Three earnings prints, one thesis: AI compute is getting monetized three different ways — and the market's rewarding all of them
Nebius jumped 34% on $582M in revenue, up 454% y/y, with adjusted net loss narrowing 64% to $33.2M — a company scaling GPU-cloud revenue while visibly closing the gap to profitability. That's the "growth efficiency" trade.
CoreWeave rose 19% despite missing revenue consensus ($2.58B vs. ~$2.61B expected) — because the backlog is the real story. $104B contracted as of June 30, and that's *before* another $25B+ signed in the first weeks of Q3, pushing total forward commitments to roughly $129B. The adjusted EPS loss of $1.03 also beat the Street's -$1.24 estimate. Investors aren't pricing this quarter — they're pricing 2028.
Riot's +4.33% came off a structurally different lever: the Anthropic deal, $9.1B guaranteed for 191 megawatts over 20 years, scaling to $16.1B with extensions. No revenue growth story here — just long-duration, locked-in power economics.
Three different products from the same underlying scarcity — GPU-hours, contracted backlog, and raw power capacity. The market's paying up for all three right now, but they carry very different risk profiles: Nebius is an execution bet, CoreWeave is a duration bet on AI demand holding, and Riot is closer to a fixed-income-like power contract wrapped in a crypto-mining wrapper.
If AI demand cools even modestly, which of these three cracks first — the one with the thinnest margins, the one with the most leverage, or the one furthest from its original business model?
Comments