I sat in on an operator panel earlier this year where one of the speakers said something that has stuck with me since.
The companies winning with AI, he said, are the ones who almost never talk about it. They treat it the way they treat electricity. It runs everything. It is in the walls. And nobody calls a board meeting to announce that the lights are on.
The companies losing with AI are the ones who made it their personality. More AI-strategy slides than AI-results slides. A chief AI officer before a clean customer master. A press release before a working pipeline.
He was paraphrasing a feeling that anyone who has sat through enough portfolio reviews this year recognizes immediately. AI has become a thing companies perform rather than a thing they run.
The two postures
There are two ways a portfolio company can hold AI right now.
The first is as identity. AI becomes part of how the company describes itself. The deck leads with it. The all-hands talks about it. The website gets a new banner. There is a roadmap, a center of excellence, maybe a new hire with “AI” in the title. The signal to investors and the market is “we are an AI company now.”
The second is as infrastructure. AI becomes part of how the company runs. It sits inside the forecasting, the pricing, the support queue, the collections process, the document review. Nobody outside the company would know it is there. The signal to investors is a number that moved.
The first posture is loud and produces decks. The second is quiet and produces results. The operators worth listening to have figured out that the second one is where the value lives, and that the first one is usually a tell that the second one is not happening.
Electricity and networking did not get a strategy slide
Think about how the strongest companies you know treat the genuinely critical technologies that came before this one.
No serious operator has an electricity strategy. They have a power bill, a redundancy plan, and an expectation that the lights stay on. Electricity is critical infrastructure. It is deployed aggressively, it is engineered carefully, and it is discussed almost never. The conversation is about what the power enables, not the power itself.
The same is true for networking, for cloud, for the database. These are load-bearing technologies that the entire business depends on. They get budget, they get attention from the people who run them, and they get almost no airtime in the board deck. They are judged on uptime and on what they let the business do, not on their own existence.
AI is heading to exactly the same place. In five years, “we use AI” will be as unremarkable a statement as “we use the internet.” The companies that understand this already are quietly wiring it into the operation. The companies that do not are still holding it up like a trophy.
If your board deck has more slides about your AI strategy than about what AI has actually changed in the numbers, you have it backwards. You are treating infrastructure as identity. That is the same mistake as commissioning a press release about your new electrical panel.
Why identity-first AI fails on contact
The identity posture does not just waste airtime. It actively delays the work that would make AI pay off, and it does so in a way that looks like progress.
Here is the pattern I see most often. The pressure to “do AI” lands on a management team. The fastest way to respond to that pressure is something visible. A demo. A pilot. A title. A line in the next deck. So the company optimizes for the visible thing, and the visible thing gets reported as momentum.
Meanwhile the actual prerequisite, the boring one, goes untouched. The model gets pointed at the company’s real data and the real data has duplicates, mismatched hierarchies, revenue figures that do not tie to finance, and customer records in the CRM that do not match the billing system. The demo that ran beautifully on clean sample data produces useless output on the operation it was supposed to improve.
This is the stall I see across portfolio AI initiatives. The model works. The data underneath it was never ready. And because the company committed to AI as identity rather than as infrastructure, the response to the failure is usually to protect the narrative rather than fix the foundation.
So the pilot gets extended. The scope gets narrowed to another clean subset of data. The board deck still says “we have AI.” What the company actually has is a demo that runs on a spreadsheet. I have written before about how the proof of concept becomes a permanent stall precisely because nobody wants to admit the foundation is not there. Identity-first AI makes that admission even harder, because the whole point of the posture was to look further along than you are.
The prerequisite nobody wants to fund
Treating AI as infrastructure has a precondition, and it is the part most companies skip.
Infrastructure has to be able to carry load. You do not run a factory off domestic wiring. You do not run real-time decisioning off a database that falls over at month end. And you cannot run AI off a data estate that cannot produce a single trusted version of its own numbers.
The reason the electricity analogy is useful is that it cuts both ways. Electricity is quiet and reliable because someone did the unglamorous engineering. The grid, the wiring, the redundancy. AI only gets to be quiet and reliable the same way, by someone doing the unglamorous work underneath it first.
That work is data readiness. A customer master that one team owns. Revenue definitions that finance, sales, and operations all sign. Product hierarchies that are consistent across systems. History that was not silently rewritten three times. None of it is exciting. All of it is the load-bearing wall the AI sits on.
This is why AI readiness starts with data readiness and not the other way around. The companies treating AI as identity are buying the appliance before they have wired the building. The companies treating it as infrastructure spent the first quarter on the wiring, which is why their AI actually runs when they switch it on. If you want a fast read on whether your data foundation can carry the load, the AI readiness assessment is built to surface exactly that gap before you spend a cent on the appliance.
How to tell which posture you are in
You do not need a long audit to know whether your portfolio company is treating AI as identity or as infrastructure. A few questions sort it quickly.
Count the slides. In the last board deck, how many slides describe the AI strategy and how many describe a result AI produced? If strategy outnumbers results, you are performing AI, not running it.
Name the number. Ask what specific metric AI has moved. Margin on a product line. Hours out of the support queue. Recovery rate on collections. Forecast accuracy. If the honest answer is a roadmap rather than a number, the AI is still identity.
Check the data first. Ask whether the AI initiatives run on the company’s real operational data or on a curated sample. If everything that works runs on clean subsets, you have validated demos, not deployed infrastructure.
Find the owner. Ask who is accountable for the result, not the technology. Infrastructure has an operator who is judged on what it delivers. Identity has a champion who is judged on visibility. The two attract very different people and produce very different outcomes.
If the answers point toward identity, the fix is not less ambition. It is redirecting the ambition. Take the energy currently going into how the company talks about AI and point it at the data foundation that would let the company actually use AI. The deck will get quieter. The numbers will get louder. That is the trade you want.
Put it in the plumbing
The instinct to make AI your identity is understandable. Everyone is talking about it, the pressure to be seen doing it is real, and a strategy slide is a lot faster to produce than a working pipeline on clean data.
But the operators who are actually pulling ahead have made a quieter bet. They are wiring AI into the operation and judging it the way they judge every other piece of critical infrastructure, on what it enables and what it moves. They are not announcing it. They are using it.
AI belongs in the plumbing. It earns its place by results, not by airtime. And the prerequisite for putting it there is the same prerequisite for every other system the business depends on, a data foundation solid enough to carry the load.
If your AI conversation is louder than your AI results, that is the signal. Stop polishing the identity. Go fix the foundation, and let the AI go quiet the way every critical technology eventually does.