Updated my $NBIS valuation at $158.74 The neocloud selloff makes sense These businesses require enormous amounts of capital. If credit gets more expensive or less available, capex becomes harder to fund, growth expectations come down, and the returns on incremental capacity get squeezed That ultimately feeds into valuation I also think this is a healthy reset for the space The theme got hot enough that almost anything tied to AI infrastructure benefited. A tougher capital environment should start separating companies with real economics, funding advantages and customer demand from those largely riding the same tailwind And "neocloud" isn't one balance sheet. Leverage, customer prepayments, contract structure, cost of capital and dilution risk can look very different across the group So I went back through my NBIS assumptions The main takeaway is unchanged: I still don't need particularly aggressive assumptions for the valuation to work Capacity I model 1.6 / 1.8 / 2.0GW of active power at YE27 Management is targeting >1GW of new capacity brought to market annually from 2027, but I'm not assuming another 1GW simply becomes active and revenue-generating every year I'm also not treating the entire secured or future capacity pipeline as though every MW immediately gets monetized ARR / MW Using an illustrative mix of the pricing buckets NBIS has disclosed, I estimate blended economics around ~$18.8M/MW The underlying pricing framework I've worked through is roughly ~$12M/MW for the existing base, ~$20–25M/MW for newer deals and $40–50M+/MW for short-term capacity My bear/base/bull cases use: $16M / $17M / $18M per active MW Even the bull case sits below that illustrative blended figure Margins Management's medium-term target is 20–30% adjusted EBIT margin I use 20% in every scenario No margin expansion in the bull case Growth I apply only a 20% revenue uplift into 2028 rather than mechanically assuming another >1GW of capacity becomes fully active and monetized That's intentionally much less aggressive than simply extrapolating management's capacity ambitions directly into revenue Dilution I use 384M diluted shares in bear and 403M in base/bull — well above the current economic basic share count The higher count is deliberately punitive and effectively assumes a much harsher dilution outcome than today's share structure Debt I assume $40B of YE27 net debt That's a deliberately conservative forward funding assumption — not today's debt balance and not management guidance The point is to explicitly account for the capital required to fund this buildout rather than pretending today's balance sheet carries NBIS all the way through 2027 Software / asset-light I assign no separate value to software, inference or asset-light upside beyond what's already captured in the core AI Cloud economics If those develop into a meaningful incremental earnings stream, that's upside optionality — not something I need to make the valuation work Finally, I discount the YE27 scenario values back at a 15% hurdle rate, rounded from the ~15.6% CAPM framework I've been using The bear case is the one I care most about. Even with 384M diluted shares, $40B of YE27 net debt, a 20% adjusted EBIT margin and a 20x multiple, the discounted value still lands above the current price NBIS also looks different from some peers in how it is funded. Leverage and customer prepayments matter when the cost of capital rises. If those advantages persist, a financing scare shouldn't hit every neocloud the same way — even if the market initially prices them together Where I could be wrong is straightforward: deployment slips, pricing weakens, utilization disappoints, or financing gets expensive enough to materially change the returns on new capacity. Any of those pull the valuation down quickly Credit risk is real. The volatility is uncomfortable. If funding stays structurally more expensive, deployment slows, utilization disappoints or pricing weakens, the economics change — and so does my valuation But a broad neocloud de-rating also creates a better environment to distinguish between what management plans to build, what the economics can support, and what I'm actually willing to underwrite I'd rather do that work during a financing scare than when every AI infrastructure name is going straight up

Jason Ader at William Blair said neoclouds are receiving "sloppy seconds" from hyperscalers and could see business pulled back if it comes to a point where they have enough capacity themselves. He believes the probability of this is low and sees a continued partnership between neoclouds and hyperscalers, but also indicates a possibility where hyperscalers dump neoclouds after enough capacity is met. $NBIS $IREN $CRWV