Clients want everything until they see he bill.
At Navistar, we automated the dealer systems.
Two of them mattered most. SYSTEM 21. And TOPS — the Truck Order Proposal System. These were the systems the dealer network ran on. Ordering. Configuration. The whole front end of moving trucks.
Automating them was the right call. Nobody argued about the build. The build was the fun part. Everybody wanted their name on the build.
My job included the part nobody wanted. I prepared the ongoing budgets.
So I tallied the new development costs first. Then I applied the industry rule of thumb. It is not complicated, and it has held up for decades:
Take 15% of what a system cost to develop. That is what it costs to keep it running. Every year. Forever.
That is not a vendor number. That comes from historic data. Fifteen percent, annually, for as long as the system lives.
Then I added the support load. Eight hundred truck dealerships. Every one of them needing Level 2 support. Every system we built became a system somebody had to answer for — every outage, every night, every franchise with a question at 4 p.m. on a Friday.
I put the numbers in front of them in a budget meeting.
The ongoing maintenance bill was bigger than the new development budget.
Here is what made it land. The dealer franchises had wish lists. Long ones. They asked for everything, because they were not paying for any of it. Free features. Free requests. Free forever.
The second the carry cost hit the page and somebody had to own it, the wish lists collapsed.
The franchises stepped back and asked a completely different question:
What actually moves the revenue needle?
Everything else fell off the list. Not because we ran a strategy workshop. Because they got the bill.
Right now, enterprises are pulling AI workloads back in-house at a run.
The drivers are real. Regulation. Usage-based cloud pricing that nobody can forecast. Latency that hurts production inference. And a compute supply chain concentrated in two countries, which is why the EU, India, Canada and the Gulf states are all building their own.
The market is moving accordingly. Gartner forecasts worldwide sovereign cloud IaaS spending at $80 billion in 2026 — a 35.6% jump over 2025. Governments are the primary buyers, with regulated industries and critical infrastructure right behind them. Gartner calls the movement geopatriation, and estimates it will pull about 20% of current workloads from global providers to local ones.
Vendors are selling into that wave hard. Private AI clouds. Sovereign factories. On-prem inference. The pitch is a clean one: lower total cost of ownership than public cloud.
Every one of those numbers is a build number.
One caution, since I would rather you hear it from me. The figure making the rounds — that 79% of enterprises have already moved AI workloads off public cloud, with 73% planning more — comes from a survey of 203 respondents commissioned by Cloudian, a company that sells on-premises storage. It may well be directionally right. But a vendor-funded number telling you to buy what that vendor sells is not research. It is a sales asset with a footnote. Gartner and IDC are the numbers I would take into a boardroom.
And here is the part that should stop you cold.
Bain's 2026 B2B Growth Agenda — a survey of more than 1,100 senior executives across 18 industries — found that roughly 90% of companies are experimenting with AI, but 60% lack the data foundation or technology to scale it. Bain's separate Automation and AI Pathfinder Survey 2026 (951 companies) is worse:
Bain's own summary of the pattern: "The technology worked. The value didn't arrive."
That 44% is not a strategy. It is a circular bet. Companies are paying for the new build with money they projected and never collected — and roughly 90% of the ones that underdelivered are increasing their AI budgets anyway.
Now they are going to add a private compute stack on top of it.
The Navistar franchises asked for everything because somebody else was paying.
Your business units are doing the exact same thing with AI right now. Every department wants an agent. Every team wants a model. Every VP wants a pilot with their name on it. And nobody is carrying the cost, because it is buried in a central innovation budget nobody has to defend.
So the wish list grows. And every item on it becomes a system that has to be answered for.
Apply the 15% rule to what you are actually building:
That last line is the Navistar line. Every one of those 800 dealerships that got a system got a support burden. Every business unit that gets an agent gets one too. Nobody is budgeting for it.
And when you fragment your stack — one model here, one vector database there, an orchestration layer stitched across the top, now split across sovereign regions — the integration cost does not add up. It multiplies. IDC predicts that by 2028, 60% of multinational firms will split their AI stacks across sovereign zones, tripling integration costs. I have lived that math already, in a multi-vendor datacenter where IBM, HP, and Sun all pointed at each other while the outage ran long.
Multi-vendor complexity is not best of breed. It is a liability you pay for in slow motion.
So run the honest math. If you build a sovereign AI stack and the annual carry is 15% of the build, you are not funding innovation in year three. You are funding ghosts.
The vendor sold you the truck. Nobody quoted you the maintenance.
This window is not arbitrary. Three things close it.
August 2026: EU AI Act enforcement arrives. Compliance stops being a slide and becomes a deadline. Urgency spikes and buyers stop asking careful questions.
September–October 2026: 2027 capital budgets get locked. Whatever number goes into that budget is the number you live with for the next twelve months.
After October: Every sovereign AI build gets funded at the build number. The carry cost gets discovered in Q3 2027, at which point it is somebody's problem and nobody's decision.
If you put the carry cost on the page before the budget locks, you own the conversation.
If you put it up after, you are the person explaining a variance.
List every AI system in flight — pilots included. Sum the development cost. Multiply by 0.15. That is your annual carry, every year, forever. Put that number on one page. Nothing in this playbook matters more than that single page.
For every business unit running an agent in production, budget the human support behind it. Who answers when the agent is wrong at 4 p.m. on a Friday? That person costs money. Name the role and the cost.
This is the Navistar move, and it is the whole playbook in one line. Charge the AI carry cost back to the business unit that requested the system. Do not run a prioritization workshop. Do not build a scoring matrix. Send the bill.
Once the units are paying, ask exactly what the franchises got asked:
Kill everything that cannot answer in one sentence with a number in it.
Do not move a fragmented stack in-house. You will import the sprawl and pay triple to keep it alive. Cut vendors first. Move second.
Keep your prompts, traces, fine-tunes and evaluation sets on infrastructure you control. Treat the model itself as swappable — because it is, and it gets cheaper every quarter. Sovereignty is not the hardware. It is the learning loop.
You are about to do a sovereignty audit, a carry-cost model, and a vendor consolidation. Every regulated company in your sector has to do the same thing in the next eighteen months, and almost none of them have run the 15% math.
Package it. The audit is a diagnostic. The carry model is a spreadsheet. The consolidation plan is a roadmap. That is a service line.
Most of your competitors will fund the build and discover the carry in eighteen months, when it is a variance report instead of a decision.
You have until October.
I work with four executives at a time as a Fractional CDO. Ninety-day minimum. We run the tally, put the carry number on the page, and get it in front of the people who sign before the 2027 budget locks.
And if you want to be known as the executive who saw the carry cost coming while everyone else was buying compute — that is what M.A.P. (Maverick Advantage Platform) does. It turns what you already know into the authority content that makes the market come to you.
Anyone can show you an AI strategy. Just ask AI.
I show you where the bill is.
M.A.D. (Maverick Advantage Design) Designs Your Brand. M.A.P. (Maverick Advantage Platform) Makes You Known For It.
Stop Reading. Start Seeing.
— Charles K. Davis
Fractional CDO
seriodesignfx.com
P.S. The Navistar franchises did not get disciplined because someone showed them a framework. They got disciplined because someone showed them an invoice. If your business units are still asking for everything, it is because nobody has sent them the bill yet.
