One idea ran through the year: when AI demand hits a physical bottleneck, the bottleneck gets repriced. The best trades followed it — memory first, then custom silicon, then the chips around them.
03 / SPEED
Fast when right.
All 63 profitable round trips of the year, by holding period and return. Hover a square to see the trade.
04 / RIDING THE CYCLE
Staying with the memory cycle.
Price change from the first purchase of the year to the best exit achieved. The theme was harvested by returning to the same names again and again, not by holding once.
Measures the stock's move between two of our own executions, not a single-trade return.
05 / CLOSING BOOK
Where the year ended.
Holdings at 30 June 2026 by market value. The whole book sat in the memory and custom-silicon chain the year had been trading.
06 / NEXT RESEARCH UNIVERSE · SEP 2026
Ten ideas for the next orbit.
Ten US-listed companies across AI, semiconductors, power, space, biomedicine and web3, with the question that matters next for each.
Astra's ten
Research watchlist dated 26 September 2026. Order is editorial, not a target weight or ranking of expected returns.
07 / A SECOND OPINION
Ten more, from Claude.
Chosen independently of Astra's list. Three names made both — they carry a ✦. Hover or tap one.
Claude's ten
Claude's list, 28 September 2026, based on public information available to Claude up to mid-2026. Research prompts, not recommendations; order is editorial.
Method and sources
Source: Stake financial-year report pack, 1 July 2025 – 30 June 2026 (US equities; no Australian equities were traded). Return on invested capital = (closing value − opening value − deposits) ÷ (opening value + deposits), in US dollars. Trade returns use first-in, first-out matching and include brokerage; positions held before 1 July 2025 are excluded from trade-level figures.
Past results do not predict future returns, and nothing here is personal investment advice.
Back to the beginning ↑
…and the next season is already falling into place.