A four-year conviction · Oct 2022 – Oct 2026

I bet on the machines
that learn.

Before the world met ChatGPT, I started building a portfolio across the entire AI stack — the chips, the data centres, the power that feeds them, the clouds that rent them and the software that brings them to life. Four years on, it has grown by 600%.

Read the journeySee the numbers
+600%portfolio growth, Oct 2022 → Oct 2026
0%
Total portfolio growth
Every $1 invested became $7
0%
Compound annual return
≈ 62.7% p.a. over four years
0
Layers of the AI stack
From silicon to software
0 yrs
Holding through cycles
Including two sharp drawdowns

The thesis

The thesis: AI is the new electricity

In 2022 most people saw artificial intelligence as a research curiosity. I saw the start of an industrial build-out on the scale of railways, electrification and the internet — and every build-out has its picks and shovels.

Rather than guessing which chatbot would win, I asked a simpler question: whoever wins, what will they need to buy? The answer was the same every time — more compute, more data-centre capacity, more power, more networking and more software to turn it all into productivity.

I invested across every layer of the AI stack:

◆
06

Applications

Copilots, agents and AI-native software that businesses pay for every month.

◇
05

Models & platforms

The foundation models and developer platforms that everything else is built on.

☁
04

Cloud & hyperscalers

The landlords of compute, renting AI capacity to the world by the hour.

▣
03

Data centres & networking

Real estate, cooling, optics and switching — the physical plumbing of AI.

ϟ
02

Power & energy

Every new AI campus needs hundreds of megawatts. Grid, generation and storage.

⬢
01

Semiconductors

GPUs, custom accelerators, memory and the equipment that makes them.

The journey

Four years, five chapters, one conviction.

Chapter 1 · 2022

Planting the seed

Buying conviction when others were selling fear

Late 2022 was a brutal market. Tech stocks had fallen hard, rates were climbing and semiconductor shares were priced for a recession. That was exactly when I began. I built core positions in the chipmakers and equipment suppliers that every AI future depends on — patiently, in tranches, while sentiment was at its worst.

  • Initiated core semiconductor and chip-equipment positions
  • Bought in tranches through the bear market rather than all at once
  • Set a four-year horizon and wrote the thesis down

Chapter 2 · 2023

The ChatGPT moment

The world woke up — and the bottleneck was compute

Within weeks of ChatGPT’s launch, every board in the world had an AI strategy and nobody had enough GPUs. Accelerator demand outstripped supply, and the companies selling the shovels reported results that rewrote the record books. I added to the compute layer and broadened into the cloud platforms racing to rent that capacity.

  • Added to AI accelerator and high-bandwidth memory exposure
  • Opened positions in the major cloud and hyperscale platforms
  • Resisted chasing speculative ‘AI-washed’ small caps

Chapter 3 · 2024

Building the factories

Following the capex into concrete, copper and cooling

AI moved from software to steel. Hyperscalers lifted capital spending to unprecedented levels, and the money flowed into data centres, optical networking, liquid cooling and electrical gear. I followed the capex — rotating part of my chip gains into the companies building and supplying the AI factories themselves.

  • Took partial profits on the strongest chip positions
  • Rotated into data-centre REITs, networking and cooling suppliers
  • Kept position sizes disciplined as valuations stretched

Chapter 4 · 2025

Power becomes the bottleneck

Holding steady through the shakeouts

Early 2025 tested everyone. Cheaper, more efficient models sparked a sharp sell-off in AI stocks, and trade tensions shook markets again soon after. I held my core and added selectively — because cheaper intelligence means more usage, not less. Meanwhile the real constraint had shifted to electricity, so I built a dedicated energy and grid sleeve.

  • Bought the dips in high-conviction holdings during the sell-offs
  • Built an energy sleeve: generation, grid equipment and storage
  • Added enterprise AI software as agents reached real customers

Chapter 5 · 2026

The inference era

From training models to running them everywhere

AI is now part of daily work. The spending has shifted from training giant models to running them — billions of queries a day across phones, factories, vehicles and robots. I rebalanced to lock in gains, trimmed what had become crowded, and leaned into the next frontier: inference at scale, edge devices and physical AI.

  • Rebalanced to bank gains and reset risk
  • Added exposure to inference, edge AI and robotics
  • Portfolio up 600% from the first purchase in October 2022

Performance

From 100 to 700 — through every dip.

Portfolio value indexed to 100 at the first purchase in October 2022. The path was never a straight line — the drawdowns were where the biggest gains were seeded.

Portfolio value (indexed)

Hover or tap the chart
Portfolio value indexed to 100 at October 2022, rising to 700 by October 2026.
01002003004005006007008002022202320242025Oct 2026700 · +600%
View data table
QuarterIndexNote
Q4 2022100First positions
Q1 2023118
Q2 2023165AI demand shock
Q3 2023158
Q4 2023196
Q1 2024262
Q2 2024318
Q3 2024296Mid-year pullback
Q4 2024352
Q1 2025318Efficiency & tariff sell-offs
Q2 2025405
Q3 2025482
Q4 2025535
Q1 2026568
Q2 2026612
Q3 2026668
Oct 2026700+600%

Where it sits today

Allocation by theme, % of portfolio

  • Semiconductors & compute30%
  • Data centres & networking20%
  • Power & energy for AI15%
  • Cloud & hyperscalers15%
  • AI software & applications12%
  • Robotics & edge AI8%

How I invest

Six principles behind the 600%.

01

Own the picks and shovels

I don’t need to know which AI app wins. I own what every winner has to buy.

02

Follow the capex

Where the biggest companies on earth spend their capital tells you where the next profits will land.

03

Buy fear, trim euphoria

My best entries came during sell-offs. My best decisions were taking profits when it felt uncomfortable.

04

Size by conviction, cap by risk

High-conviction positions get more capital — but no single holding is allowed to sink the ship.

05

Think in years, not days

A four-year horizon let me ignore noise that shook out short-term traders.

06

Rebalance with discipline

Regular rebalancing turned paper gains into real ones and kept the portfolio aligned with the thesis.

Outlook

What’s next: the next four years

The first chapter of AI was about building capacity. The next is about using it — everywhere. These are the themes I’m watching closely.

Inference everywhere

Billions of daily AI queries need efficient chips and distributed data centres.

Physical AI & robotics

Models that see, move and manipulate are leaving the lab for factories and warehouses.

Energy for intelligence

Generation, grid and storage remain the hardest constraint on AI growth.

Sovereign & Australian AI

Nations — including Australia — are building local compute, creating a homegrown opportunity.

Let’s talk about the future.

I share what I’m learning about the AI economy with fellow investors, founders and builders. If you’d like to compare notes, I’d love to hear from you.

hello@myaionline.com.auLinkedIn