101-I Built an Index: newtype AI Index

Yesterday, I built an index for investing in the AI industry chain, called the newtype AI Index. The repository is here: https://github.com/newtype-01/newtype-ai-index

The newtype AI Index is a net asset value curve. It starts on July 17, 2026, with an initial value of 100. It contains seven companies: Nvidia, Micron, Vertiv, Constellation Energy, TSMC, Marvell, and Nebius.

These seven were selected according to the framework of the “seven layers of the AI industry chain” I introduced in my course — from the bottom layer of power, to chips, to memory, to data center electromechanical systems, to cloud infrastructure, one layer at a time, selecting the key company in each layer and then assigning weights according to importance. Altogether they add up to 100%, with no cash, meaning it is pure AI exposure.

Every trading day, this curve is updated once. You can see whether it performs better or worse than the semiconductor ETF (SOXX), or better or worse than the Nasdaq 100 (QQQ).

Only by outperforming SOXX does my framework prove useful — because outperforming SOXX means “actively picking stocks with the seven-layer framework + actively assigning weights” is better than “blindly buying the entire semiconductor ETF.” If it can’t outperform, then it means it hasn’t generated excess return over this period, and I should reflect on that.

The first month’s performance was underwhelming: from July 17 to July 31, the newtype AI Index fell 3.51%, SOXX fell 3.24%, and QQQ fell only 1.06%.

Why do this?

The first motivation is that I want to turn “AI industry chain investing” from “telling a story” into a judgment that can be tested.

This year, I’ve created the “AI Industry Chain Investing: The Big Picture” course (already live, free within the community), built a low-frequency signal system (the methodology has been made public), and published monthly reports (released only within the community). But all of these things share one common problem — they are all text.

Text can be polished, explained, and retold after the fact. Readers cannot tell at a glance whether “this thing actually works.”

This index is an honesty test I set for myself.

It distills all the judgments — which companies to pick, what weights to assign, when to rebalance — into a single curve. The curve does not lie. Whether the methodology is useful or not, if you stretch out the timeline, the curve will naturally give you the answer. And because the entire repository is public, Git timestamps are verifiable, and component changes must be committed before the effective date, everything is traceable.

Writing content and selling courses provide feedback that is too soft — a reader liking a post or buying a course does not mean your judgment was correct. But an Index is different. Every day, it grades you with real money and real prices. If it underperforms, it underperforms.

The second motivation is simpler. A lot of people have asked me, “How would you allocate it yourself?” I couldn’t explain it clearly before, because the methodology did not have a unified output surface. Now it does.

The newtype AI Index is the distilled version of this methodology. Anyone can open the repository and see the current judgment: which companies are in it, what the weights are, and why those weights are what they are.

It is not my live portfolio (my live portfolio does not exactly match the index), but it is the “target form” that my methodology wants to show you.

The third motivation is that I want to provide a reference template for people in this industry who do similar work: if you also do industry research, can you put your judgments in a place like this, where timestamps are verifiable?

Our industry has far too many “armchair quarterbacks after the fact” and “selective track records,” and far too few “publicly verifiable continuous records.”

So, the newtype AI Index is the combination of three things:

A commitment to honesty, a condensed output surface for a methodology, and a demonstration.

It is not a product. It does not accept AUM, and it does not manage money on behalf of others. It is just a curve, along with all the judgments behind the curve that can be tested.