
“Both can be true.”
It’s a saying I find myself repeating often. And yet, keeping two seemingly contradictory ideas in the same headspace is not easy.
For example: building new roads can relieve traffic and end up increasing it in the long run. (Braess’s Paradox, for the uninitiated).
Giving people more choice often makes them less likely to choose anything at all (Paradox of Choice, or as I call it - The Cheesecake Factory Paradox).
You can fiercely love your kids and still regard bedtime as the greatest invention in human history. (Henceforth known as the “I love you, and please become unconscious” Paradox).
A company can build a product that genuinely changes the world and creates an overvalued bubble.
That’s right. We’re talking about the railroad bubble of 1873. No… wait… the internet Dot-Com bubble of 2000. No… wait… You get the idea.
Of course, the contradiction we are referring to today involves AI - the popular debate over whether AI is valuable, or whether there is an AI bubble.
The answer? Yes. Both can be true.
I have been called both an AI evangelist and an AI skeptic. In this case, both are true.
Last year, we published research on how Delta can use AI in revenue management alone to increase revenues by up to 10%. The pushback from the traditional pricing industry was, um, lively to say the least. (For those keeping score: Delta’s Q3 passenger revenues were up 15% year-over-year for many reasons -including a now very secretive AI pricing strategy).
I have also been told that I “simply don’t get it” when speaking to travel executives about how overheated AI adoption has become. Granted, I fully understand the pushback when suggesting the role for which the “Head of AI” was hired isn’t as important as people think. I mean, what else are they going to say?
But that in no way makes either call any less true. This seems an increasingly obvious thing to say, but it’s quite remarkable how often it is overlooked.
For the record, yes, AI is valuable; very valuable. But have we ever asked HOW valuable?
And that’s just the question we ask with our lead chart this week. How does the investment made into AI compare with other famous infrastructure investments over time?
There are some real society-progressing infrastructure investments to compare through history. The Suez and Panama Canals were built for less than $25 billion combined in today’s dollars (though this does not include the human cost, which was very high). The largest project for fusion power? $30 billion for ITER.
The Apollo Program - the very program from which we get the term “Moonshot” - was a whopping $330 billion in 2026 dollars. The entire U.S. interstate system - one of the best infrastructure investments to connect and enable the world’s largest economy - was $500 billion. Half a trillion.
Saudi Arabia’s ill-fated (ill-conceived?) NEOM project was originally slated for $1 trillion USD… before work was almost entirely stopped.
Now consider these historical projects in the context of AI infrastructure and compute spending. $1.1 trillion, based on conservative estimates… just for 2026. Contacts within Nvidia privately put that number closer to $1.4 trillion, while Gartner puts it closer to $1.6 trillion. For comparison, the entirety of China’s Belt and Road initiative is estimated to be $1.3 trillion.
AI is valuable, but is it four moonshots valuable this year alone? 2027 is anticipated to see another $2.4 trillion in AI infrastructure investment. Is AI worth another seven moonshots next year?
Already, we intuitively know the answer, even if we are hesitant to be called a “skeptic” while simultaneously prompting ChatGPT to create another social media image destined to be considered “slop” by half of those who even notice it. AI is valuable… and it’s overrated (not to mention underpriced).
What does this have to do with aviation?
AI poses a threat to commercial aviation. No, we’re not referring to cyber attacks (though we’re not in any position to suggest that’s not a threat). We’re not talking about lost jobs. We are talking about the impact on demand.
Those who have heard me speak at any of the various conferences this year have heard the words, “The tail is wagging the dog.” The U.S. market is acting weird. The stock market usually reacts to economic sentiment. Only not recently. Despite challenging economic data, the stock market continues to go up and to the right. Meanwhile, passenger demand is more closely following the S&P 500 than any economic data.
But that knife cuts both ways. If passengers are willing to pay premium fares based on stock market returns, one would expect those fares to stop being paid during a stock market correction.
It did. Kind of. Remember Liberation Day 2025? That day when Americans were supposed to be freed from, well, we’re not exactly sure. But the stock market responded with a brief near-bear market. Makes sense. But something happened to passenger bookings at the same time. They entered an even deeper bear market.

Looking back to the near-bear market of 2025, many airlines mysteriously withheld revenue guidance during this period. Bookings tanked, along with the stock market - briefly. By May, a delay in the tariffs was announced, which immediately produced a noticeable “V” shape in the stock market. The market returned to highs as if nothing happened, and passenger bookings returned, again, as if nothing happened.
We don’t hear much about this brief panic, but it was a real concern for the airlines… until it wasn’t. Does this mean the market is driving passenger demand? Not necessarily. We don’t know if those passengers were just waiting for things to calm down to book, or if they were looking at dwindling 401k accounts, concerned about them ever recovering. But recover they did, as did passenger bookings.
And yet, correlation is not causation, but would you bet your retirement against it?
Which brings us back to the AI bubble. Is eleven moonshots’ worth of investment a bubble? Probably. Is the circular investment and booked revenue from which growth potential is so liberally being calculated overestimated? Probably.
The point is, the risk of it being a bubble is very high. And if it is a bubble, it will hit the market - that very same market that has shown a tight correlation with all of those sweet, sweet premium fares our industry has gotten so used to.
What then is propping up the stock market that may be propping up premium revenues if not AI?
This all reminds me of a famous nursery rhyme I used to recite to my children when nearing that magic hour of bedtime:
“How many moonshots would an AI fund fund if an AI fund could fund moonshots? It would fund, it would, every moonshot it could—and fund as many moonshots as an AI fund could fund if an AI fund could fund moonshots.”
Eleven. The answer is eleven.
For now.
How Visual Approach uses AI
Oh, we use it. Just because we think it’s overvalued doesn’t mean we don’t hit daily usage limits as fast as a toddler finds uncapped Sharpies.
But we are very intentional about HOW we use it.
AI informs our models, but AI is not our model. We do a lot of coding in the background to build various models and database queries. But our AI output goes into Excel as source data, on top of which we build our models.
Why Excel? Because Excel is how we show our work. It’s human-readable. It’s traceable. It forces transparency. It forces the human element back into the equation. It informs our truly valuable models - the HI models (Human Intelligence).
We don’t design our research to deliver answers. We design our research to deliver understanding. We deliver a competitive edge by doing what AI cannot - thinking differently. Between Gueric, Svetlana, Brett, and yours truly, the context window on our HI models exceeds 85 years.
Today’s AI models are incredible at bringing countless skill sets to competence. It’s how LLMs work. It’s how interpolation works. AI is competent at many things. But our objective isn’t to be competent at everything. Your own LLM can do that. Our objective is to be experts in the commercial aviation market. Our value starts at the AI output.
Which means nothing from Visual Approach ever goes to our clients directly from any AI model - whether in-house or a frontier model. AI only informs HI. You will never see procedural or AI-generated analysis from us. You can do that yourself for $20 / mo, or a quick perusal of LinkedIn will yield plenty. You’ll know it when you see it. It looks like everything else.
Just as importantly, nothing proprietary from our clients ever goes into a public model. Non-public datasets and client work stay strictly inside isolated, zero-retention environments - never used to train someone else’s algorithm.
When AI helps us write code to query schedules, capacity, or financial filings, a human verifies the logic and reconciles the output against primary sources before a single chart is published. We run the queries. Real research, not faith.
Every thesis, forecast, and groan-inducing joke in our research is written—and stood behind—by a human. While AI hallucination is an inevitable side effect of probabilistic engines, our HI hallucinations are usually a result of bad take-out food. Similar, and yet… not.
Visual Approach Research is AI-informed Human intelligence. We’re not regurgitating what your own $20 / mo subscription can give you (which is generating the same as it is for your competitors). Artificial Intelligence creates a brief edge, until your competition thinks to ask the same models the same question.
Human Intelligence creates deep understanding. Deep understanding creates a powerful competitive edge.
And we invite you to inject our Human Intelligence as context into your AI models. We are already working on an MCP server to allow clients to do just that. We are offering our PowerPoint files of full presentations with embedded data, so you can see the nuts and bolts.
Your competitors use the same AI. Our objective is to give yours better Human Intelligence.
In other words: real competitive edge.

You should do a chart on…

We like to create valuable charts. But it’s not easy to come up with new ideas amid the endless hours spent delivering data-driven edge to our customers. In our quest to provide a valuable weekly newsletter, we can keep guessing what you find most valuable, or you could just tell us.
If you have an idea for data visualization, reply to this email and let us know what analysis you’d find most valuable. We’d love to hear from you and will happily name-drop.
ACCESS OUR DATA AND ANALYSIS
We provide bespoke analysis to investors, lessors, and airlines looking for an edge in the market.
Our approach to analysis is data-driven and contrarian, seeking perspectives to lead the market, question consensus, and find emerging trends.
If a whole new approach to analysis could provide value to your organization, let's chat.
If you were forwarded this email, score!
As valuable as it is, don't worry; it's entirely free. If you would like to receive analyses like this regularly, subscribe below.
Then...
You can pay it forward by sending it to your colleagues. They gain valuable insights, and you get credit for finding new ideas!
Win-win!
Contact us
Have a question? Want to showcase your organization in a sponsored analysis? Reach out.
It’s easy. Just reply to this email.
Or, if you prefer the old way of clicking a link, we can help with the hard part: contact

