What if cloud computing becomes the fully integrated steel mill of the future? What would disruptive mini-mills look like?
I was musing on resiliency issues recently, which led to an interesting line of thinking that I’d like to share.
I use the mini-mill analogy because it’s often cited in business school stuff about strategy, and there’s a big section about it in Clayton Christensen’s oft-cited work The Innovator’s Dilemma. You can read the appropriate back-story on mini-mills here.
Cloud As Integrated Steel Mill
An integrated steel mill produces steel products from raw iron ore. Mini-mills do not. They use scrap steel as their input, so they skip a big (and expensive) part of the supply chain. Mini-mills thus cannot take over all of the market, because someone has to make the steel in the first place, but they have about a 20% cost advantage, with a penalty for quality.
Let’s imagine the cloud companies, like Google, Facebook, AWS, Azure, etc. are integrated mills. In this analogy, they take raw materials of infrastructure components (CPUs, storage devices, network silicon, code, developers) and turn it into computers that run code. It costs a lot to build an AWS Availability Zone, so there’s a certain minimum scale you need to have to make it cost effective to compete in that market, just like integrated mills. The margins are pretty good, at least in some segments, and better at the high-end enterprise computing market than down at the ‘free introductory tier’ level. Again, pretty similar to the steel market.
Ok, so what if you could re-use existing cloud ‘scrap’ to build a lower-cost, smaller scale cloud? We kind of see that already, with technologies invented at Google, Facebook, etc. getting released into the wild via patent disclosures, public papers, or outright release as Open Source. There are whole companies, some quite large, built on ‘scrap’ from Google, with a few tweaks of their own. They’re not quite mini-mills outright. They’re more like the suppliers of the components to make your own mini-mill, in some ways.
A closer analogy to a mini-mill would be a cloud provider with less scale and features, but cheaper. It wouldn’t need to have all the bells and whistles of AWS, because AWS is trying to build something to appeal to many market segments, from small to enterprise, just like the integrated steel mills. A cloud mini-mill might make good margins at smaller scale by not having to spend billions each year on R&D to outpace Azure, instead tracking a year or two behind as features arrive in open-source versions.
A cloud mini-mill might be localised to a single geographic region, providing the data sovereignty required by that region. Like the EU, where the safe-harbour framework has been declared invalid, and who don’t like having data exfiltrated back to the US because the US Department of Justice says so? One big advantage to a local cloud mini-mill is network latency, because the laws of physics aren’t likely to change quickly, and light only moves so fast.
There are a bunch of issues with doing this, such as the minimum economic scale required to run a data-centre. The large companies who buy in bulk can get better prices out of their providers than smaller places can, generally, even if they’re both housed in a NextDC or Equinix facility. Similarly, they have price advantages on buying the raw hardware (CPUs, storage, network) for building the plant. I don’t think it’s possible yet for a smaller player to build a cost-competitive cloud mini-mill, at least, not with a ~20% per-unit margin advantage.
But that doesn’t mean it’ll never be possible, and we’ve only really just started the first wave of fully integrated cloud mills. And they’re basically a service bureau from decades ago, so the rent-vs-buy pendulum has already swung from rent to buy once before. It could again, just as businesses used to run their own generators, but now buy power from the grid, yet rooftop solar means we’re starting to generate our own power close to where it’s used.
It’s a thought.