The Best Thing About AI?

I’ve been thinking about AI all wrong.

For the last couple of years, I’ve mostly looked at it through the lens of hype.

The trillion-dollar data centres. The breathless founder interviews. The nonsense LinkedIn posts. The Sam Altman prophecies. The endless parade of startups claiming they’re about to “reimagine” email, meetings, spreadsheets, therapy, education, customer service, romance, death, and probably toast.

And yes, a lot of it is absurd. A lot of it is overfunded nonsense. A lot of it will disappear.

Good. Because I’m starting to think that might be the best part.

What if AI really is transformational, but not in the way investors think?

What if it changes everything, improves almost everything, spreads everywhere, makes our lives better, gives ordinary people extraordinary tools...

And still doesn’t make any one company particularly rich?

That would be quite funny, wouldn’t it?

Silicon Valley spends trillions building the most powerful technology in human history, and the final winner is Dave from Swindon who can now build an app, write a legal letter, translate a contract, learn Spanish, fix his boiler, create a business plan, and explain his blood test results before lunch.

How about that for a good luck story?

The Wrong Question

Most of the AI conversation is obsessed with the same question:

Who wins? Will it be OpenAI? Google? Anthropic? Meta? Nvidia? Microsoft? Some startup run by a 24-year-old in a black hoodie who says “agentic workflows” too often?

Everyone wants to know which company captures the value. That’s the phrase, isn’t it? “Captures the value.” It sounds so clean. So clever. So financially serious.

But maybe that’s the wrong question.

Maybe AI is one of those technologies where the value doesn’t get captured neatly by a company. Maybe it leaks. Maybe it spills. Maybe it spreads too quickly, too widely, and too cheaply for anyone to own it properly.

Maybe the big winners from AI won’t be shareholders. Maybe the big winners will be us.

The users.

The people pressing buttons, asking questions, building things, learning faster, working better, saving time, saving money, and upgrading our lives while investors fight over whose margins are least terrible.

That’s the theory I can’t shake.

AI might be incredible for humanity and but a lousy business model.

The Vaccine Problem

Vaccines are one of humanity’s greatest inventions.

They have saved hundreds of millions of lives. They changed childhood. They changed public health. They changed the basic risk profile of being alive.

They are, without exaggeration, a miracle.

And yet, as an industry, vaccines are not where pharmaceutical companies make their giant fortunes.

Not compared with chronic medications. Not compared with oncology. Not compared with drugs people take every day for years.

A vaccine is too effective. That’s the problem.

You get it, it works, and then you don’t need another one for ages. Terrible business model. Wonderful human outcome.

Imagine pitching that to investors:

“We’ve invented a product that prevents enormous suffering, protects society, and only needs to be used once in a while.”

“Great. How often does the customer come back?”

“Well…every 10 years or so?”

Exactly.

Vaccines create vast social value. They don’t necessarily create vast shareholder value. That matters.

Because we often assume that if something is important, someone must make a fortune from it.

But history says otherwise.

Some things are too useful, too essential, or too widely distributed to become easy monopolies.

AI might be one of them.

The Airline Problem

Then there’s aviation.

Flying is ridiculous when you stop to think about it.

You sit in a metal tube, drink a bad coffee, complain about legroom, and six hours later you’re on another continent.

It is one of the greatest technological achievements of the modern world.

It made the planet smaller. It connected families, businesses, cultures, and economies. It turned what used to be a life-defining journey into something you do for a stag weekend.

And yet airlines?

Awful businesses.

They go bust constantly. They merge. They get bailed out. They fight over tiny margins while being crushed by fuel prices, regulation, labour costs, maintenance costs, airport fees, weather, geopolitics, pandemics, and customers who somehow expect to fly to Barcelona for £29.

The experience changed humanity. The industry barely makes money. Again, there’s the pattern. Huge value to users. Very little value to companies.

We love flying. We just don’t love paying what it actually costs. And because everyone offers roughly the same thing, the competition becomes brutal.

That might be where AI is heading.

A miraculous capability that becomes expected, cheap, and everywhere.

Amazing for us.

A nightmare for anyone trying to defend a premium price.

The PC Problem

Remember PC manufacturers?

Gateway. Compaq. Dell. IBM. Hewlett-Packard.

For a while, they were the future. They were putting computers on every desk and in every home. They were changing work, education, design, entertainment, administration, everything.

And then what happened? The product became a commodity.

A box. A screen. A keyboard. A processor. A slightly different logo.

Margins collapsed. Differentiation disappeared. Consumers compared specs and prices. Nobody cared that much who made the machine, as long as it worked.

The personal computer transformed the world. Most PC companies did not become world-owning profit machines.

Apple is the exception people always reach for, but even there, the story isn’t really “PC manufacturer gets rich.” Apple became Apple because of the iPhone, the App Store, services, brand, ecosystem, taste, and a level of consumer devotion that borders on religious behaviour.

The beige box was never the goldmine. I mean, yes, the value was real. But the profits moved somewhere else.

So maybe AI follows the same path? Maybe AI becomes a beige box? Powerful, useful, necessary, but totally interchangeable.

At the moment, I use both Claude Code and ChatGPT. I don't care which one. I switch seamlessly. They both cost me £200 a month!

But once something becomes interchangeable, the money gets harder.

Why AI Might Defeat AI

Here’s the weirdest part.

AI might be the first industry where the product destroys the moat.

Normally, if a company builds a clever feature, it has a little window of advantage. Maybe months. Maybe years. Competitors have to notice it, understand it, build it, test it, ship it, market it.

That lag creates value.

But with AI, the lag shrinks.

If one company launches a brilliant new feature, everyone else can ask AI to analyse it, replicate the workflow, write the code, improve the interface, draft the launch copy, and deploy something similar at terrifying speed.

AI helps competitors copy AI. That’s hilarious and brutal.

You build an advantage, and the very technology you’re selling helps everyone else erase it.

So where does the moat come from?

Data? Maybe.
Distribution? Yes, to a point.
Brand? Possibly.
Regulation? Sadly, probably.

But the pure technology advantage might be thinner than people think.

Every breakthrough becomes training material. Every new interface becomes inspiration. Every clever workflow becomes a template. Every expensive feature becomes tomorrow’s free plugin.

The better AI gets, the harder it may become for any single AI company to stay special.

That’s not a bug.

That’s a gift.

Open Source Is the Wrecking Ball

Then there’s open source. That's what "local models" are called. I'm trying some of these local models from Google and from OpenAI. They run on my laptop (ie no internet needed. The model is loaded onto my laptop). They're free. And they aren't bad at all.

This is the bit that should terrify investors and delight everyone else.

Because while some companies are spending staggering amounts of money building closed models, open-source models (i.e free models) keep getting better.

They may not always be the absolute best. They may not win every benchmark. They may not have the slickest interface or the fanciest safety team or the most dramatic launch video.

But they don’t need to be the best.

They just need to be good enough.

Good enough to run locally.

Good enough for businesses to customise.

Good enough for developers to build on.

Good enough for schools, charities, researchers, small companies, hobbyists, and governments to use without paying a toll to one dominant platform.

That changes the game.

Because once a free or cheap version is good enough, the premium version has to work very hard to justify itself.

Most people don’t need the world’s best AI.

They need an AI that writes the email, explains the spreadsheet, helps with the presentation, summarises the meeting, drafts the policy, translates the note, and tells them why the washing machine is making that noise.

If the free version does that well enough, the paid version has a problem.

But the paid version also has another problem.

It costs a fortune to serve.

The Cost to Serve Is Not a Detail

Software used to be beautiful because the marginal cost was close to zero.

Build once. Sell many times. Lovely.

AI is different.

Every question costs money. Every image costs money. Every video costs more money. Every agent running around your inbox costs money.

The product doesn’t just sit there.

It works.

It consumes "compute". It uses energy. It needs chips, servers, cooling, engineers, safety teams, lawyers, and enough electricity to make a small country nervous.

That makes the business model awkward.

Users want AI to be cheap or free because software trained us to expect that.

Companies need AI to be expensive because serving it is expensive.

That gap is what will (in my opinion) cause OpenAI and Anthropic to go bust sooner rather than later. They are not charging us what it costs them. They are subsidising millions (maybe billions) of people. What happens when they decide they can't do that anymore?

Of course, costs will come down. They always do. Models will get smaller, faster, more efficient. Hardware will improve. Inference will get cheaper.

But when costs come down for one company, they come down for everyone.

That helps users.

It doesn’t necessarily help margins.

Again, we win.

Maybe This Is the Point

I find this strangely comforting.

Because so much of the AI conversation feels like it belongs to billionaires, CEOs, venture capitalists, and people who say “compute” as if they’re discussing theology.

But maybe the real story is much simpler.

Maybe AI becomes a public utility in all but name.

Not owned by the state. Not centrally planned. Not perfectly equal. But so widely available, so heavily competed over, so hard to monopolise, and so useful that most of the benefit flows outward.

To students. To small businesses. To parents. To teachers. To doctors. To writers. To builders. To lonely people trying to understand a letter from the council. To someone with a good idea but no technical team. To someone with a diagnosis they don’t understand. To someone who wants to learn, create, fix, research, write, ask, test, build, and finally have access to the kind of support that used to be expensive.

That’s the story I want to believe in.

Not AI as god.

Not AI as doom.

Not AI as a trillion-dollar land grab.

AI as leverage for ordinary people.

The Short Trade

Now, does that mean I’m short AI stocks?

Emotionally, yes.

Financially, I’m not giving investment advice because I’m not insane and I don’t want angry emails from people called Nigel who bought Nvidia at the top.

But as a theory?

It’s tempting. Maybe the right position is. Long humanity. And short the assumption that any single AI company captures the upside.

Because those are not the same trades.

AI can be world-changing and still disappointing for investors. It can transform education, work, creativity, healthcare, research, and daily life, while destroying the margins of the companies that provide it.

That sounds contradictory only if you think capitalism and human progress always move together.

They don’t.

Sometimes the best technologies escape. They leak into the world. And become tools, infrastructure, habits, expectations. Nobody gets to own the whole thing.

And thank God for that.

So What Should We Do?

If this theory is right, then the smart move for most of us isn’t to obsess over which company wins.

It’s to use the tools.

Properly.

Not as toys. Not as shortcuts. Not as little cheating machines for writing better emails.

As leverage.

Learn with them. Build with them. Experiment with them. Let them make you more capable. Let them remove friction. Let them compress the distance between idea and execution.

Because if the companies can’t capture all the value, then the value is sitting there for us. Right now. That’s the opportunity. Not to worship AI or to fear it.

And yes, don't buy every stock with “AI” in the investor deck. At least I wouldn't! Just to pick up the tool and use it better than the person next to you.

Takeaways for the Week Ahead

Use AI for one thing you normally avoid.
Admin, research, planning, writing, learning, analysis. Don’t use it where you’re already strong. Use it where you’re stuck.

Stop asking which AI company wins.
Ask how you win. What can you now do faster, cheaper, or better than you could six months ago?

Watch for commoditisation.
If five tools all do the same thing, don’t get emotionally attached. Move to what works.

Remember the real prize.
The miracle isn’t that one company might become worth $5 trillion. The miracle is that millions of ordinary people may suddenly get access to intelligence, creativity, tutoring, translation, coding, research, and support at almost no cost.

That’s the good luck story. AI might not make the companies rich.

It might make the rest of us more powerful. That's my hope.

And that’s a much better ending.

Until next time,

Jon "Free Chatbots for All" Kaye

Choosing to live more deliberately

P.S.

I've built two apps now using AI. From scratch. Without any coding knowledge. I think they're pretty good. If you want to see what George Orwell predicted about AI over 75 years ago, then check out one of those apps. It's called Bookmind.