The AI infrastructure race is entering its next phase. It is no longer simply about who can buy the most GPUs.
It is increasingly about who can secure power, land, data-center capacity, financing and the ability to bring that capacity online fast enough.
That shift is creating one of the most interesting corners of the AI infrastructure market: NeoClouds and former Bitcoin miners transforming themselves into AI infrastructure landlords.
Names such as $CRWV, $NBIS, $IREN, $CIFR, $WULF, $RIOT and $CORZ are increasingly sitting between the world’s largest AI companies and the physical infrastructure required to run their workloads.
And the deal flow is becoming difficult to ignore.
1. The Biggest Bottleneck Isn’t GPUs Anymore
For years, the AI infrastructure thesis was straightforward:
AI → GPUs → NVIDIA
But the bottleneck keeps moving.
Now the equation increasingly looks like:
AI models → GPUs → data centers → electricity → transmission → cooling → land → financing
And that creates a major opportunity for companies that already control some of those scarce inputs.
This is why former Bitcoin miners have become so interesting.
Bitcoin mining required:
- cheap electricity
- large amounts of power
- grid interconnection
- land
- substations
- transmission infrastructure
- data-center-like facilities
AI data centers require many of the same things.
The difference?
AI workloads can potentially generate significantly more predictable, long-duration contracted revenue than Bitcoin mining.
That’s the pivot.
2. The Bitcoin Miner → AI Landlord Transformation
The most interesting part of this entire theme may be the transformation of companies that were previously valued primarily on their Bitcoin mining economics.
$WULF — TeraWulf
TeraWulf is probably one of the clearest examples.
The company signed a 20-year lease with Anthropic covering approximately 401 MW at its Justified Data Campus in Kentucky.
The agreement is expected to generate approximately $19 billion of contracted revenue over the initial term, with two additional five-year extension options potentially taking the total to roughly $33 billion. Initial capacity is expected in the second half of 2027, with the full 401 MW targeted for early 2028.
This is an enormous change in the economic profile of the asset.
The important point isn’t simply “$19 billion.”
The important point is:
401 MW × 20 years × high-quality AI customer.
That’s infrastructure monetization.
3. $RIOT — Another Bitcoin Miner Becomes an AI Data-Center Developer
Riot’s Anthropic agreement is another important data point.
Riot signed a 20-year lease covering 191 MW at its Rockdale, Texas campus.
The initial contract is worth approximately $9.1 billion, with two five-year extensions potentially increasing total value to approximately $16.1 billion. Riot expects to deliver the first 96 MW in December 2027 and the full 191 MW by June 2028.
Even more interesting:
Riot already has an AMD relationship.
Together, its AI data-center contracts cover 241 MW and approximately $9.8 billion of long-term contracted revenue.
This changes how investors should think about RIOT.
It isn’t simply:
Bitcoin miner
It increasingly becomes:
Power + land + interconnection + data-center developer.
That’s a completely different valuation framework.
4. $IREN — One of the Most Advanced NeoCloud Transformations
IREN is another company I think investors should pay close attention to.
The company has moved aggressively from Bitcoin mining toward AI Cloud.
The Microsoft relationship alone represents approximately $9.7 billion of contract value through 2031, with GPU capacity being delivered in tranches. Horizon 1 was delivered and accepted by Microsoft in August 2026.
Then there’s NVIDIA.
IREN signed a five-year, $3.4 billion AI Cloud contract with NVIDIA, while NVIDIA also received rights that could allow it to invest up to approximately $2.1 billion in IREN under specified conditions.
And IREN says it has a 5 GW secured power portfolio with substantial additional AI capacity under development.
This is exactly the kind of transition that makes the NeoCloud theme so interesting:
Bitcoin → Power → Data Center → GPU Cloud → AI Infrastructure
5. $NBIS — The Pure-Play AI Cloud Angle
Nebius is somewhat different.
It isn’t simply converting Bitcoin infrastructure into AI infrastructure.
It is building a more vertically integrated AI cloud platform.
NVIDIA committed approximately $2 billion to Nebius, while the companies announced a partnership targeting deployment of more than 5 GW of NVIDIA systems by the end of 2030.
Microsoft’s agreement with Nebius has an estimated contract value of up to approximately $17.4 billion through 2031, subject to deployment and availability conditions.
Nebius has also expanded its relationship with Meta, with agreements representing up to approximately $27 billion in total contract value.
This is important because it demonstrates something bigger:
NeoCloud demand isn’t coming from one AI company.
It is increasingly coming from multiple hyperscalers and AI developers competing for the same infrastructure.
6. $CRWV — The Most Obvious Customer-Concentration Story
CoreWeave is arguably the poster child for the NeoCloud movement.
And the numbers explain why.
Meta’s expanded agreement with CoreWeave is approximately $21 billion through 2032.
CoreWeave also has a major OpenAI relationship, with an agreement under which OpenAI committed to pay up to approximately $6.5 billion through May 2031.
NVIDIA has also invested $2 billion in CoreWeave, while the companies are targeting more than 5 GW of AI factories by 2030.
This creates an interesting flywheel:
NVIDIA supplies GPUs → CoreWeave buys/deploys them → AI companies rent the capacity → NVIDIA owns equity exposure to CoreWeave.
That’s powerful.
But it also introduces risk.
CoreWeave itself warns that customer concentration is likely to remain significant because of the long-term nature of its contracts.
So investors need to separate:
contracted demand
from
economic diversification.
They aren’t the same thing.
7. $CIFR — The Power-First Model
Cipher is another name that illustrates why power availability is becoming the real asset.
Its Black Pearl project involves approximately 300 MW of data-center capacity under a 15-year AWS lease.
The project has approximately $5.5 billion of contracted revenue, with Amazon guaranteeing base rent and operating expenses under the agreement.
That’s a very different business model from simply mining Bitcoin.
The underlying asset is essentially:
secured power + interconnection + land + data-center infrastructure.
The AI/cloud tenant provides the monetization.
8. $CORZ — The Massive Optionality
Then we get to Core Scientific.
AMD and Core Scientific announced a partnership with the potential to support up to 2.5 GW of leasable capacity, anchored by more than 500 MW beginning in 2027.
The initial agreements cover approximately 530 MW across five sites, with more than $14 billion of potential base contracted revenue.
This is where investors need to be careful.
2.5 GW is not 2.5 GW of current revenue.
It’s an expansion ceiling.
That’s still extremely valuable because it demonstrates potential scale, but optional capacity should never be valued the same way as contracted, financed and operational capacity.
9. NVIDIA Is Playing a Different Game
One of the most fascinating pieces of this entire ecosystem is NVIDIA.
NVIDIA isn’t merely selling GPUs.
It is increasingly investing throughout the AI infrastructure stack.
Examples include:
$2B investment in Nebius
$2B investment in CoreWeave
up to ~$2.1B investment rights in IREN
while simultaneously becoming a major customer or infrastructure partner.
This creates a circular ecosystem:
NVIDIA → finances infrastructure → infrastructure buys NVIDIA GPUs → AI companies rent infrastructure → AI demand drives more NVIDIA GPU demand.
That doesn’t automatically make the demand artificial.
The demand is real.
But investors need to understand the financial structure behind the headline numbers.
10. The Deal Map Has Three Layers
I would break this entire opportunity into three baskets.
Basket #1 — Power / Infrastructure Owners
$WULF
$RIOT
$CIFR
$CORZ
$HUT
These companies are increasingly monetizing:
- power
- land
- transmission
- substations
- data-center campuses
- existing interconnections
The Bitcoin mining heritage is becoming less important.
The infrastructure is becoming more important.
Basket #2 — NeoCloud AI Providers
$CRWV
$NBIS
$IREN
These companies sit closer to the compute layer.
They aren’t simply landlords.
They are building and operating AI cloud infrastructure.
This creates potentially higher revenue opportunities — but also higher capital requirements, GPU depreciation, financing needs and execution risk.
Basket #3 — AI Giants
NVDA
MSFT
META
GOOG
AMZN
OpenAI
Anthropic
These companies are the demand engines.
Their contracts tell us something extremely important:
They are willing to lock up infrastructure years ahead because they are worried they won’t have enough compute.
That is arguably the strongest signal in the entire map.
11. The Most Important Number Isn’t Contract Value
This is where investors can easily get fooled.
A headline saying:
”$19B AI contract”
sounds enormous.
But that doesn’t mean $19 billion of revenue arrives tomorrow.
TeraWulf’s $19B Anthropic agreement spans 20 years.
Riot’s $9.1B agreement also spans 20 years and begins with phased delivery in 2027–2028.
IREN’s $9.7B Microsoft contract runs through 2031 and is delivered in tranches.
Therefore, the investor’s real questions should be:
How much is contracted?
How much is financed?
How much is under construction?
How much is energized?
How much is generating revenue?
What is the expected EBITDA/NOI margin?
Who funds the buildout?
That’s much more important than simply adding up headline contract values.
12. Financing May Become the Next Bottleneck
This is one of the biggest risks investors should not ignore.
AI infrastructure is extremely capital intensive.
Data centers require:
- land
- substations
- transformers
- cooling
- networking
- buildings
- GPUs
- power infrastructure
- construction financing
And financing conditions matter.
Recent reporting has highlighted rising AI-related debt issuance and growing lender scrutiny around power availability, project delays and data-center financing.
This creates an interesting paradox:
AI demand can be enormous while individual NeoCloud companies can still struggle financially.
Why?
Because demand doesn’t automatically equal available capital.
13. The Biggest Risk: Overbuilding
There is another side to this trade.
NeoClouds are benefiting from extraordinary demand today.
But investors should remember what happened to telecom infrastructure during previous technology booms.
The question isn’t:
“Will AI require more compute?”
I believe the answer is obviously yes.
The harder question is:
How much infrastructure will ultimately be economically profitable?
Reuters recently compared the NeoCloud boom with the early-2000s alternative telecom buildout, highlighting concerns around capital intensity, reliance on NVIDIA and uncertain long-term utilization.
That’s the bear case.
If too much capacity gets built, pricing power eventually falls.
The winners will be the companies with:
lowest-cost power + best locations + strongest customers + strongest balance sheets + fastest execution.
14. What I Would Watch Going Forward
1. Power availability
Forget theoretical gigawatts.
Look for:
approved + interconnected + energized MW.
That’s the real asset.
2. Delivery milestones
Watch whether companies hit:
- construction milestones
- energization
- commissioning
- customer acceptance
- revenue commencement
A signed contract is only the beginning.
3. Financing
Track:
debt → interest expense → dilution → capex → free cash flow.
A company can have billions of contracted revenue and still need enormous amounts of capital to build the infrastructure.
4. Customer diversification
One giant customer can make a company look stronger than it actually is.
I prefer the trajectory toward:
Microsoft + NVIDIA + Meta + Amazon + AI labs + enterprise customers
rather than dependence on one tenant.
5. Power economics
The long-term winner may not have the most GPUs.
It may have the cheapest reliable power in the right location.
That is why the former Bitcoin miners deserve attention.
15. The Bigger Picture
The AI infrastructure stack is becoming clearer.
Silicon → Data Center → Power → Grid → Compute → AI
And every bottleneck creates another investment opportunity.
The first wave was:
GPUs
Then:
Networking
Then:
Optical
Then:
Data centers
Now:
Power + grid + energy + cooling + infrastructure
NeoCloud sits right in the middle.
That’s why the transformation of Bitcoin miners is so interesting.
They already spent years acquiring something that AI desperately needs:
power infrastructure.
The market may have originally valued them for the Bitcoin they could mine.
The next valuation regime could increasingly depend on the AI compute capacity they can host.
Bottom Line
I don’t view $WULF, $RIOT, $IREN, $CIFR, $CORZ, $NBIS and $CRWV as simply another group of AI stocks.
They represent different ways to own the physical infrastructure behind the AI compute boom.
The most important distinction is:
Not all MW are equal.
Not all contracts are equal.
Not all backlog is equal.
The winners should be the companies that can turn:
Power → Data Centers → GPUs → Customers → Cash Flow
with the least friction.
And that’s why I keep coming back to one investment principle:
Follow the bottleneck.
AI needs compute.
Compute needs data centers.
Data centers need power.
And whoever controls the scarce infrastructure may capture a surprising amount of the AI value chain.
NFA. DYOR.