Last week’s outages showed what we rent. Here’s how to know where you stand: Cost, Control, Exposure.

An ancient legend warns us about mighty idols built on feet of clay.
Last week, a series of profound outages bared the clay feet holding up global business and the US’ infrastructure. Over a 48-hour period starting on November 12th, outages hit key platforms and spread like wildfire. Thursday, Microsoft’s Teams and Outlook were unavailable to users for more than a day. Friday, the FAA lost its radar and communications, stopping all air traffic in the US for four hours. Next, Workday went down and hoped to have most customers restored by the next day.
For Microsoft and the FAA, explanations were thin, contested, or slow to arrive. Systems stopped, and the customers were just told to be patient.
Governments, courts, hospitals, banks, and airlines stopped, not because anything was wrong with them, but because they are all built on someone else’s systems.
That’s the bargain of modern infrastructure. We don’t run it. We rent it. We rent out data centers, HR systems, secure communications, and even our national air traffic infrastructure.
Renting can be a major advantage. It outsources things to experts. It shares costs with other consumers. It gives you access to the latest and greatest. But last week also reminded us: you are only as strong as your weakest link.
Like all systems, these core platforms exist to maximize shareholder value, or, in the FAA’s case, decades of deferred upgrades and political budget fights. Their priorities and yours might not always align.
Right now, your board is reading the same headlines you are. Your investors are asking the same quiet question: could that happen to us?
When you outsource infrastructure, you transfer operational responsibility, but you do not necessarily transfer business risk. In almost all organizations, a distance exists between the people making outsourcing decisions and the people bearing the consequences.
If you are living in that rent vs. own world, this is the moment to know where your infrastructure, and possible consequences, actually stands. You need an answer for CCE: Cost, Control, and Exposure.
Cost
The Success Tax

Cost is a key driver in deciding whether to rent or own.
Using someone else’s platform usually has lower starting costs. You can quickly and cheaply bootstrap an application in Lovable or a dashboard in Tableau Cloud.
If you want to bring these capabilities in-house, setup costs are higher: hardware, software, talent, etc.
But third-party platforms structure costs differently. They have to operate the service AND make a profit, which comes out of your margin. If you use these services for back-office internal operations, they will probably scale slowly and won’t be a major cost driver for your business.
If these capabilities are inputs to your products, though, then the more products you ship, the more platform you need to buy. Their fees become a success tax. The better you do, the better they do, scaling their costs with your consumption.
In manufacturing terms, the goods and services used in your products aren’t overheads; they are inputs. They scale with your business.
Manufacturers, restaurants, and such look for cheaper suppliers as much as they look for new sales.
At some point on the graph, your success tax costs more than setting up and running it yourself. You passed the inflection point.
Do you know where your inflection points are?
Google asked that question early on. Rather than buy servers, it built its own. Standard server power supplies of the time ran at 50 to 70 percent efficiency; by spending a little more per machine, Google pushed its servers as high as 92 percent. On a single box, that’s a rounding error. Across hundreds of millions of dollars of hosting, it’s a “success tax” windfall.
The Inflection Point Moves
In tech, the inflection point historically drifts left over time.
20 years ago, I needed a $12M Oracle/Sun cluster for big data analytics, plus a staff of highly qualified, expensive experts. Today I can process much more complex data projects on a $3,000 laptop running open-source Spark, Pandas, and Jupyter Notebooks. Not only has it gotten cheaper to set up, but it is cheaper to run. With access to Hugging Face, even I can do analysis that used to require a staff of experts, all with a pretty interactive notebook!
The boom in AI and open source, combined with the relentless march of Moore’s law, is changing the landscape and pricing models faster and faster.
If your rent vs. own model is more than a year old, it probably is out of date.
AI is the extreme example. While companies announce multi-million-dollar budgets for next year’s service agreements, self-hosting is collapsing setup costs. Colibri, an open-source inference engine released this summer, runs a 744-billion-parameter model on a laptop with about 25 GB of RAM by streaming the model from disk instead of holding it in memory. It isn’t fast. But a year ago, that required a data center. Tools like Ollama and llama.cpp are pushing in the same direction.
The AI inflection point isn’t just trending toward cheaper; it’s a pole jump. The newer the tech, the more savings potential.
What are your major cost drivers? Who are your major vendors? Which ones are still a good choice, which ones are getting close to crossing over, and which ones have already crossed?
Control
Does your customer care?
We outsource a lot to gain the advantage of dedicated products and experts. Some are commodities, like cloud computing or authentication systems. Others are specialties that aren’t central to your company’s mission, like HR or payment systems.
You don’t need to control how these systems work; they just need to work as advertised. You are paying for the promise that, as the market evolves, your suppliers will evolve with it and, by renting, you’ll keep access to the latest and greatest.
However, some capabilities are core to your business. You would do things differently in-house than a vendor selling general solutions. You need products that are unique to you because they’re part of your differentiation to customers.
If it is a capability your customers or stakeholders would notice a difference in, you need to own it to control it.
Few customers would notice if you swapped out one cloud provider or logging system for another. They would very much notice if your shopping cart got easier or your support bot could actually help them faster than irritate them.
Apple is the extreme example, controlling everything from the M-chips to their hardware to their OS; they need the full control to deliver their distinctive user experience.
Netflix owns its own CDN because, for a streaming product, good content delivery tailored to their needs is a key product differentiator.
Sometimes Control Is Binary
When I worked at Lottovate, we ran a white-label platform for charity lotteries, all of which required gambling and government certification.
At the time, no major cloud provider would host what they considered gambling services. Even if they would, they couldn’t provide the hardware security required for certification. Our core draw system had to use certified, physical random number generators, be air-gapped from the network, be physically secure (some servers were hosted in a modified vault), and be set up to be immutable. The whole system booted and ran from a DVD-ROM on a machine with all the ports glued shut.
In that case, our ability to control the environment was our selling point.
Sometimes Control Isn’t Binary
Sometimes It’s Not
The other mistake is control, but at the wrong level.
Lyst is the world’s largest fashion search engine. Our core product is search and recommendations, so we need to control the search engine.
I have seen too many teams do this. They start too low-level. They build their vector databases or RAG models. They roll their own search engine on top of their own databases. They develop in-house query processors. And all of this is BEFORE they get to the real work of searching and recommendations.
At Lyst, we worked with an open-source platform that we could control; we managed it in-house, but we took advantage of (and contributed to) a community of experts. We built on a flagship search product and stayed on top of it. But we chose a modular, extensible product. It let us control the parts specific to Lyst.
We had our own modules for language processing, recommendations, graphs, weights, etc. We spent more of our time on building AI/ML modules and training them and retraining them, as we liked to say, at the speed of fashion.
We didn’t need to create our own vector database or build our own search engine; we needed to know what was commodity search and what was our special sauce.
The real question isn’t do we need control? It’s at what level do we need control?Usually the answer is a platform you can extend, not a foundation you have to pour.
Exposure
Exposure to Someone Else’s Priorities
Cost and control are decisions you make on a good day. Exposure is what you live with on a bad one.
Some exposure is geopolitical. In October, Starlink imposed a speed limit on its terminals in Ukraine, cutting connectivity for fast-moving drones in the middle of a war. SpaceX said Ukraine had broken its terms of service. Whatever the rights and wrongs, a nation’s military capability turned out to depend on one company’s decision.
Europe noticed. In June, the European Commission launched a tech sovereignty package. American companies take roughly 80 percent of professional cloud spending in the EU, and as one official put it: “We want to be sure nobody has a kill switch.”
Exposure to Someone Else’s Mistakes
Most exposure is more mundane, and that makes it more dangerous.
Workday wasn’t hacked. It was testing its disaster recovery. Services in its backup site connected to live production databases and left customer data in an inconsistent state. The safety net pulled down the trapeze. From Friday evening until late Saturday morning, employees at more than 11,500 organizations couldn’t clock in, check their pay, or finish benefits enrollment, in the middle of enrollment season.
None of those 11,500 companies approved that test. None of them could see it coming. They found out when their employees did.
That’s exposure. You can audit your own change process. You can’t audit theirs.
Revenue Loss vs. Relationship Loss
Exposure isn’t just about how likely a system is to fail. It’s about how close that failure sits to what your customers see and care about.
In April 2025, a cyberattack forced Marks & Spencer to halt online orders for around seven weeks.
For non-UK readers, M&S is a brick-and-mortar clothing, food, and home retailer. While they do online shopping, most customers shop in real life. If their app can’t take orders, they can still shop in person.
It was painful, but not core to M&S’ relationship with their customers. First-half profit fell by more than half, but has since recovered. It inconvenienced customers, but it wasn’t a betrayal of trust. It damaged revenue, but not their standing with customers or their relationships.
In short, the exposure was painful but survivable.
Imagine the same outage for a digital-native service. Imagine if Uber had gone offline for seven weeks. There is no easy alternative way to access their service. People who had depended on Uber would have been left in the lurch, scrambling to find other transportation apps.
There are many options. Users would have found them, switched, and established new habits.
Uber could come back after seven weeks with an apology, an explanation, maybe even vouchers for free rides; but the damage would have been done. It wouldn’t have only hurt revenue; it would have hurt the customer relationship.
For digital-native businesses, that level of exposure could be extinction-level.
Where are you exposed? Whose decisions affect you that you don’t control? And what does each exposure put at risk: your revenue, or your right to stay in business?
Where This Leaves You
Every idol stands on something. The question is whether you know which of your feet are clay.
Another outage is coming. That isn’t pessimism; it’s the nature of shared infrastructure. The only question is whether it will be a bad day for you or an extinction-level event.
Renting isn’t the mistake. Not knowing what you’ve rented is.
So run a CCE audit this quarter. List your critical third-party dependencies, and ask three questions of each:
- Cost: Have you passed the inflection point where running it yourself is cheaper? Has new technology moved that point since you last looked?
- Control: Would your customers notice the difference? If so, at what level do you need control, and can you extend a platform rather than rebuild?
- Exposure: If it disappeared for a week, would you lose revenue or lose the relationship? Could you survive it?
After last week, they’ll ask you these questions. Your board, your investors, and your executive team are nervous, and they’ll look to you for certainty.
You don’t want to be working out the answer in that meeting. You want to say: We’ve already looked at this. We have a system. We know where we stand, what we’ve decided, and where we’re going next.
That’s not a technology answer. That’s leadership.
