Losers in the cloud revolution

The internet is changing the way every business is run. Aivars Lode avantce

Losers in the cloud revolution 

he shape of the IT industry is due for a shake-up as cloud becomes a utility, due to the greater influence that cloud providers can wield over IT suppliers.

The winners are likely to be those companies that can serve the peculiar needs of huge infrastructure operators, like Amazon or Google, or those that can give advanced software tools to businesses to let them spend less on costly hardware.

Judging by the results of a round of interviews with industry figures and analysts, combined with recent technological shifts, the companies that have the most to lose are those saddled by legacy businesses selling technologies developed in another time for another market. 

Established hardware makers

Original equipment makers (OEMs) like Dell, HP, EMC and IBM that specialise in selling datacentre equipment to large enterprises are going to face difficulties for two reasons: the trend for small companies to opt for cloud services, and the disruption of the datacentre infrastructure market by the move to equipment commodification.

Chip companies stand to be hurt as well, although to a lesser extent: the cloud means fewer companies buying ever-larger amounts of chips, which will hit overall margins as these massive buyers can demand volume discounts.
Chip companies stand to be hurt as well, although to a lesser extent: the cloud means fewer companies buying ever-larger amounts of chips, which will hit overall margins as these massive buyers can demand volume discounts.


The mechanism for this is fairly simple: if a company opts for hosted email, like enterprise Gmail, and they were using on-premise Microsoft Exchange previously, then when they go to the cloud they no longer need their Exchange server. The equipment spend disappears and the cloud service has eaten the on-premise expenditure.

A recent Gartner report found that as companies went to the cloud for cost savings this could sometimes cannibalise on-premise equipment spend. 

Whenever cloud-based IT service companies, like Salesforce, report growth this comes at the expense of spending on on-premise software and hardware. Along with this, these types of 'as-a-service' applications have a heavy emphasis on mobile devices, so they can drive either IT department spending on mobile devices or a slacking of rules to allow workers to take part in the BYOD (Bring Your Own Device) trend.

"Enterprise [IT] is being squeezed out in a barbell," says Chris Swan, the chief technology officer of client experience at UBS. "On one end [there is] commoditised consumer technology [and] at the other end commoditised datacentre technology."

This trend means modern enterprises are spending more on client devices and cloud services and less on the middle layer — the on-premise computers, network, server and storage gear.

Along with cloud cannibalisation, some of the world's largest infrastructure operators — Google, Facebook, Amazon Web Services, various major banks — have taken a look at OEM datacentre equipment and found it wanting. Instead they have begun designing their own. 

Facebook, for instance, designs its own server and storage gear to be manufactured by Asian technology manufacturers like Quanta, Foxconn or Wistrom (known in the trade as 'Original Device Manufacturers', or ODMs, from their heritage of manufacturing consumer products like netbooks).

Search giant Google has been designing stripped-down servers for many years.

"For the last few years as far as I know [Google] have never purchased servers from a branded vendor," says Giorgio Nebuloni, research manager for servers for IDC’s European Systems and Infrastructure Solutions division.

Amazon is thought to do the same, and recently launched a low-cost cloud storage technology named 'Glacier' that's believed to be based around custom Amazon-designed equipment.  

Companies doing it for themselves

Intel can attest to how enthusiastic these companies are about building their own equipment: In 2008, HP, Dell and IBM made up 75 percent of Intel's server processor revenue. Four years later, eight companies are on that list, with Google ranking at number five, Diane Bryant, head of Intel's datacenter group, told Wired Enterprise in September.

By building their own equipment cloud companies can sidestep typical OEMs and achieve significant cost reductions. Not only does this reduce the pool of customers available to the large OEMs, but it also bites into their margins as they buy proportionally fewer components so get lower price reductions. 

The commoditised-equipment trend is also visible in smaller cloud companies, like web hoster and OpenStack-supporter Rackspace. 

As of today, "at least 10 percent" of Rackspace's servers are either built to the ODM model or bought from OEMs' datacentre-specific wings, such as Dell's Datacentre Solutions Unit, John Engates, Rackspace's CTO, says.

"In a number of environments we've chosen to buy equipment from some of the guys like DCS, but are also working with [ODM] companies," Engates says. 

Even if Rackspace buys directly from an OEM, the equipment it buys will be stripped down and lack many features that are found on traditional OEM datacentre equipment, so it will cost less and generate less money for the OEM.

This means more revenue to the Asian ODMs, which are branching out: in May Quanta launched a US subsidiary to sell configurable datacentre infrastructure that undercuts traditional suppliers. 

"Our flagship cloud product, which is our fastest growing product, is based on DCS or ODM gear," says Engates. "Over time the trend is that less and less will be enterprise-branded gear and more and more will be [DCS-style] OEM or ODM."

Industry response

The Dells, HPs and IBMs of the IT industry have not been blind to this shift, and have been attempting to service large cloud customers by producing equipment for them through specialised divisions, such as Dell's Datacentre Solutions Unit. They have also been targeting medium-sized enterprises that would otherwise go to the cloud with appliance systems that wrap storage, servers, networking and software together.

This strategy addresses the commodity hardware shift in two ways. The specialised divisions target large infrastructure buyers with products tailored around their needs. Then, the appliances are sold to smaller companies that want to use a cloud architecture, but do not have the expertise to build their own gear. 

Such a strategy has its dangers. One is a margin-sapping price war with Asian ODMs on datacentre equipment; another is that appliances are vulnerable to open-source software tools. As these improve, smaller companies can build their own appliances by buying gear from an ODM and putting mature open, free software on top. Open source is strong in infrastructure management, with many large IT-based companies seeing the benefit in collaborating in the open development of key technologies traditionally forming rich revenue streams for OEMs and systems vendors. 

"The challenge in the long term, perhaps three to five years time, is how you can differentiate and compete against some of these [Asian ODM] competitors that are able to deliver at lower prices," Nebuloni says. 

The software-defined networking apocalypse

Networking looks set for a shake-up due to the arrival of software-defined networking. SDN lets companies separate the network's data plane (the part that forwards packets around the network) from the control plane — the more sophisticated layer that dictates, monitors and controls the overall structure of the network.

This technology means companies can outsource intelligence of the control plane to commodity servers while leaving the less sophisticated forwarding work to less intelligent — and therefore cheaper — networking equipment. This has the potential to bite into the revenues of established networking companies like Cisco, Juniper and Brocade as many of their products wrap complex control plane features directly into the networking gear.

By separating the two, SDN means companies can buy less networking gear and, as with storage and compute, can even go as far as to design the networking equipment themselves, as Google has done with its adoption of the OpenFlow software-defined networking protocol for wide-area network access.

The networking companies are attempting to deal with this via new products and acquisitions: Brocade recently announced plans to acquire Vyatta, and Cisco is developing its own SDN technology. 

However the academic community is busily working on the OpenFlow networking standard, and virtualisation king VMware acquired SDN expert Nicira for over a billion dollars to help it take advantage of this shift.

Although network companies are developing their own SDN technology, there are signs that these companies are hobbled by their legacy commitment to combining hardware and software, whereas SDN startups come from an all-software world. Cisco prefers to emphasise software and hardware working in unison — the same line Oracle adopted after it acquired Sun Microsystems — but SDN requires very little sophistication of the networking hardware. 

Meanwhile, the open source OpenStack cloud platform, which is backed by over a hundred major IT companies including HP and Rackspace, has SDN-style networking capabilities thanks to the 'Quantum' networking code that was contributed by Nicira prior to the VMware acquisition. Rackspace have based their cloud on OpenStack, as have HP, so over the coming year we will start to get good information on how SDN works in production. 

Just as the storage and server makers face competition from a new breed of hungry companies, established networking companies are dealing with the same problem.

The chip margin slump 

For this reason Intel has been investing ever-larger amounts in hard-to-commoditise areas, like its massively multicore Xeon Phi coprocessor and associated HPC technologies. However, while these areas are becoming more specialised again after a period where pure commodity processors had edged out the custom designs of the early supercomputers, the market is nowhere near big enough to compensate for the loss of mainstream enterprise. 

AMD, meanwhile, has embraced ARM's low-power RISC processor architecture in the hope it can appeal to power-conscious datacentre operators that have the resources to port their applications from x86 to RISC. Intel hopes to make progress here too with its Atom processors, but the momentum in low power is very much with the ARM architecture and its near-total dominance of mobile and embedded technologies. It too is unlikely to compensate Intel for the loss of margin and reduced OEM influence in the enterprise.