Kevin Kelly on Personal AI, Eric Seufert on Google and Chatbot Ads, & Why I Don't Save Everything for Later — 3-Bullet Sunday


Hey Reader - Thomas here.

Good day to my fellow compounders!

In 3-Bullet Sunday, I share with you the three most interesting piece of article, video or podcast I came across for the week on investing and growth philosophies.

Enjoy!


Chasing the Northern Lights This December

In early December, I'm going to the Nordics for the first time in my life, to see the Northern Lights.

Apparently the sun runs on a cycle of about 11 years. It peaked in late 2024, but the years right after a peak can still bring big displays, so I'm hoping I'm not too late. I'm heading to Levi in Finnish Lapland and Tromsø in Norway for the lights, plus Stockholm and Helsinki.

If there's anything you highly recommend I do or eat, hit the reply button. When I went to Croatia and Barcelona, you guys gave me wonderful recommendations!

And if there's anyone you think I should meet, let me know as well.

Compound Steadily,

Thomas

P.S. If you've finished The Lunch Break Investor, a short review on Goodreads would mean a lot.


​

I. How Long Will AI Stay This Centralised?

Right now the most reliable way to get a smarter AI is to build a bigger one. In AI lingo, that's the scaling law, and it's why the best models run in data centres with millions of chips and gigawatts of power, costing tens of billions of dollars, which very few companies can afford.

So Kevin Kelly thinks this is probably as centralised as AI will ever get. By his numbers, better model designs and training methods have halved the compute needed for a given capability about every 8 months, and once you add the chip improvements from Moore's Law, the cost per capability halves roughly every 6 to 7 months. He then works out roughly when a model as good as today's best could run on a device you own, and when an individual could afford to train one from scratch.

He also thinks the giant data centres will probably disappear, that they were just temporary scaffolding for the first round of tech. If he's even half right, being able to afford them won't be much of a moat for very long.

The second half is about the AI Kelly wants, one he's trained himself on his life's experiences, his influences, his work and the people he most admires. Today's frontier models are trained on everything for everyone, which in his view makes them generalists aiming at the average. In his words, "We don't want the best AI, we want the ONLY AI."

​Click here to read​

II. Eric Seufert on Google, Chatbot Ads and AI Agents

Search was built to get you off Google as fast as possible. Search did its job if you clicked the first link, and Google made the most money when that link was also an ad. Eric Seufert of Mobile Dev Memo argues AI Overviews flipped that around, so now Google wants to keep you there asking follow-up questions, which means more chances to show you an ad.

AI Overviews now has more than 2.5 billion monthly users and AI Mode more than 1 billion, and Google's search business did $63 billion in Q2, up 17%.

Drew Cohen, the host, isn't fully convinced. Chatbots cost more to run, and ads inside an answer could make people trust it less. Seufert calls that a design issue and points out that ChatGPT's ads were on pace for $1 billion a year less than 200 days after launch. Drew's bigger question is whether Google can go from one of the best business models ever built to something just as good.

They also get into AI agents, online travel and how Apple could make money from AI. In my Booking report for paid Insider Stocks members, I said how Meta charges for Muse, its AI agent, matters more to Booking than anything Muse can do today, since a fee on every booking would hurt and an auction is the game Booking already plays well. If you want to hear what Seufert thinks, skip to about the 36-minute mark. The travel part starts around 50 minutes.

​Click here to watch​

III. Why I Don't Save Everything for Later

Chuang asked me on the Do More podcast about memory dividends, an idea from Bill Perkins' Die With Zero that I love. A lot of us think about passive income only in money. But at the end of our lives it's the memories and experiences we look back on with the people we love, and our time is finite no matter what.

It's also what the final chapter of my book is about, which happens to be my editor's favourite. Investors like me think a lot about delayed gratification. But we don't get to choose how long our health holds up, or how much time we get with the people we love, so some things can't wait until we decide we're ready.

Chuang had his own version. He owns a Malaysian REIT that pays him dividends four times a year, but he can recall a trip to Vietnam with his kids any day he likes.

We also get into why financial freedom to me is the privilege to do what I love, why I start with businesses I can't live without, and how much cash I keep.

​Click here to watch​

That's all I have for you today!

Till next week.

Thomas

Founder of Steady Compounding

Connect with me:

​

Steady Compounding

I write about investment concepts, business breakdowns and timeless lessons from super investors. Featured on Business Times, Channel News Asia (CNA) and more. Read by over 10,000 investors.

Read more from Steady Compounding

Hey Reader - Thomas here. Good day to my fellow compounders! In 3-Bullet Sunday, I share with you the three most interesting piece of article, video or podcast I came across for the week on investing and growth philosophies. Enjoy! Chasing the Northern Lights This December In early December, I'm going to the Nordics for the first time in my life, to see the Northern Lights. Apparently the sun runs on a cycle of about 11 years. It peaked in late 2024, but the years right after a peak can still...

Hey Reader - Thomas here. Good day to my fellow compounders! In 3-Bullet Sunday, I share with you the three most interesting piece of article, video or podcast I came across for the week on investing and growth philosophies. Enjoy! Chasing the Northern Lights This December In early December, I'm going to the Nordics for the first time in my life, to see the Northern Lights. Apparently the sun runs on a cycle of about 11 years. It peaked in late 2024, but the years right after a peak can still...

Alphabet Q2 2026: The $707 Billion Footnote This quarter is the clearest sign yet that Alphabet is transforming into a different company and there’re a lot of misconceptions surrounding its Q2 2026… so let me go through them one by one. Advertising has gone from 78% of revenue in Q3 2023 to 68% in Q2 2026, with Google Cloud coming in strong to change the revenue mix. And it’s not because search has fallen or anything like that, they weren’t disrupted by AI, in fact, the opposite has happened,...