Thursday, October 1, 2026

Back to today

From X

@convequity 3 clicks

Convequity on X

Renting AI capacity protects Microsoft ($MSFT) if the AI buildout busts. It leaves Microsoft exposed if compute gets scarce. On Q2 2026 data, Microsoft's implicit and off-balance-sheet infrastructure commitments were 17 times its headline borrowings according to our AI Bubble Barometer. 1. Microsoft signs a multi-year deal to rent capacity. The partner, a specialist AI cloud such as $IREN or Nebius ($NBIS), typically raises the capital, builds the site, buys the chips and owns the assets. 2. Construction risk, the risk that the GPUs lose value, and most of the financing sit with that partner and its lenders. 3. Microsoft swaps capital spending today for rental payments later. Compare Alphabet ($GOOGL) at 10.2 times. Almost all of its off-book commitments are purchase orders, $811bn of them, for chips and equipment Google will own. We read Microsoft's choice as less than full confidence in the buildout, and Google's as the opposite. In our view the market underweights the risk that compute gets scarce. The partners own any capacity beyond their Microsoft contracts and can sell it to whoever pays more. Microsoft would then have to wait or pay up, as it owns fewer of the data centres and chips it uses. That means higher compute costs and tighter margins. On our reading, Microsoft wins if AI demand falls short, because its partners carry most of the asset risk. If compute gets scarce, Microsoft loses, because renting gives it less control of supply, and Google wins because it will own the chips it ordered.

Not So Fast

Entravision, the portfolio’s largest position, fell more than 40% from its highs as Q3 returns dropped to 37.8% YTD.

ELME · long

Elme Communities

Elme Communities, a REIT, is nearing the end of a long liquidation process.

Zentek Research Note

Zentek shipped its first ZenGUARD air-filter order to a Canadian federal facility as its Albany graphite PEA confirmed a US$3.85 billion…

The Next Consumer is a Computer

Meta’s Muse illustrates how persistent AI agents shift compute from per-query to per-user, driving heavier GPU, CPU and memory demand.

Publications we follow

Every publication on this site, hand-picked.

110 publications

Newsletters 93

X accounts we follow

Every account on this site, hand-picked.

72 accounts

Investor letters we follow

189 firms

Sectors

Ordered by how often our editor reaches for them.