Neoclouds are not one trade. They are three, wearing one name.

A cluster map of the neocloud complex against the AI megacaps

Everyone buckets the neoclouds together as a single AI-infrastructure theme. The tape disagrees. Under one label sit different assets with shared AI-compute direction but very different torque, valuation support, and drawdown behavior. Treating them as interchangeable is how a basket blows up in one direction while you thought you were diversified.

Market structure note. Correlation data uses adjusted daily closes through June 22, 2026. Not investment advice.

The neocloud complex has had a violent run. Across the names below, the trailing 63-day move spans roughly 36 to 141 percent. That kind of synchronized rip is exactly what convinces people they are looking at one trade: buy the theme, own the basket, ride the AI buildout. The problem is that the synchronized direction masks very different valuation regimes and very different return magnitudes underneath it.

Plot the names by how they actually trade, and they separate cleanly. The horizontal axis is beta and torque, from the low-volatility anchors on the left to the high-beta speculative names on the right. The vertical axis is what the market prices each name on, from hard earnings and EBITDA at the bottom to pure sales multiples and momentum at the top. The clusters fall out on their own.

Cluster one. The infra-operator that trades like a hyperscaler.

CoreWeave sits apart from the rest of the neoclouds and closer to the megacap block, and that is not an accident. It is the only name in the group the market prices the way it prices a real infrastructure business: on cash earnings power, contracted backlog, and capacity utilization. The tell is in the multiples. CoreWeave trades around 10 times sales with a positive, finite EV to EBITDA near 32 times. Every other neocloud is either burning EBITDA outright or carrying a triple-digit EV to EBITDA that cannot be valued on earnings at all.

That makes CoreWeave the quality expression of the theme. It carries the hyperscaler model: sell committed compute under multi-year take-or-pay contracts to named counterparties. The risk that comes with it is also hyperscaler-adjacent in the wrong way. The balance sheet is heavily leveraged, the GPU base depreciates fast, and the customer concentration is severe. The reported EBITDA is real but adjusted, sitting on top of a capex treadmill that the bears argue flatters it. You are buying a scaled operator, not a cheap one.

Cluster two. The hypergrowth name priced on optionality, not earnings.

Nebius sits high on the vertical axis because it is the higher-torque expression of the AI-compute trade. It tracks CoreWeave directionally: the NBIS / CRWV daily return correlation is roughly 0.73 over 21 trading days, 0.75 over 63 days, and 0.78 over 126 days. But the return magnitude is not remotely the same. Over the same 63-day window, NBIS was up about 141 percent while CRWV was up about 37 percent.

That makes Nebius neither a clean miner proxy nor a clone of CoreWeave. It is the levered growth expression of the operator cluster: same AI-compute direction, much higher valuation sensitivity, and far less of an earnings floor. That explains why a name up more than 140 percent in 63 trading days can simultaneously show heavy distribution in the order flow. Parabolic names meet supply at the top, and a momentum-and-optionality valuation has little underneath it when the momentum pauses.

Cluster three. The miner-pivot pack that whips on headlines.

IREN, Applied Digital, Cipher, and TeraWulf form their own tight cluster in the upper right: high beta, smallest caps, most stretched valuations, and a shared lineage as crypto miners pivoting into AI hosting. They move together, but they do not move on individual fundamentals. They move on AI-hosting-contract headlines. A single large deal landing or slipping can swing the whole pack, which is why intraday round-trips of 10 to 15 percent are normal here rather than alarming.

This is the torque sleeve of the neocloud trade. More upside if a transformative hosting contract lands, more downside if it does not, and very little in the way of earnings to cushion the gap. It is a sentiment instrument, not a quality one, and it should be sized that way.

What the map actually tells you about correlation.

Here is the part that matters if you hold more than one of these. Within a cluster, the names move together, but correlation does not mean equal return. NBIS and CRWV have a strong directional relationship, yet NBIS has been the much higher-torque name. The miner-pivot pack is tighter internally: recent 21-day pair correlations include CORZ / HUT near 0.90, CIFR / IREN near 0.87, CIFR / WULF near 0.87, and IREN / WULF near 0.83.

So a basket of CoreWeave plus Nebius plus the miners is not a diversified neocloud position. It is layered exposure: one lower-torque infrastructure operator, one high-torque AI-compute growth name, and one concentrated miner-pivot headline trade. The neocloud basket is correlated enough to rip with the AI tape, but the magnitude and failure modes are different enough that equal-weighting them creates concentration risk in disguise.

The one-line version.

If you want quality and you accept leverage and customer concentration, CoreWeave is the infrastructure operator. If you want the same AI-compute direction with much higher torque and less valuation support, Nebius is the levered growth expression. If you want maximum headline beta, the miner-pivot pack is the trade. What you should not do is buy all of them and call it diversification, because the map says you would be stacking related factor exposure while pretending you spread the risk.