency of DC (a 19th century Betamax story?).

To the final point, it may well be true that we end up with an oligopoly of providers, and decide that neither choice nor competition is required in the market for compute, storage and services. If that is the case, it is also true that these small number of global players are likely to vertically integrate, and will design and build their own hardware / software ecosystem to support their services. This is indeed already the case with the web-scale operators like Google and AWS, neither of whom buy from traditional server vendors (although they still have a number of server manufacturers they rely on to build to spec…).

However, each of these three points are more complex and not the end of the story just yet.

In the future, no one will buy servers

This may well be the case for many, however whilst the internet age was in full swing leading up to the turn of the millennium, and AWS was launched ten years ago, we are still in an era where on-premises and self-managed datacentre infrastructure makes up the majority of deployments.

For example, in the shiny new world of big data analytics, a survey by ESG in 2015 found that only 21% are considering public cloud deployments, with a further 10% considering hybrid on-prem / cloud solutions.

Server shipments grew again in 2015, in particular with hyperscale platforms in North America. It is also worthy of note that the top five server server vendors by revenues are the very players who do not sell to the the web-scale cloud providers.

This is not exactly the picture of an industry in decline.

Cloud is commodity

This is an argument usually made by people who haven’t had the pleasure of deploying a farm of servers from a linux command line. It is indeed true that for end-users, IT is an increasingly consumer-led experience, especially with the acceptance and prevalence of SAAS. The user interfaces are better than ever, accessibility is global and flexible with an increasingly mobile usage.

Large-scale deployments of infrastructure still require skills, however. Even if you outsource the datacentre, servers, and network to AWS, you still need to manage the OS, application and data layers, security services and of course figure out how to scale out and/or up.

Add to this the fact that you have to do some pretty neat commercial deals at scale with the likes of AWS or Azure in order for the economics to work. In fact, for most mid-sized companies, cloud actually becomes more expensive when you look at the total cost of ownership. In a previous life I ran a hosting and cloud company, and it was amusing to see how customers would react when we did a side by side comparison of a managed solution compared to cloud in dollar terms. Cloud only won if they were so large that Amazon would talk to them about price, and if the workload was truly variable (e.g. the key unique selling point of “cloudiness” was always flexibility — both operationally and commercially).

However, this also assumes you take the baseline operational efficiency of your cloud provider as the optimal case for your workload. As evidence from the big data industry shows, when your application or demands scale up, the economics of optimisation come into play. Ask yourself, for example, why Netflix uses AWS for the member website, iPad and other devices management, but as soon as you hit the play button, the content comes from dedicated CDN platforms. It comes down to optimising the use case — and for that you cannot truly commoditise. There are simply too many variables.

In order to answer these different workloads, you need variety in supply and supplier alike.

Only a handful of web-scale clouds will survive

This thesis relies upon a couple of key assumptions.

The first and most important is that consumers are happy relying on a small number of providers and trust them to keep innovating, driving costs down, and improving service. The evidence is that this kind of oligopoly hasn’t fared so well in telecoms. So the question is, why would the consumer of such infrastructure tolerate it? Smaller niche hosters continue to grow, services migrate from what are seen to be legacy systems to AWS and then boomerang back to either on-prem or managed hosting providers who can assure the suitable IOPS for chunky databases or other applications.

It is just that this “boomerang” scenario is less well publicised, as no one wants to publicly acknowledge poor planning or strategic failure in deciding to move to cloud.

Hence the growth in “enterprise cloud” solutions from the mid-market players who are focussed on delivering services geared towards the use-case of such enterprise applications.

To be able to move to web-scale clouds seamlessly assumes your application was written with web technologies in mind (e.g. distributed datasets like hadoop, and load balanced front-ends in Python or PHP).

Safety in numbers?

As adoption has increased, the reality is that cloud is coming close being able to sell on the basis that “no-one ever got fired for buying IBM”. If you host with Microsoft Azure, some might see themselves free from operational responsibility if their platform falls over. So the fact that increasing numbers are attracted to cloud or hosted services is not a surprise. And the advantages of outsourcing to infrastructure specialists are well documented and understood. The question is whether we will entrust such infrastructure to 3 or 4 players, or if the multi-trillion dollar industry will still have a healthy variety in supply.

The key question is whether a small number of suppliers can answer all the questions not just operationally or in service innovation, but also from a security and jurisdiction perspective. Since Snowdon, many companies already refuse to host with a US-based player. To sell to the German markets, Microsoft is now working with incumbent local players like T-Systems.

All of this adds up to the fact that between the clear need for innovation, variety in supply chain and services, it is difficult to see how a small number of very large players can sensibly support the global IT landscape.

Server Vendors Are Dying

When you look at the woes of HP and Dell, you would have thought Rome was burning and that taking firms from public to private then public again was akin to tuning your violin.

The truth is perhaps easier to see by looking at another sector completely. It is true that the pharmaceuticals industry is dominated in revenue terms by global players like Astra Zeneca, Novartis, and GSK. They are however having real challenges around product innovation and pipeline, relying on acquiring startups who take new molecules to stage I or II trials, derisking the innovation pipeline substantially.

The big pharm guys then use their industrial strength and processes to get the new drugs over the line, with FDA or EU approvals, as well as established sales channels to assure access to markets.

In the IT infrastructure space, it may be more worthwhile considering a similar story, where the large manufacturers like Dell, Lenovo and Quanta (who make the kit for Facebook) look to startups for new ideas and platforms — whether in server technology, storage, network or now hyperconvergence (horrible term that it is). The startups act as market makers in new areas or with new ideas, and the big guys come in and help scale up their route to market.

The long and the short of it is that this enormous industry is not going to disappear in a puff of smoke, regardless of the well publicised growth of public cloud providers. There will always be a need to match market demand to a varied and innovative supply chain — it is unlikely that by relying on an oligopoly of cloud operators the IT sector will manage to do just that.