Given the massive investments in AI infrastructure ($700B estimated for this year alone), it seems like these 3 AI instrastructure build-out companies are possible candidates for our investment dollars. The problem I’m having is trying to get a uniform comparison matrix.
I’m purposefully skipping a number of things, as they don’t matter for these “story stocks.” I don’t care about last quarter’s revenue nor profit. What I care about are their contracts for the build-outs, how far along are they on those build-outs, and how profitable those contracts potentially are. Likewise, I don’t care about the “NeoCloud” AI software infrastructure business claimed by CoreWeave and Nebius - that’s likely to remain small dollars from small companies, and there’s no doubt the raw-metal hardware revenue will swamp all of these companies in the next 1-4 years.
So, what’s the proper comparisons? Use of terms like “Active Power” versus “Contracted Power” versus “Secured Power” are confusing and likely measuring different things at different companies.
Anyone got a good uniform comparison matrix for these companies?
I really liked that one…not sure if he’s published an update or not…maybe you’re better at searching for it than I have been.
Also, it’s missing the #1 metric I value, which is 2026 ending ARR expected. I’m not sure Coreweave gives this, but you can probably estimate around $15b. Nebius gave a $7-9b range when they reported the September quarter and affirmed it with the Dec quarter. Iren did likewise with a $3.4b estimate.
Just as important IMO, all of that has to be balanced vs mkt cap (with increasing share counts, esp for Iren) and debt.
Thanks - I see “active power” there, but not “contracted” nor “secured.”
I would think, but could be wrong, that “contracted” is what the contracts with MicroSoft, Oracle, Meta, etc. are for - promises of server capacity to be delivered of the next N years.
And I would think that “secured” is what the companies themselves have arranged and locked in with power companies and local agency permits.
So, for instance, a company that has a large contracted power number but a low secured power number is at a higher risk than a company whose secured power is at least as high as its contracted power.
All these companies are losing money, but I’m surprised that losses aren’t listed in that chart.
I found this video, which has a bunch of numbers, but again, the categories aren’t consistent across the comparisons:
@Smorgasbord1, out of curiosity, why do you view these companies as story stocks?
I only follow Nebius and do not view it as a story stock. Revenue, ARR, capital expenditures, debt, net income, free cash flow, etc., are all highly relevant to evaluating the company and its prospects as it grows.
I agree that power is an important consideration. Sometimes terms like these are defined in earnings calls or supplemental information. If not, I suggest writing investor relations for each company asking for a more detailed definition of the terminology they use.
Yet, of those, only debt showed up on the table Bear forwarded. And the company continues to have negative EBIT. But, promises a “medium-term” EBIT target of 20% to 30%. That’s a story.
The forward-looking numbers aren’t based on the past. They’re based on new deals struck just last year and the expectations from those deals, assuming no execution risk, some of which is outside of their control.
For instance, since these companies signed their deals with hyperscalers last year, memory prices have shot up through the roof. Who adsorbs the price increases? The hyperscalers or the “NeoClouds?” Do we even know the economics of these deals - what it’s costing them to secure power, build the physical buildings, run cooling, hire personnel, buy the servers, install and configure them, maintain them, etc?
Nebius wants us to think their NeoCloud is their future, but right now it’s clear that their future is really renting out AI hardware. They don’t break out that I could find revenue from their AI Cloud versus server rental. I’d love to see those numbers.
Nebius said in the recent ER Call: “Our main strategic focus is on – is to scale our core AI cloud business, which is our multi-tenant AI cloud.” and "we are, as you know, much more focused on our AI cloud business. "
They even sound defensive on the transcript:
"Again, this is real business. These are real customers of them paying real revenues. "
But, then the numbers tell a different tale: "We are reiterating our annualized run rate revenue of $7 billion to $9 billion by the end of 2026. "
So, where does that revenue come from? Not the AI cloud business:
“we exceeded the high end of our 2025 ARR guidance and showed more than $1.2 billion ARR, because we already contracted more than 2 gigawatts of capacity and are on track to exceed 3 gigawatts this year, because we have already delivered all of our capacity for the Meta contract, because we are on track to deliver the capacity for Microsoft through the course of 2026…”
While they hype their AI cloud infrastructure software, what’s really bringing in the revenue are the bare metal rental deals.
They expect to have an ARR of $7B to $9B by end of this year. However, they also expect Capex costs of $16B to $20, of which they can fund only 60% from their balance sheet, existing operations and commitments. You have to believe in the story of borrowing money to fund the build-out to capture future payouts.
I’m not saying it’s a bad story, but it is a story, and the economics are unproven.
BTW, did you notice they just changed their depreciation schedule from 4 years to 5 years? Oh, that they’ve already set-up for an ATM equity raise for up to 25 million shares (which they haven’t tapped yet)?
These companies haven’t proven themselves in the new business models they’ve set-up for themselves. The stories they’re telling sound reasonable in some cases, but they are stories.
Of the three, Nebius irks me the most because they tell a story of AI infrastructure software when what’s really driving results, and will for the forseeable future, are the server rental fees they collect.
One tidbit from the Nebius call: When asked about the high Capex, COO Ophir Nav went on a long winded response that revealed an interesting stat:
• Securing power is less than 1% of their Capex
• Building the data centers themselves are 20% of the Capex
• The rest of the Capex is deploying the servers
So, it’s not the cost of securing power that’s an issue (and probably applies to everyone), it’s the politics and local requirements that are the issues. Hence my initial question about “secured power.”
I agree these are all highly relevant. I also agree there’s more story than substance to the current neocloud leaders. Full disclosure, I own 13% NBIS and 7% IREN, so I’m certainly not shying away from either (with the secondary disclosure NBIS’s size is not due to pushing it so much as owning it since before the MSFT deal and having it hold up much better than many other holdings this earnings season).
Yes, each is seeing real revenue and cash flow, but those numbers pale in comparison to the CapEx and buildout requirements still to be met. Both deserve credit for being able to show real present numbers. At the same time, both deserve close scrutiny as they work to close the gap between today’s revenue and still quite dreamy projections of their future.
IREN has positioned itself as a bare metal provider who will find its long-term edge through power efficiency. NBIS has positioned itself as a bare metal provide who will find its long-term edge through additional AI stack services. Not to oversimplify, but the most recent report for each was pretty much reiterating that dreamy future (again, a future with both bold and demanding timelines) while posting present bare metal numbers that were meh at best.
While I wouldn’t call either a meme stock, I’m also quite aware both are running story-heavy playbooks at this point.
To pre-empt pedantry, they are all pretty deeply FCF negative with severe doubts that this will (ever?) change, but your point holds (that current levels pale in comparison to forward expectations). If they hit their 2026 ARR targets (and analyst revenue estimates), NBIS will grow revenue (from its current nigh-irrelevantly low level) by something close to 1,000% YoY.
I’m still grappling with how power factors in for these companies. I think it might give us the best insights to the future.
This from Nebius:
Last quarter, we said we would secure 2.5 gigawatts of contracted power by the end of the year. We are already at more than 2 gigawatts now in February, which is why we are raising our forecast for 2026 to more than 3 gigawatts. And we are well on track to deliver 800 megawatts to 1 gigawatt of those as available data center capacity, just like as we spoke about last quarter.
That’s securing, which is different than actually delivering as “Data center capacity.” Nebius said:
we have already delivered all of our capacity for the Meta contract… we are on track to deliver the capacity for Microsoft through the course of 2026, exactly as planned.
and
We are well on track of achieving our goals that we mentioned about 2026 of this 800 megawatts to 1 gigawatt goal around year-end.
which got clarified further as:
And we are well on track to deliver 800 megawatts to 1 gigawatt of those as available data center capacity…
My concern here is that we keep getting talk of “watts,” but that’s like saying “miles” without saying whether those miles are planned, mapped, or actually driven.
Question 1: How many watts of Data Center Capacity does Nebius have today? What was the Meta contract for? Nebius says that Meta capacity is fully delivered today. So is the 800-1000 Megawatts a number of additional capacity, or total Data Center Capacity?
Does anyone know the 2026 contracted power in the Microsoft deal? We know it’s spread out over years (5?), but I don’t think we know the levels for each year.
IREN says:
We have secured more than 4.5 gigawatts of power, stood up 810 megawatts of operating data centers
So IREN and NBIS have about the same Data Center Capacity today (~800MW).
One difference is that Nebius already has customers for its capacity while IREN apparently doesn’t, as implied by this statement:
We have 810 megawatts of existing data centers that can be immediately leveraged for AI cloud deployments.
And clearly, Nebius has more signed contracts than Iren today. However, the story on Iren is that they have a ton of secured power (45GW) and since demnd isn’t an issue (they say), they’re just looking for the best partner/deal.
While Nebius’s ARR is $1.2B today, they promise $7B-$9B by end of year, thanks to the Microsoft contract. IREN is already at $2.3B, but their EOY target is only $3.4B.
On the debt side, Nebius hasn’t said where 40% of the Capex needed for the Microsoft build-out will come from.
Iren said:
Financial year-to-date we have now secured $9.2 billion from customer prepayments, convertible notes, including the $2.3 billion issued in December, GPU leasing arrangements and the dedicated GPU financing for the Microsoft contract.
Of that $1.9B was a Microsoft pre-payment.
Iren says the Microsoft build-out Capex will be $5.8B plus another $3B for Verizon’s build-out Capex. Iren has $3.6 billion in underwriting commitments at an interest rate of less than 6%, which they say will fund approximately 95% of GPU-related CapEx for the $9.7 billion AI contract with Microsoft. So, it appears Iren knows 95% of its needed Capex while Nebius only know where 60% of its needed Capex will come from. Of course, Nebius’ Capex is probably larger due to its larger signed contract deals.
The Iren story is based on this, again from the ER call:
Our $3.4 billion ARR target for the end of calendar 2026 reflects utilization of only around 10% of our 4.5 gigawatts of secured grid-connected power capacity. That means the vast majority of our portfolio remains available to support additional deployments. With demand continuing to build that secured capacity gives us the ability to keep engaging customers on new large-scale opportunities and to extend growth well beyond the 2026 target.
If securing power is indeed a constraint (if not a major cost), then perhaps Iren has an advantage. But if anyone, including Nebius, can secure power easily and quickly, then it’s harder for me to see Iren’s potential advantage.
Regarding Nebius, you can find most of the information from their share holder letter and earning transcripts.
As Nebius said in share holder letter, they have ~170MW active power today (VS 100MW guided). Active power is generally the source of truth about the power that generates revenue, rather than contracted or connected power.
Nebius said “We also remain well on track to reach 800 MW to 1 GW of connected power by the end of 2026.” This should be the total capacity, not additional. We should expect the active power is below this number because connected power includes the power that is not active yet.
Nebius never disclosed how much power is used for Meta deal, but DataOne (Nebius’s colocation partner for NJ site) ’s CEO did publicly talked about the Microsoft deal, which will be supported by 350MW total utility capacity (50MW for phase 1 is already delivered and the other 300MW phase 2 will be delivered by November per DataOne CEO). Please note the difference between the total utility capacity and IT load here. IT load = total utility capacity / PUE. The PUE of Nebius’s NJ data center is estimated to be below 1.2, so the 350MW utility capacity is roughly equal to 300MW IT load. $IREN disclosed that their Microsoft deal needs 200MW IT load. So this is something we can directly compare to understand the unit economics of each deal. Since Texas is a lot hotter than New Jersey in summer time, so typically the average PUE there is much worse.
If securing power is indeed a constraint (if not a major cost), then perhaps Iren has an advantage. But if anyone, including Nebius, can secure power easily and quickly, then it’s harder for me to see Iren’s potential advantage.
I’m not sure of securing power in terms of having permits or contracts, but building new power plants is definitely very very time consuming right now. From a quick Google search, the lead time of large gas turbine power generators right now is ranging from 3 to 7 years while smaller or modular units may take 12 to 18 months. That being said, I’m dissatisfied with $IREN’s conservative ARR guidance considering the energized power it’s going to have by April and I started to suspect that there could be other constraints in addition to power that is slowing $IREN down. Unless $IREN significantly revise their ARR guidance to the upside in the upcoming quarters, I don’t buy in their power being any game changer.
Anyone got a good uniform comparison matrix for these companies?
I don’t have any simple good metrics to use, but I would generally look for:
ARR (source of truth of the execution)
unit economics of large deals (e.g. expected margins)
Debt terms that they manage to get(e.g. rates)
Operating margins, EBIT margins (I understand that they are negative but I want to see up trends)
I’m actually paying less attention to power, because:
power is only a proxy while ARR is the source of truth.
I might be wrong but I tend to think that power is only the short term constraint in 2026 / 2027 and won’t last forever. Only half year ago, Nebius was guiding $1GW contracted power by end of 2026. Now, they are guiding $3GW+. Although these newly contracted power probably won’t generate revenue until 2028, I did not see any major bottleneck yet for Nebius to secure these power contracts.
In addition, there’s one key differentiator of Nebius that we should be aware of, the Sovereign AI. Since Nebius is an European company, it’s often a natural choice for EU to build sovereign AI with. Currently, Nebius’s roadmap is way more geographically diverse than $IREN’s as we can see in the following screenshot from the share holder letter.
Lastly, I respectfully disagree that software stack is useless. Nebius’s Microsoft deal and Meta deal should only generate a little over $4B ARR based on the contracted numbers. Nebius’s leadership clearly said in earning call that, although they are open to hyperscaler (bare-metal) deals, they have a preference to AI cloud deals. Let’s give them some time to see what types of deals will fill the remaining gap of 2026 $ARR before we conclude that the story is all about mega bare-metal deals.
When $IREN was trying to emphasize that they did have software capacity, they said this in the earning call:
one of the contracts we are negotiating at the moment is a multibillion-dollar contract where we need to bring a software solution.
And I really like Nebius’s quote:
The AI start-ups today are the future enterprises.
I’m currently on holiday so am only just catching up with all of the really great threads currently on the board. To answer @Smorgasbord1 re the differences between active, contracted and secured power, this is how I’ve always understood it….
Contracted power: Secured by binding agreements but still needs site build-outs or utility work to be usable. It’s a forward-looking metric showing what Nebius has locked in via signed contracts, not current operational capacity.
Connected power: Includes all capacity that is built out with utility connections and can be switched on quickly as soon as GPUs are installed—ready to go live at any time.
Active power: The portion of capacity where GPUs are already installed, running, and delivering compute to customers—what is currently operational.
