The AI Power Race: 10 Neocloud Stocks Turning Megawatts Into Billions
The artificial intelligence infrastructure boom is creating a new kind of scarcity.
It is no longer only about GPUs.
It is about power.
As hyperscalers, AI laboratories and enterprises race to deploy increasingly large AI clusters, access to electricity, data-center capacity and advanced GPUs has become one of the most important constraints on AI growth.
That creates a new question for investors:
Which AI infrastructure companies are actually converting their power capacity into revenue most efficiently?
A new August 2026 analysis from The Futurum Group attempts to answer that question by comparing ten major AI neocloud and infrastructure providers using a particularly interesting metric:
contracted revenue per megawatt.
The metric is not perfect. Contract structures, duration, ownership models and GPU economics differ substantially between companies. Some figures in the analysis are estimates rather than company-reported financial metrics.
But the comparison reveals something important about the emerging economics of AI infrastructure.
Having access to gigawatts of power is valuable.
Turning those gigawatts into high-value AI compute contracts may be even more valuable.
The AI Infrastructure Power Ranking
According to Futurum's August 2026 analysis, estimated contracted revenue per megawatt looks roughly like this:
The ranking immediately shows why comparing AI infrastructure companies solely by megawatts can be misleading.
Some companies control enormous power pipelines but have monetized only a relatively small percentage of them.
Others have already converted much of their available infrastructure into long-duration AI contracts.
That difference could become one of the defining variables separating winners from losers in the AI infrastructure boom.
1. IREN: The Revenue-per-Megawatt Leader
IREN sits at the top of Futurum's ranking at approximately $40 million of contracted revenue per megawatt.
That is substantially above the rest of the group.
The important part of the IREN story, however, may be what has not yet been monetized.
The analysis estimates a total secured power platform of approximately 5 GW, while only around 350 MW is currently associated with contracted AI and high-performance computing capacity.
That potentially leaves roughly 4.65 GW of secured power that has not yet been converted into contracted AI capacity.
For investors, this creates significant operating leverage.
If IREN can continue signing large AI customers while maintaining attractive economics per megawatt, its existing power portfolio could become considerably more valuable.
But the risk is equally clear.
Secured power is not the same thing as operational data-center capacity, and operational capacity is not the same thing as contracted revenue.
The investment thesis therefore depends heavily on execution.
2. Galaxy Digital: From Crypto Infrastructure to AI Compute
Galaxy Digital represents one of the more unusual transformations in the group.
The company is widely associated with digital assets, but its Helios infrastructure has created exposure to the AI data-center opportunity.
Futurum estimates approximately $29 million in contracted revenue per megawatt, placing Galaxy second in the ranking.
Its approved capacity is estimated at roughly 1.4 GW, with approximately 526 MW contracted and another 830 MW still seeking tenants.
That makes future tenant announcements particularly important.
If Galaxy can convert a meaningful portion of the remaining Helios capacity into long-term AI infrastructure contracts, the economics of the facility could become an increasingly important part of the company's valuation.
3. CoreWeave: AI Infrastructure at Massive Scale
CoreWeave is different from many companies on this list.
It is already operating at enormous AI infrastructure scale.
Futurum estimates approximately 5.2 GW of total power capacity, including contracted capacity and additional options or letters of intent, with roughly 3.7 GW already contracted for AI/HPC workloads.
Estimated contracted revenue per megawatt is approximately $28 million.
That puts CoreWeave slightly below IREN and Galaxy on the efficiency metric, but its sheer contracted scale changes the comparison.
CoreWeave is not merely sitting on potential future capacity.
A substantial portion of its infrastructure pipeline is already tied to AI demand.
For investors researching CoreWeave stock, the central question therefore shifts from proving demand toward execution, financing, customer concentration and the economics of continuously expanding an extremely capital-intensive infrastructure platform.
4. Hut 8: Massive Power Optionality
Hut 8 may have one of the largest power portfolios in the entire comparison.
Futurum estimates approximately 8.7 GW in its development pipeline.
Yet contracted AI/HPC capacity is estimated at approximately 949 MW.
That leaves roughly 7.7 GW of potential uncontracted capacity, by far one of the largest remaining opportunities in the group.
Estimated contracted revenue per megawatt is around $28 million.
The potential upside is obvious.
If Hut 8 can convert even part of that enormous development pipeline into economically attractive AI infrastructure agreements, its current contracted footprint may represent only a fraction of its eventual platform.
But investors should distinguish carefully between pipeline capacity and operating, revenue-producing capacity.
That distinction matters enormously when valuing infrastructure companies.
5. Applied Digital: Construction Becomes the Catalyst
Applied Digital ranks fifth at approximately $26 million of contracted revenue per megawatt.
Its total platform is estimated above 2 GW, with approximately 1.41 GW contracted and around 600 MW of remaining capacity.
The investment story increasingly revolves around execution.
Data-center capacity under construction does not generate the same economics as capacity that has been completed, energized and handed over to a customer.
For Applied Digital stock, construction milestones and customer deployments therefore become critical catalysts.
Successful delivery can progressively convert infrastructure investment into contracted cash flow.
Delays can have the opposite effect.
6. TeraWulf: Strong Economics Without Owning the Entire Compute Stack
TeraWulf is estimated to generate approximately $25 million in contracted revenue per megawatt.
Its platform is estimated at roughly 2.9 GW, with approximately 522 MW contracted.
The distinction here is the business model.
Companies operating powered-shell or colocation infrastructure generally capture less economic value per megawatt than full-stack GPU cloud providers because customers may own the computing hardware themselves.
That can reduce revenue per megawatt.
But it can also change capital requirements and risk.
Investors therefore should not automatically assume that higher revenue per megawatt means a superior business.
The capital required to generate that revenue matters just as much.
7. Nebius: Intentionally Waiting for Better Economics?
Nebius may be one of the most interesting cases in the entire ranking.
Futurum estimates approximately $20 million to $25 million in contracted revenue per megawatt, with roughly 5 GW of total power capacity and approximately 2 GW contracted.
But the story is more complicated than the ranking suggests.
Nebius has been rapidly expanding its global infrastructure footprint while signing major AI infrastructure agreements.
The strategic question is whether some of the company's uncommitted capacity can eventually be sold at higher economics than its existing long-term contracts.
That would mean unused capacity is not necessarily evidence of weak demand.
In some cases, it could represent valuable inventory in a market where AI compute remains scarce.
For investors analyzing Nebius stock, the distinction between contracted power, connected power and active computing capacity is particularly important.
8. Core Scientific: Infrastructure Contracts Over Full-Stack Cloud
Core Scientific is estimated at approximately $21 million of contracted revenue per megawatt.
The company has roughly 2.7 GW of total power capacity, with approximately 1.12 GW contracted for AI/HPC infrastructure according to Futurum's analysis.
Its economics reflect a model focused more heavily on infrastructure contracts rather than operating the entire GPU cloud stack.
That can produce lower revenue per megawatt than a full-stack AI cloud provider.
But again, revenue efficiency should not be confused with return on invested capital.
The underlying capital structure and contract economics ultimately determine shareholder value.
9. Cipher Mining: Lower Revenue per MW, Different Economics
Cipher Mining sits near the bottom of the ranking at approximately $13 million per megawatt.
Its total power platform is estimated around 2.8 GW, with approximately 700 MW contracted.
The lower revenue-per-megawatt figure illustrates one of the biggest differences across the AI infrastructure industry.
In powered-shell colocation arrangements, the customer can own the expensive silicon.
That leaves less revenue for the infrastructure provider.
But it can also mean substantially different capital requirements.
For that reason, Cipher should not simply be compared with a full-stack GPU cloud operator using revenue per megawatt alone.
10. Crusoe: Significant Capacity, Limited Financial Visibility
Private AI infrastructure company Crusoe rounds out the group.
Futurum estimates approximately 2.6 GW of total capacity and roughly 1.2 GW of contracted AI/HPC infrastructure.
However, because the company remains private and contract economics are not fully disclosed, a reliable contracted-revenue-per-megawatt figure is unavailable.
Its inclusion still demonstrates how quickly the competitive landscape is expanding beyond traditional hyperscale cloud providers.
The Bigger Story: Electricity Is Becoming an AI Asset
The most important takeaway from this comparison is not which company ranks first.
It is what the ranking says about the AI economy.
For years, investors focused primarily on semiconductor supply.
That remains critical.
NVIDIA GPUs, high-bandwidth memory, advanced packaging and networking equipment continue to determine how quickly new AI clusters can be deployed.
But another bottleneck is emerging underneath the semiconductor layer:
electricity.
A company may be able to purchase GPUs.
That does not mean it can immediately find hundreds of megawatts of power, secure land, obtain interconnection agreements, construct data centers, deploy cooling systems and bring an AI cluster online.
As AI models become larger and inference workloads expand, access to power increasingly resembles a strategic resource.
That changes how investors should analyze the AI infrastructure ecosystem.
The Metric Investors Should Really Watch
Revenue per megawatt is useful, but it should never be analyzed alone.
A more complete framework would examine:
Contracted revenue per MW
How effectively does the company monetize available electrical capacity?
Percentage of power already contracted
A company with 5 GW of power but only 300 MW contracted has a very different risk profile from one with most of its capacity already committed.
Capital expenditure per MW
Higher revenue means little if achieving it requires disproportionately higher investment.
Contract duration
A five-year hyperscaler contract has different economics from short-duration GPU cloud pricing.
Customer concentration
Microsoft, Meta and other hyperscalers can provide enormous contract visibility, but dependence on a small number of customers creates concentration risk.
Power status
Investors should distinguish between secured, contracted, connected, energized and active power.
These are not interchangeable.
Compute ownership
Owning GPUs can dramatically increase revenue per megawatt, but also increases capital intensity, financing requirements and technology-obsolescence risk.
Full-Stack AI Cloud vs. Powered-Shell Infrastructure
This may ultimately be the most important distinction in the ranking.
Full-stack AI cloud providers can potentially generate much more revenue from every megawatt because they monetize not only electricity and data-center infrastructure but also GPUs, networking, storage and software.
The tradeoff is capital intensity.
A powered-shell provider may earn less revenue per megawatt but avoid purchasing billions of dollars of rapidly depreciating GPUs.
That means investors should resist the temptation to declare IREN automatically "better" than Cipher simply because the estimated revenue-per-megawatt figure is three times higher.
The correct question is:
How much shareholder return can each company generate from every dollar of capital invested?
That requires analyzing revenue, margins, financing costs, depreciation, contract duration and utilization together.
What Could Drive the Next Re-Rating?
The next stage of the AI infrastructure trade may increasingly revolve around conversion.
The market already knows that these companies control large power pipelines.
What could change valuations is evidence that those megawatts can be converted into economically attractive contracts.
That makes several developments particularly important:
New hyperscaler and AI-lab contracts.
Higher contracted revenue per megawatt.
Faster conversion of secured power into energized capacity.
Improving utilization rates.
Lower financing costs.
Expansion without excessive shareholder dilution.
Stronger free cash flow as completed campuses begin generating revenue.
The companies with the largest uncontracted power pipelines could therefore have significant upside.
They also carry significant execution risk.
The Investor Takeaway
The AI infrastructure race is evolving.
The first phase was about GPUs.
The next phase may increasingly be about power, infrastructure and monetization efficiency.
IREN currently stands out in Futurum's analysis with the highest estimated contracted revenue per megawatt.
CoreWeave stands out for the enormous amount of capacity it has already contracted.
Hut 8 stands out for the scale of its remaining power pipeline.
Applied Digital and TeraWulf offer different infrastructure economics.
And Nebius represents an increasingly interesting combination of full-stack AI cloud capabilities, hyperscaler contracts and rapidly expanding global power capacity.
There may not be one universal winner.
Different companies are effectively making different bets on where the economics of AI infrastructure will ultimately concentrate.
For investors, that means the next question should not simply be:
Who has the most megawatts?
It should be:
Who can turn those megawatts into the highest sustainable return on capital?
That is where the real AI infrastructure competition is beginning.
Continue Your Research
Use Be The Investor's research tools to go deeper:
Research CoreWeave stock
Research IREN stock
Research Nebius stock
Research Applied Digital stock
Research Hut 8 stock
Explore the Stock Screener
Analyze portfolio risk
Read the latest market news
This article is for informational and research purposes only and does not constitute investment advice. Figures derived from third-party research may include estimates. Investors should verify company disclosures and financial filings before making investment decisions.