By Russell Bristow
Most of the money wasted on global networking is spent in things that aren’t put to full use. It goes in the fixed contract costs, all the extra renewal costs of cloud subscriptions and services and in the three years of extra telco capacity bought up front because the alternative was going back to the board and asking again next year.
That is the problem we have brought Alkira in to work on. Alkira now sits alongside Meter in CloudCoCo's Network-as-a-Service portfolio as our latest technology alliance partner. Meter looks after the network inside the building. Alkira looks after the backbone between buildings, clouds and regions.
Why global networking costs what it does
The usual explanation is bandwidth prices, and that is mostly wrong. Bandwidth has been getting cheaper for years. What has not got cheaper is the structure wrapped around it.
A typical global estate carries a few things at once. Long-term circuit commitments signed when the site map looked different. Colocation space that was taken out to host kit, most of which has since moved to a public cloud. Firewalls, routers and load balancers bought as appliances and virtual, each on its own refresh and renewal cycle and its own support contract. A managed service provider charging for change control on a network the customer is no longer really using the way the contract assumed.
None of these are bad decisions. They were all sensible on the day they were made. The cost shows up because the business changed and the contract could not keep up.
I have sat with IT leaders who can tell you to the penny what their MPLS spend is and cannot tell you what it costs to connect a new office in Frankfurt next month, because the answer depends on three suppliers and a procurement cycle. That gap is where budget disappears.
The colocation exit nobody finished
A lot of organisations started leaving colocation five or six years ago and stopped halfway. The applications moved. The data moved but the network did not, because the colo was also the place where everything met: the carrier handoffs, the internet breakout, the security stack, the private links into cloud.
So you end up paying for a data centre that no longer hosts data, purely because it is the junction box. Rack space, power, cross-connects, remote hands, and a carrier contract that renews quietly every year.
Finishing the exit means moving the junction box itself into software. That is what Alkira does. The connectivity points, the routing, the segmentation and the security insertion all run as a service, provisioned in a console rather than assembled from kit in a cage.
The saving is real, and the second-order effect matters more. Once the network is not tied to a physical location, adding a region stops being a project and can be done in software.
China, and how to connect
Every multinational IT leader I speak to who operates in China describes the same experience. It works, eventually, and nobody can quite explain the process.
Connectivity into and out of mainland China carries requirements that do not apply anywhere else: licensed in-country providers, ICP filing, local presence rules, and performance that falls apart if you try to serve Chinese users from Hong Kong or Singapore and hope for the best. The usual workaround is a dedicated circuit, an in-country partner, a separate contract and a separate set of people to call when it breaks.
Alkira has built China connectivity as part of the same platform rather than as a bolt-on. Same console, same policy model, compliant in-country termination. For a business running sites in Shanghai and Shenzhen alongside everything else this is not only much simpler but also much better value, and much easier to manage operationally.
AI traffic is the thing breaking the old sums
The networks most organisations are running were sized for people opening applications. The traffic now includes model training pulls, inference calls crossing regions, and data being moved between clouds because that is where the GPUs happen to be available this quarter.
That traffic is bursty and it is expensive in egress charges. It does not fit a five-year committed circuit, because nobody knows what the volume will be in eighteen months, let alone five years. Buying for the peak means paying for idle capacity most of the time. Buying for the average means the thing falls over when it matters.
Consumption-based networking fits the shape of the problem better. You scale when the workload scales and you stop paying when it stops.
Network as a Service Economics
I joined CloudCoCo in July to build something useful for IT leaders that would ensure we brought value as a reseller partner. Every technology vendor we sign has to pass a simple test: does it make the customer's job easier, does it save them money, and can we stand behind it when something goes wrong at two in the morning?
Alkira passed that test because the economics are legible. An IT director can look at a current colo and circuit bill, look at what the same connectivity costs as a service, and make a decision without needing a spreadsheet model built by a consultancy. We like suppliers whose value is easy to check.
The other reason is that this is a space where a smaller, faster team can genuinely out-serve a large systems integrator. The big Service Providers make good money from the complexity. We make money from removing it. Those are different incentives and customers notice which side of the table you are sitting on.
Where to start
If you are looking at this, the first piece of work is not technical. Start with looking at the cost of China connectivity or the real cost of Co Lo including the connectivity charges or the sprawl of virtual cloud firewalls and load balancer subscriptions and then ask yourself are you getting good value from the spend. If you are doubtful, then the CloudCoCo team and I are happy to sit down with anyone working through this, whether or not it leads anywhere. The conversations are usually the useful part.
Leave a comment!