1) Compute is scarce. Jevon's Paradox you've heard this argument a million times. The question is if you believe it or not. That's your decision as an investor.
2) $NBIS's bread and butter isn't hyperscalers. It's AI startups and enterprises. They've seen 4 customers compete for every GPU cluster coming online.
$NBIS has depended on hyperscaler contracts to supercharge their financing and increase their return on equity (minimize dilution etc.)
What hyperscaler bare metal contracts enable $NBIS to do is build faster and take the rest of the money and build for their own higher margin customers (operating leverage is at the center of the $NBIS thesis).
3) So why won't customers just choose hyperscaler compute over $NBIS?
The point here is more subtle. The hyperscalers of the world do not build specialized AI clouds. They are very much general-purpose service clouds with a host of non-AI solutions. Things like object storage, serverless functions, managed databases, CDNs, identity services, IoT platforms, and hundreds of other things that have nothing to do with training or serving a model.
AI is a feature of their cloud. It is not the thesis of their cloud.
We see this differentiation manifesting itself in real-time. $NBIS is winning in enterprise. Something stated by Roman Chernin himself. You also see $NBIS customer success stories such as Revolut and now Reflection AI.
4) So what does this all mean together?
$NBIS's AI-first cloud captures and services the leaders of the AI-native startup world. $NBIS is selective (given demand / supply imbalance) to choose customers in a VC like fashion that will be the Fortune 500 of tomorrow.
As these companies grow, scale, their compute demand scale too (Revolut is one example commonly cited) and $NBIS expands their revenue alongside them.
By starting now, $NBIS enters a fly wheel of continuous improvement by having their FDE's work closely with AI startups that are defining SOPs and requirements in real-time. This head start and advantage compounds leading to a increasingly more specialized and sticky ecosystem.
This is something generalized hyperscalers cannot compete with although they may be able to supply more compute as a whole.
Finally, with these customers $NBIS is able to monetize GPUs (or tokens) at a much higher rate. The structural advantage in unit economics and fantastic operating leverage is what sets $NBIS apart from other "Neo Clouds".
$NBIS wave 4 landing right in the pocket
38.2% fib + prior wave iv low + 50 day MA all stacked at $200-220
Hold it → wave 5 projects $327-361 Lose $132 → count dead
NFA