Real Assets and Smarter Funds: A Diversification Primer

2026-05-27

The inference cost curve is moving 4x faster than the training one — and that breaks most AI business plans

Frontier-model training got the headline numbers in 2024 and 2025. Inference, the part that actually meets revenue, was the line item nobody wanted to chart. That has flipped. Between April 2024 and April 2026, the per-million-token cost of running GPT-4-class inference fell roughly 92%, on a curve compounding twice as fast as the training-cost curve fell over the same period. This column is about what that re-prices, who it benefits, and why the API margin assumption inside most 2025-vintage AI business plans is about to fail.

2026-05-26 CapexArmsRacePicksAndShovelsAggregatorTheoryStrategy

The Nvidia margin paradox: why FY27 datacentre profit may peak before revenue does

Nvidia just posted 85% datacentre growth, but a quieter line in the same release flagged the structural reason gross margin is set to compress through FY27 — and it isn't competition. It's the Hopper-to-Blackwell transition, hyperscaler ASIC orders, and CoWoS-L allocation. A look at the three numbers that will decide whether Nvidia's profit pool widens or narrows from here.

2026-05-25 CapexArmsRacePicksAndShovelsSemiconductorsStrategy

ETFs, Mutual Funds and Index Funds Compared

2026-05-22

Chart Patterns and Technical Analysis: What the Charts Say

2026-05-22

Cloud Spending Trends in 2026 Every Developer Should Track

2026-05-21

AI Layoffs and the Reskilling Imperative: A Practical Guide

2026-05-21

The warehouse-robot economics nobody is modelling: why swarm coordination — not arm dexterity — decides the next decade

Symbotic, AutoStore, Geek+, and the new wave of Amazon Sequoia deployments are routinely framed as a robotics-hardware story. The hardware is the cheap part. The expensive, defensible, and woefully under-modelled part is the coordination layer — the multi-agent scheduler that turns 4,000 bots into one productive system rather than 4,000 expensive collisions. A look at the gross-margin gap between hardware-led and coordination-led deployments, and why the latter is set to dominate.

2026-05-19 EmergentCoordinationSwarmIntelligencePicksAndShovelsStrategy

The hyperscaler capex cycle has crossed $600bn — here is the under-discussed line item that decides whether it pays back

Microsoft, Alphabet, Amazon, and Meta will jointly commit roughly $612bn of AI-related capex in calendar 2026 — more than the entire US federal R&D budget. The bullish framing is that this is a generational productivity bet. The bearish framing is that it is a bubble. Both miss the question that actually decides the IRR: utilisation. A look at the four-quarter trend in inference-utilisation rates, why the curve flattened in Q1, and what it means for the 2027 capex guides.

2026-05-12 CapexArmsRaceAggregatorTheoryPicksAndShovelsStrategy

Value Investing in an AI-Dominated Market

2026-05-05

1234