I invested in Nebius Group, I was just a naive trader who doesn’t know anything about AI data centers, neoclouds, or hyperscalers, it was just an options selling play. But, I started researching about the AI data center trade more a year ago and even though there are some of the headwinds, I would be saying: When big AI companies are spending on the AI play, it will lust.
NBIS was one of my big investment, I bought huge number of shares, (huge for me) right before they announced the Microsoft deal. There are many reasons these big companies are actually outsourcing the data centers and one of the reasons is that these companies are looking for the risk-averaging. They could terminate the contracts anytime they want to.
Am I an expert? No. Not in a million years. I don’t know, besides just the basics, of how an AI data center is made or what components are there except for the GPU, racks, and power. And, that heat is producing and that the GPUs is an asset that is depreciating.
The Liquidation:
A lot of companies, almost everyone in the AI data center play has liquidated their stocks. NBIS, IREN, WULF, RIOT, ORCLE, and many others. They use this money to build more data centers. With the money raised, they need to show that it works.
Here is how much they raised: The money includes the money they raised in senior secured notes and convertible notes.
- TeraWulf: ($WOLF): Raised 3.2 Billion dollars.
- Bitfarms: Over 1 Billion dollars.
- Cipher Mining: Over 1.1 Billion dollars.
- Marathon Digital: Over 1.15 Billion dollars.
- Iren: 875 Million dollars.
- CleanSpark: 1.15 Billion dollars.
- Core Scientific: 400 Million dollars.
- Applied Digital: 2.4 Billion dollars.
- Hut8: 150 Million dollars.
Miners have raised over $15 Billion dollars in the last year till end of the 2025. These selling of notes will be continue as most of these companies are unprofitable and they need more money to support the deals they are signing.
The good thing about these funding is that it is expiring in 2030s. But, another important thing here is that how they will need more funding to acquire more GPUs and other equipments plus construction if they get more deals. For example, Iren recently raised over 2 Billion dollars because they need to fund their scaling to support the deal they signed with Microsoft.
The power constrain:
Here is the video that better explain how Electricity prices are hugely impacted by the AI data centers:
<A Video here> Embed.
One of the moat of the AI data center plays is how cheaper they have the power contract. Electricity is costing more and more and it is also made the power available to common American households more expensive.
There is already a push from the senators to ban data centers and the huge electricity it consumes plus the noise that it generates. For example, one of the video of Bernie Sanders where he said about unprecedented threat from AI.
Here is a graphical view of the US based data centers:
It is also a moat to have the access to power and contracts with electrical supply companies. One such company is IREN. I have heard numerous number of times that how having a signed contract with a company for power is a huge moat, in my opinion, it is a huge moat especially for the future when other data centers have to pay more for the same kWh compared to companies that have cheaper contracts.
Why big companies are awarding contracts to hyperscalers?
A. They are outsourcing risk.
B. It is hard to build a data center, so the demand might be high enough for them to manage so outsourcing it to the possible non-competitors.
C. The AI race. You need GPUs to train and if you wanted to win, you need to produce the best possible models.
Companies that pivoted from Bitcoin mining centers to AI data centers, they just have the bare metal. Which means, it is hard for individual researchers to train their models. For this, NBIS has the bare metal plus the software they need.
What could go wrong? ‘
While I do investment, I always find out what could go wrong. In the AI data center race, most of the risks are outward then inward. It is more Macro. For example, the Mag7 and other big players reducing their AI Capex is one of the huge risk for all the data centers.