A filter I applied to the listed AI universe left three companies standing on my list of ten serious candidates. The filter asks for something any ordinary business would take for granted: revenue paid from customers’ operating budgets, large enough to show up in results.
Everyone talks about AI as an investment. Very few people go and look.
I went, with one condition in hand.
I ran out of candidates.
The Signal
The question I brought was practical. If these tools are producing real economic results, somewhere there are listed companies collecting that money, and it should be possible to find them. So I set a single condition to qualify. The company had to sell something built on AI that a customer pays for out of an operating budget. Money paid because the tool saves money or earns money this quarter. And that revenue had to be large enough to move the company’s results. A normal condition. Any bakery meets it.
The first thing the filter removed was the buildout chain. Chips, energy, cooling, networking, memory. Trace any invoice in that chain and it ends at the capital expenditure budget of a handful of companies building data centers. The revenue is real and enormous. It also comes out of one pocket, and that pocket is a construction budget, approved on faith that the applications will arrive later. Under my condition, the entire chain counts as a single candidate, and a conditional one.
The second thing it removed was the megacaps. Several of them have serious AI revenue. The trouble is arithmetic. When AI tools produce a small percentage of a very large company’s revenue, the size swallows the result. You can believe the story completely and the company still fails the filter, because whatever the tools earn disappears into everything else it sells.
After those two cuts, my list of ten serious candidates held three names.
That ratio is the finding. There is more capital looking for AI exposure than I have seen in my 25 years in this industry. What ran out was listed businesses that already turn these tools into operating revenue a customer willingly pays. The bottleneck sits on the idea side of the market, and the market keeps pricing it as if it sat on the capital side.
Then I looked at what survived. One applies AI to fleet operations. The other two sit in cardiac diagnostics and factory inspection. Boring companies, by the feed’s standards. Each of them sells to an operating budget, and each sits at an intersection between AI and an older industry with its own physics. I did not go looking for convergences. The filter produced them, at the same intersections this newsletter tracks.
I recognized the pattern because I had already lived it at a smaller scale. I built a regenerative farm from scratch using these tools, and the farm taught me the filter before the market did: anything that does not change what happens in the soil stops getting paid by the second season.
An economy pays for applications. It tolerates infrastructure while it waits for them. The listed market currently holds a long shelf of the second and very little of the first, and that imbalance is the most useful thing my search produced.
The Application
Five cases where an operating number moved.
UPS deployed ORION, route optimization software it built itself, completing rollout across its US fleet in 2016. It reports about 100 million miles and 10 million gallons of fuel saved yearly; UPS’s own estimate at full deployment was 300 to 400 million dollars a year. Buyer and seller share one body. It predates the current AI wave; the point is the economics, savings landing in its own fuel budget every year.
Mayo Clinic’s EAGLE trial, a pragmatic cluster-randomized study across 120 primary care teams in 45 sites, screened 22,641 adults. Clinicians using Mayo’s AI-ECG algorithm diagnosed low ejection fraction, a weakened heart pump, in 2.1% of patients versus 1.6% with usual care, a third more new diagnoses. Published in Nature Medicine, May 2021.
CITIC Dicastal, the world’s largest aluminum wheel maker, put AI vision inspection into series production at its Ameur Seflia plant in Morocco, the named quality lever in a plant-wide program that cut defects 31.1%. Validated by the World Economic Forum’s Global Lighthouse Network in January 2025.
Farmers running John Deere’s See & Spray covered more than 5 million acres in the 2025 season with sprayers whose cameras open nozzles only over weeds. Around 31 million gallons of herbicide mix saved, a roughly 50% average cut in non-residual herbicide use, average yield up 2 bushels per acre. Deere-reported, from fleet telemetry.
Insilico Medicine’s Phase 2a trial of rentosertib, a pulmonary fibrosis drug whose target and molecule were both generated with the company’s AI platform: 71 patients, twelve weeks, double-blind. The 60 mg arm gained 98.4 mL of lung capacity; placebo declined 20.3 mL. Published in Nature Medicine, June 2025; a 320-patient Phase III started in July 2026. Small trial; lung function was a secondary endpoint; safety, the primary endpoint, was met.
The Noise
The diversification story. A basket spread across chips, energy, cooling, networking and memory looks like five separate positions. Follow the invoices: every company in that chain is paid from the same capital expenditure budgets of the same handful of data center builders, so the five lines rise and fall on one decision. Diversification means your risks answer to different owners. These five answer to a single pocket, and the pocket belongs to someone else’s board. Five names for the same bet is coverage of nothing.
The Question
Take everything you plan to spend on AI next year, in your company or your own work.
Of that total, what percentage depends on the buildout continuing at its current pace?
Write the number down before you read on.
Now What?
Four things to do with your own budget, in one afternoon.
Split your AI spend into infrastructure and application. Two columns. Compute, tokens, storage and platform fees on one side; tools that change an operating number on the other. Most budgets I have seen never make this cut, which is why they cannot answer the question above.
Trace each provider to the pocket that pays it. If a vendor’s own revenue rides on the buildout capex chain, its pricing and its roadmap move with that chain. So does your dependency on it.
Stress test the application column. Assume the buildout cools next year. Cross out every line whose value disappears with it. Whatever remains is your real AI budget; the rest is exposure you were carrying without naming it.
Demand an operating number from every application line. Miles cut, defects caught, gallons unsprayed, diagnoses found. A tool that cannot name the number it moves is infrastructure wearing an application’s badge.
What I’m Watching
A Medicare payment rail for algorithms. In July 2026, CMS proposed a “Software as a Medical Service” payment category in its 2027 rules, designating 36 billing codes: the first systematic Medicare reimbursement rail for algorithmic diagnostics. Final rules are expected around November 2026. The day diagnostic software has a reliable payer, cases like the Mayo trial above stop being outliers.
Ping An’s earnings line. In its first-half 2026 results, published August 20, the insurer reported RMB 57.3 billion, around 8 billion dollars, in sales made with its AI tools, 81% of customer service volume handled with them, and RMB 7.11 billion saved in claims fraud detection. Company-attributed figures with the methodology undisclosed. Applied AI now carries line items in an official earnings report. That is new.
A clearance template for conversational medical software. UpDoc V1.0 became the first FDA-cleared device with a conversational LLM front end, for insulin titration in type 2 diabetes; cleared December 23, 2025, and identified publicly in June 2026 by regulatory analysts at Innolitics. The cleared function is narrow. The template it establishes is the signal.
Years of AI as the only conversation, and when you finally go looking for where to put money, the shelf offers you the pick and the shovel. The monetizing part exists. I listed five pieces of it above, collected on delivery routes, in primary care clinics, factory lines, sprayer booms and trial sites, by organizations the feed rarely names. The conversation and the cash are sitting in different rooms. Most of the attention is still in the wrong one.
This is the Javier D’Ovidio Newsletter. Emerging tech without the hype. Real signals for strategic decisions.
If this edition helped you see something you were not seeing before, forward it to one leader who needs to be in the conversation.
Subscribe at newsletter.javierdovidio.com so you catch the next edition. Follow Javier D’Ovidio on LinkedIn for daily signals on the convergence.
How I make this
The judgment in this newsletter is mine. Every edition is the output of a framework I have built over 25 years in technology. I map what is happening across five forces. Then I filter signal from noise, and run each thing through one question: how does this serve life.
I use AI as a tool in that process. It helps me sweep sources across the five forces and draft. It does not decide anything. I choose the topic. I verify every figure and claim against primary sources before it goes out. I call what is signal and what is noise, and that call rests on having watched several technology waves arrive and get dismissed. I edit every edition against my own writing standards. The framework and the final call are mine, and I am accountable for every claim here.
That is the point of this newsletter. AI is a tool a person uses. The person stays responsible for what it produces.
Buen día !
Buenísimo Javier;
Recién me suscribí al newsletter y está muy bueno el contenido !
“La conversación y el dinero están en habitaciones diferentes”.
Para mí, ese es el gran titular.
Muy buena publicación.
Saludos,
Nacho