I am old enough to remember a fair amount of the late 70s and early 80s. Full disclosure: I have a pretty good recollection of the 90s and the two and a half decades after that as well. But back then, while I remember plenty of moments, I do not think I understood the bigger geopolitical picture. I was yet to be a teenager, so I will give myself a small pass.
The 1979 fuel crisis was the second major shock to hit airline growth in less than a decade, and it arrived at exactly the wrong time. Air travel was becoming mass transport. Not quite normal, perhaps, but no longer the preserve of the very wealthy either. Then fuel, which airlines had treated for years as a background cost, moved right to the centre of the P&L.
That changed the industry. Manufacturers stopped selling range and speed as the main event and started selling efficiency, reliability, and cost per seat. The Boeing 757 and 767 arrived in the early 80s as leaner twin-engine aircraft that were much cheaper to fly per passenger. Airbus built much of its challenge to Boeing on the same logic, eventually getting to the A320.
It was not a tidy transition. Braniff went. Laker went. Plenty of thin-margin operators went with them. But the airlines that came through the other side were different businesses. By the late 80s the industry was growing again, only this time the growth was built on efficiency rather than glamour.
So, on to AI
I think we are in the early stages of a similar shock now. Tokens were easy to ignore when the bills were small, the usage was experimental, and most people were still treating AI as a clever chat box. That phase is ending.
For any business that depends on AI at scale, token cost is starting to look less like a line buried in a software budget and more like fuel. You can still fly, but you cannot pretend the fuel does not matter.
The big providers will respond in a fairly predictable way, because aviation has already shown the pattern. First come the more efficient aircraft: smaller, faster, cheaper models that do most of the useful work for a fraction of the cost. Not as glamorous as the frontier flagships, perhaps, but much easier to operate at volume.
Then come the service tiers. OpenAI, Anthropic, Microsoft, and the rest will all end up with their version of the same airport. Cheap, no-frills inference for workloads where price matters more than comfort. Premium, governed, supported AI for enterprises that need reliability, controls, auditability, and someone to call when it matters. The same firms will sell you the efficient aircraft and fly you in every class of cabin.
The change that happens because it can
There is another move coming too, and it will happen for the simplest reason: because it can.
A growing share of low-complexity inference will move off cloud and API infrastructure altogether and onto the device in your hand. Not everything. Not the hard enterprise work. Not the work that needs access to large real world understanding. But a lot of everyday questions do not need a round trip to the cloud.
How do I fix a puncture? What is the best way to open a best man's speech? How do I get a stain out of a shirt before a meeting? That sort of thing will increasingly be handled locally, or by very cheap inference that does not need the full enterprise treatment.
In airline terms, that is the low-cost carrier. It is not glamorous. The seats are tight, the boarding queue is a bit of a mess, and nobody is pretending it is a first-class lounge. But most of the time it gets you from A to B, and for a lot of journeys that is exactly what people need.
So the market splits. The price-sensitive workloads get funnelled through the shed that is Terminal 1. The national carriers keep building out the first-class lounges in Terminal 2. No-frills commodity inference one way; premium, governed, enterprise-grade AI the other.
Both terminals will be busier than ever. They just will not be the same business.