Nebius to deliver AI infrastructure to Microsoft

[PaulWBryant]

@stonccs

Regarding „one time charges“: A GPU under full utilization has a life expectancy of 3-5 years.

This is why I have avoided Coreweave and Nebius (and any other aspiring neoclouds). If you get $50b of annual revenue at some point, but your annual expenses are $75b, you don’t really have a viable business.

I’ll use Corweave as example because it’s the furthest along (most scaled):

So we see that the cost of revenue is only about 26% of revenue. A 74% gross margin. Good start. And Sales and Marketing is tiny: awesome. Gen and admin is also very reasonable. What’s this “technology and infrastructure” that amounts to 55% of revenue? (Which was up 19% sequentially) Well, that’s largely made up of the depreciation @stonccs mentioned above. That’s only going to keep growing as they buy more and more Blackwells etc. Lastly what’s this 267m interest expense which is the main reason they’re losing a lot of money each quarter (even with now over $1b of quarterly revenue)? Well that’s just what they currently pay each quarter on the debt they’ve taken so far. And long term debt just increased 51% SEQUENTIALLY. So expect interest expense to go higher and higher.

So the point is, though their revenue is now ~$5b annually and will no doubt double and triple from here, they’re still losing money. Will that change when they get to $50b annually? I just don’t know.

I could be wrong, and these companies could reach some level where they eke out a tiny profit. But think about that. Applovin has around $5b of revenue and already has more profit than Coreweave might have at $50b revenue. Because Coreweave (and Nebius and all the other aspiring neoclouds) is a CAPITAL INTENSIVE business. I.e. Margins will not be what Applovin or any other asset light business is able to enjoy. Not even half. Maybe not even 10% of what a software business can have.

And not to sound like a conspiracy theorist, but why are the hyperscalers (AMZN, GOOG, MSFT) partnering with these companies? MSFT is actually Coreweave’s largest customer in addition to the Nebius deal that sparked this thread. My thought is that, these companies burn money building out capital so that the hyperscalers don’t have to spend EVEN MORE CAPEX than they already are. Billions of revenue for NVDA (who also partners with Coreweave and Nebius), and less capex for the hyperscalers. Everybody wins? Except Coreweave/Nebius investors??

Again, I could be wrong, and this could just be the early days of gigantic businesses scaling. The “who wins” question many of you have discussed here is still open. Hard to see Coreweave and Nebius and Iren and Terawulf and whoever else ALL becoming $50b or $100b revenue companies. Will there be 10 of these? 100?

All this to say, there are just too many questions.

Just my take, feel free to have your own!

Bear

[SailingDev]

@wpr101 I always enjoy how you lay these out, and I admit I steal a lot of your ideas .

I’ve noticed the same wave of hype on Twitter around NBIS, and yeah, it does feel a bit “sus” how often the “superior engineers” angle gets repeated. That’s not actually where my impression came from though. Yandex has been known for strong engineering: they went head-to-head with Google in Russia on search, built MatrixNet (an early ML framework that later evolved into CatBoost), and scaled out a full ecosystem of services (maps, ads, ride-hailing, cloud). Their hiring pipeline drew heavily from ACM ICPC medalists and Olympiad winners, so the talent base was objectively very high, as you also acknowledged.

That said, I agree with you that marketing narratives can get manufactured. You rarely hear Nvidia or Astera talking about “superior engineers”, the results just speak for themselves.

On your point about IREN, I think the optionality they have, between mining and HPC, is a big advantage. IREN is a medium position too for me. Thanks to you!

[Smorgasbord1]

@wpr101

From what I can gather Nebius has 100 MW of “active online data center capacity” and 220 MW of “connected power” as of September 2025. While IREN currently has 810 MW of connected power, with +1GW coming online in April 2026. HIVE has 430 MW of connected power and is a sub 1B market cap. My view here is that IREN and HIVE have done the harder work of electrifying the grid, while Nebius has focused heavily on sales and very heavily on marketing.

This highlights concerns I have about using MegaWatts as a true capacity metric. Getting enough power is one thing, having operating servers in place using that power is another thing, not to be overlooked.

First, we need to understand the term “connected power.” What does that actually mean? I found this link: https://finance.yahoo.com/news/nebius-1-gw-capacity-target-140700738.html, which describes it as: “active or ready for GPU deployment,” while this link: Nebius Stock: Set For Hyperscaler Glory (NASDAQ:NBIS) | Seeking Alpha, describes it as: “active power plus fully provisioned capacity that can go live immediately once GPUs are installed.” That defines the term with terms like “active power,” which means GPUs that are live and running, and “fully provisioned capacity,” which means resources that are installed, allocated, and ready for use.

@wpr101

My view here is that IREN and HIVE have done the harder work of electrifying the grid…

I hate bringing this up again, but these companies are not generating nor supplying electricity, nor do they “electrify the grid.” They consume power to run their data centers.

It feels like some here are thinking this is “utility power capacity,” which is the maximum amount of electricity the entire facility can draw from the grid at any given time. Obviously, that needs to be large enough to handle what these companies want to install in their data centers, but getting that capacity is first up to the actual electricity providers to the grid, then the work of wiring up that data center for power, and then the work of purchasing, installing, provisioning the actual servers. But, utility power capacity is not what these companies are talking about.

In essence, what’s going on is that these companies are using the power consumed by their servers as a shorthand for compute capacity. It’s like using engine displacement as an indicator of automobile car performance. All things being equal, a larger engine will produce more horsepower, but all things aren’t always equal. Just as turbo charging can get a smaller engine to produce more hp, a data center’s PUE (the ratio of which is the electricity the entire facility draws from the grid to the power consumed by the GPUs) is a metric showing data center efficiency.

And then there’s the issue of GPU power efficiency. One of the important metrics for new GPUs from Nvidia is their power efficiency - not just faster compute, but higher compute per watt consumed.

I wonder why we’re getting “connected power” numbers instead of the actual numbers of H100s or B200s, etc. installed. Could it be that they’re worried talking about how many Hoppers they have in a world where Blackwells are shipping and Rubins are coming?

One aspect being discussed here that’s puzzling to me is the claim that Nebius’s software stack is an advantage when Microsoft is now by far their biggest customer. Surely no-one needs to educate Microsoft on software or building and operating data center efficiently. If anyone is buying “bare metal” for data centers, it would be the likes of Microsoft, Amazon, and Google. Certainly no hand-holding needed there.

[Iamnzane]

stonccs

@Iamnzane

It is all about ARR, not one time charges cost of equipment and cost to install.

Regarding „one time charges“: A GPU under full utilization has a life expectancy of 3-5 years. Maybe more with liquid cooling - but even then they will be technically obsolete due to newer models coming out. GPUs are quickly depreciating assets.

It is good to bring up depreciating assets. Significant changes are in place 2025 for the Trump OBBB. The accelerated capital recovery opportunities under 100% bonus depreciation, QPP expensing and Section 179 represent a significant shift in the landscape of capital cost recovery. Keep in mind that NBIS has big 2025 outlays in property and with equipment so I would expect a faster depreciation schedule. This may impact the taxable net income and earnings for NBIS in a significant way given the huge capital outlay. For our near term investment thesis, I really do not care as I am not investing based upon expected earnings anytime soon. But looking at the net income, cash flow along with ARR are important at this stage of NBIS. Somebody with a better accounting background can extrapolate this further than me.

So now with Microsoft, we have X40 larger order backlog into the much longer horizon. We now know how much revenue growth will be coming for 20+ quarters into the future. The size of one order is x200 larger than Q2 Revenue.

I am very bullish on NBIS.

-zane

[Smorgasbord1]

I’m surprised that no-one seems to be talking about an obvious customer concentration issue with Nebius and its gorilla customer Microsoft.

Yes, it’s a 5 year contract with revenue numbers, and yes it’s guaranteed as long as Nebius meets the delivery milestones in the contract. Does anyone here know what those milestones are and how easy/hard it will be for Nebius to meet them?

[MFChips]

@BearBear re: depreciation (excellent points that I’d like to explore more as it pertains to NBIS):

For NBIS, I see revenue grew 90% QoQ while Depreciation grew 53%. Some % of new equip hits financial reports prior to revenue hitting - perhaps as much as a quarter or more? In such a high rev growth environment where equipment has to be procured and then stood up (setup, tested then burned-in, etc) I wonder what the lag to revenue is? I surmise it is longer than a quarter.

If in this most recent quarter, they received $105.M in revenue, which is $50M more than the quarter before. What if they took delivery of an additional $200M in GPU infrastructure to support the revenue 2-3 quarters out already under contract? So they would have $25M in depreciation on something not yet stood up. What if this was true for 2-3 quarters? How long from delivery of the goods from NVDA to full-revenue-receipt rate? How many months of depreciating assets are on the financials that are not yet contributing to revenue and how fast is that expense growing to support future growth?

[twillo]

With regard to the landmark MSFT NBIS agreement … The following is from @MVanbrunsc47513 in an X post today:

As an early Nebius $nbis investor having entered just three days after the Yandex relisting in October 2024, I believe it’s the right moment to assess potential execution risks and bottlenecks. Right now, sentiment feels sky-high, and I’m seeing some extremely bold statements being made. I’m preparing a direct contact with Nebius to raise critical questions and stress-test my thesis. I’ll provide an update once I’ve shared my remarks. Staying sharp and critical is essential at this stage.

In plain language my keypoints will be the contract details (price indexation, SLA details, pass-troughs etc etc) and how it will affect their strategy to become “the AWS of neo clouds”. If nebius truely wants to compete with hyperscalers for cloud customers why would Microsoft help them, would’t they cap that in the contract? Lets give some insight in which rigour and stance my critical but open view will be..

I have doubts if nebius $nbis has traded autonomy for validation: it gets the Microsoft brand stamp and cash flow, but almost certainly gave away indexation protections, pricing flexibility, and maybe freedom to compete. The risk is they’re locked into being a high-capex wholesale provider rather than evolving into the “AWS of neoclouds.” Why does it matter and what could be in the contract if you think it trough? Indexation risk: If power or GPU prices inflate and Nebius didn’t secure escalators, their margin erodes. Microsoft locks in cheap compute; Nebius eats the cost risk. And also asymmetry: Microsoft is simultaneously a customer and a hyperscaler. It doesn’t want Nebius to become another AWS; contractual clauses (MFNs, exclusivity, ROFRs) likely box Nebius in. Nebius may end up a supplier, not a competitor in that scenario.. There are a lot of deeper questions that need to be cracked to validate the real “value” of the Microsoft contract instead of just revenue. Does the agreement restrict Nebius from marketing certain competing public cloud services to Microsoft enterprise accounts for a period? Any co-sell or white-label constructs that effectively make this Azure capacity? Are there price-parity or most-favored clauses that force your public cloud prices to sit at or above what Microsoft pays? Again.. Lots of strategic and competitive questions about the value of the contract and Nebius if nebius sold validation for autonomy..

How do i get to these questions. Simple: Pattern recognition. What have we learned from earlier hyperscalers contracts in how they are structured. Hyperscalers prefer fixed or very limited escalation when they buy wholesale capacity whether it’s GPUs, bandwidth, or renewable energy PPAs. They push inflation risk back onto the supplier. In data center leases, escalation is often CPI-linked (2–3% annually), but in hardware/compute offtake deals, they tend to push for fixed rates per unit of compute delivered. In past large contracts (think Azure–CoreWeave, AWS–Bloomberg, Google–Anthropic), the hyperscaler often negotiates MFN (Most favorite nation) pricing: they get the best rate available.

Sometimes there’s exclusivity by geography or workload type limiting the supplier’s ability to sell the same capacity to a hyperscaler’s direct competitors. Volume optionality is also a big thing. Hence the broader price range of the contract stated up to 19.4b. The big buyers prefer optionality: commit to minimum baseline usage but keep expansion rights at locked-in pricing. This lets them scale up cheaply if demand surges, while the supplier bears idle risk if demand doesn’t materialize. For now, this is just a sneak peek at the questions I’d like to explore in more depth with Nebius. Once I’ve received their answers and had time to reflect on them, I’ll share my findings.

https://x.com/MVanbrunsc47513

[drew1618t]

Smorgasbord,

Your right to directly compare IREN energy usage as NBIS should be done skeptical. But I think your also missing part of the point. From most of Nvidia’s presentations, Nvidia emphasis the increased compute per watt. They focus on the limit for compute is not in the chip but in the power required to run it. It takes longer to build a power plant to fuel a data center than it takes to build all the chips inside. IREN has a contract for a large amount of electricity at a very cheap rate. This is where the opportunity lies for them. They can build an HPC or mine Bitcoin or whatever. They have secured a limited resource at a really cheap rate.

WPR mentioned that this is not well geographically located and it might be a detriment for IREN. In normal edge computing you would be right but in inference models people are ok waiting a few seconds for a result that the .5 second travel distance is not as much of an issue as it was in the past.

Drew,

Long IREN

[Smorgasbord1]

@drew1618t

From most of Nvidia’s presentations, Nvidia emphasis the increased compute per watt. They focus on the limit for compute is not in the chip but in the power required to run it.

Yes, but this aspect is nuanced. Blackwell’s maximum power draw (~1,000-1,200W) is typically much higher than Hopper’s (~700W), so retro-fitting Blackwell servers into data centers current running Hoppers can be tricky, especially in regards to cooling. And the higher-end Blackwells now require liquid cooling. But, there are Blackwell systems with the same 700W maximum power draw, too.

@drew1618t

It takes longer to build a power plant to fuel a data center than it takes to build all the chips inside.

For the US, sure. For China, probably not. China brings more new power online in a month than the US brings online in a year. That’s more than an order of magnitude difference, and why China gets away with using older generation AI chips, or slower, more power hungry Chinese AI chips from Huawei. Throwing power, cooling, and networking at the problem is easier/cheaper for them. Will that last? I dunno.

@drew1618t

IREN has a contract for a large amount of electricity at a very cheap rate…WPR mentioned that this is not well geographically located and it might be a detriment for IREN.

Yes on both/all accounts. IREN has historically also had curtailment contracts, which is fine for bitcoin mining, but potentially a problem for inference usage. I suspect they have different deals for different data centers though.

As for the size of Iren’s data centers, here’s a recently article:
https://www.datacenterdynamics.com/en/news/iren-doubles-gpu-fleet-as-ai-cloud-expansion-continues/

{Iren} procured an additional 4,200 Blackwell B200 GPUs, taking its total number of GPUs at its ‘Prince George’ site in British Columbia, Canada, to 8,500.

This fleet of GPUs comprises 800 H100s, 1,100 H200s, 5,400 B200s, and 1,200 B300s.

IREN expects its total mining capacity to be ~50 EH/s, while the Prince George facility hosts a 50MW power capacity - the company says that’s enough for ~20,000 Blackwell GPUs.

Good to see actual Nvidia numbers here. Note the “EH/s” metric is hashrate (Exa-Hashes per Second), which is the computational power of a blockchain network (usually Bitcoin) indicating the number of complex calculations performed each second to secure the network. So, that’s different than MW of power capacity, which to me makes that last quoted sentence confusing - one is for bitcoin mining for the entire company, the other for HPC usage in one of its data centers.

[drew1618t]

Smorgasbord,

I agree china can build more electric plants and can use less efficient chips. But none of the companies we are talking about in this thread are getting access to the energy in China. So I am going to focus on energy production outside of china. For most countries power production has not kept up with the AI demand. Energy is one of the major limiting resources with GPUs.

I think it will be easier to find ways to use cheap electricity than it will be to find cheap electricity. History has shown companies with cheap energy have a strong competitive advantage. Which is why I placed my money on a company with plenty of cheap electricity.

Drew,

[wpr101]

@Smorgasbord1

I hate bringing this up again, but these companies are not generating nor supplying electricity, nor do they “electrify the grid.” They consume power to run their data centers.

Interesting, I thought IREN was actually installing solar panels, but it seems that part of the data center build is not them. The term they use is “energization” of the grid which I guess means connecting to the grid to draw power from. I’m still a bit confused by the statements they have made that they improve the strength of the grid from a stability standpoint. I’m wondering if they are purely a consumer for the grid how that strengthens rather than stresses the grid. It seems they have some way to give excess power back.

I think it’s probably easier to refer to what the management is saying to explain this part, highlighting some areas which could be relevant.

The industry is now talking about 50 to 1,000 megawatt clusters with customers asking for 250 megawatts or more across multiyear rollouts. All of this is happening in live time and the scale and the intensity just continues to climb month-on-month. So grid connectivity, this is really a gating factor.

A greenfield hyperscale site can take up to 5 to 7 years just to secure that transmission access and energization. So that is the critical bottleneck. And it’s where we clearly believe IREN has a competitive advantage along with a few other points, which we’ll get to. So we are clearly well positioned to meet this demand. We’ve got 2.9 gigawatts of secured power capacity already contracted, including those large-scale grid connections at Childress and Sweetwater.

We own our own land, which means we’re not reliant on M&A. We’re not reliant on third-party developers, and we can move quickly when customer demand materializes and is contracted and importantly, retain the upside on the infrastructure development. So I guess my message here is the demand is real. It’s growing more and more real week by week, and it’s infrastructure constrained. We’ve got the land, we’ve got the power.

And importantly, we’ve got the engineering and the execution capabilities to capture a good share of this growth, and we’re making it happen right now. So – if we zoom out briefly and look at how we are positioned to address one of these biggest barriers to adoption. So it’s not just about the GPUs, it’s about the ability to deploy those GPUs with power, cooling, permitting and speed, and that’s where most of the market is getting stuck today. So we’ve spent years assembling the fundamental and foundational ingredients for this. We’ve secured those large-scale sites that I mentioned.

---

@Smorgasbord1

I’m surprised that no-one seems to be talking about an obvious customer concentration issue with Nebius and its gorilla customer Microsoft.

Yes, it’s a 5 year contract with revenue numbers, and yes it’s guaranteed as long as Nebius meets the delivery milestones in the contract. Does anyone here know what those milestones are and how easy/hard it will be for Nebius to meet them?

Good point here, I’m wondering the same thing. On the surface the contract value blew away expectations for the market, but does Nebius have any capacity left to sell to anybody else?

I’m curious about the milestones they will need to hit. Something to keep in mind with these hardware companies is they can become dependent on one or two hyperscalers for most of their business. Credo had an 86% customer with Amazon two quarters ago. That company Applied Optoelectronics had a huge contract with Amazon in 2017, but then Amazon cancelled it and the stock cratered.

The last thing I’m wondering about is Microsoft specifically. Bear pointed out above they are the lead customer from CoreWeave as well! With Microsoft eating up all of this capacity, where are the other hyperscalers in this field?

One of the reasons I’m bullish on IREN, is I believe they are looking to sell allotments in 200 MW+ of deals, and their bottleneck will be acquiring machines, and not likely related to the complexities of electricity.

[twillo]

I’ve been doing some research on $EOSE. They’ve guided full year 2025 revenue to between $150M to $190M, compared to only ~$16M reported in full year 2024. Analysts are projecting revenue of $477M in 2026 and $1B in 2027. They are ramping production of their proprietary Z3™ battery technology, an aqueous zinc battery system offering a non-flammable, fully recyclable, and long-duration alternative to lithium-ion, with safety and lifespan advantages suited for grid-scale and renewable integration. The reason I bring this up, re: IREN, is that these batteries can help reduce grid curtailment, which I believe is a big issue in the part of Texas where IREN is located because wind and solar energy is prevalent, but if there is no room on the grid to transport the electrons, the energy is essentially lost. I’ve been wondering how IREN was going to help ERCOT with it’s curtailment issues, and $EOSE batteries came to mind.

From Perplexity: “Demand Response and Curtailment Programs: IREN has a stated policy of participating in ERCOT demand response programs and curtailing data center loads during periods of grid stress or high market prices—turning off compute workloads automatically or reducing consumption as needed to support grid reliability. This makes AI and mining loads highly flexible and grid-supportive.”

It seems that IREN will need to either curtail workloads, which is a nonstarter, or have a battery type solution (or other source of power generation, like a turbine) so that they can draw on the batteries when ERCOT asks them to curtail loads

Note that “curtailment” has multiple meanings, from power generation sources (wind, solar, etc) being asked to curtail supply, to customers being asked to curtail usage.

Addendum: I asked @FransBakker9812 about the curtailment issue, and this is how he responded: https://x.com/FransBakker9812/status/1968668133725515851

“We have very little insights on that. The company has stated backup (diesel) generators and UPS systems. From what I found out after diving deeper, is that they plan to be able to run more than 24 hours without power from the grid. But the official disclosures are very limited in details. We don’t even know if they have a flexible or firm load connection agreement. Under SB6, they will have to comply with all the regulations and mandates when it comes to curtailment or the kill switch. There is a windfarm that interconnects with the SW1 substation, but that’s probably not something that can help $IREN, outside of a very unlikely situation of no more 345kV coming from the grid. I guess we will find out more in the coming months.”

[stonccs]

Exactly, this is a big difference between BTC mining and HPC: the latter can’t just be shut off when the power gets curtailed, because IREN will have Service Level Agreements (SLAs) with their customers (e.g. guaranteed 99,9% uptime).

That’s why they will need an Uninterruptible Power Supply (UPS), i.e. generators. These generators are of course in big demand and have lead times of 12 months or more. That’s what IREN means when they say “Procurement of long lead high voltage equipment”, like in their July monthly update.

On the other hand, this makes me appreciate competitors who are developing datacenters in the northeast instead of Texas. For example, Bitfarms (BITF) who is not as far along as IREN, and not as experienced in infrastructure development, but who is building a datacenter in Pennsylvania - an area where Google and Amazon are investing too. That’s why I took a small speculative position in BITF in addition to IREN.

[twillo]

This is the CEO of $EOSE being interviewed on CNBC on how battery solutions benefit the Grid in general, and data centers

https://x.com/bert_gilfoyle/status/1968748455678947369

[GDavenport]

Twillo,

Please take commentary about EOSE to another thread. A write up with business performance and growth drivers would be very appropriate there.

[Iamnzane]

Here is a comment posting from seekingalpha also posted on FB concerning NBIS cap rate and depreciation projections. Not all the basis and math details are there IMO. It appears NBIS has taken on convertible debt at very attractive conditions versus CRWV. As an owner of commercial NNN property, I certainly understand cap rate and I am happy with my current 5-6%. A projected NBIS 30-40% cap rate is rather obnoxious and unheard of.

Nebius’ Monster Rally Could Be Just The Warm-Up (NASDAQ:NBIS) | Seeking Alpha

====> [@TheTruthIsOutThere2021](https://seekingalpha.com/user/54165349) This is a story-line stock I love and heavily committed. Thos@TheTruthIsOutThere2021 long value investors know how to read in between the line and have a skill of mental-Math without using a spreadsheet will tell you roughly (yet with certainty, except the timing) that big profit will be there as the author laid it out for us. If you do not see what he meant, you must be new to Nebius. Go back 7 months and read all the headlines only, and then read the company latest quarter and their relevant information, this will take you a week. Then perhaps you will NOT miss the $450 days ahead of us, from $99 now, it is a x4.5 gain, not too bad.

Instead of asking general smart question (your questions and other skeptics’ questions are all the right questions to ask). But you are obliged to find the answers on your own, don’t take it from the super bulls like myself.

Give you some key points here however. (1) Author said a great terms on the notes — we have 4 notes total, will save $200M annually which CRWV has to payout due partly to their big debts carried so far. That is the saving you will see in the future recurringly up to 2032. (2) From myself, explained to others on another post, GPU leasing business can have an equivalent of 30-40% cap rate annually (using REIT’s terms, ask ai if you are not familiar). If we model the TRUE depreciation curve as well as the tax-saving schedule of 100% 1st year by OBBB or even the 5-6 years depending on what both NBIS/CRWV will elect to do to best fit their paper-loss/gain, we do not know until the CFO decide on the fly. We know it is profitable, very profitable. Skyscraper cap rate is 6-7%, residential is now 5%, Net lease commercial 8%, nothing in the world has 30-40% cap rate. (3) Some of us is talking about “800 in 28” => $800 by year 2028 … I believe it will not be that fast, perhaps $1000 by year 2032 … you can do some DD way this crazy target is possible or not, that will make you a good investor in NBIS or in other stocks.

still riding the bull

-zane

[Chamonyx]

I see lots of words, but have difficulty making any sense out of them - and I am not going to pay for SA premium to read the original article.

I would agree wholeheartedly that CRWV is much less attractive from a margin/cash flow/RoE point of view than NBIS.

However, there are some significant doubts relating to the NBIS deal with MSFT in terms of what the actual benefits to current shareholders are likely to be.

Also, NBIS has to deliver in NJ to start collecting from MSFT (who already had a deal with CRWV, which they could have extended if CRWV could have delivered on). The NJ host/builder is already making excuses for being late.

I am out of NBIS for now, until they can prove they can deliver something to shareholders.

[Buffjan2]

However, there are some significant doubts relating to the NBIS deal with MSFT in terms of what the actual benefits to current shareholders are likely to be.

Would you be willing to share a list or description of the significant doubts?

[stocknovice]

@Chamonyx

I am out of NBIS for now, until they can prove they can deliver something to shareholders.

Hi @Chamonyx. What in your opinion has management not delivered based on the timelines we’ve received thus far?

I had concerns in Q4 when they estimated $170-180M in ARR and only delivered $90M while emphasizing the pie-in-the-sky $750M-$1B ARR guide for this year. However, revenue and ARR have since ramped right on time with that guide now standing at $900M-$1.1B. And regardless of the contract terms as we know them so far, MSFT committing such a large sum lends public credibility to the underlying business.

As you say though, they must deliver in NJ. Can you post where you found that project is running behind?

[wpr101]

@Buffjan2

Would you be willing to share a list or description of the significant doubts?

It is specifically mentioned in the original post the builder is behind schedule. This was new information to me and seems like it should be relevant to Nebius investors. The statement seems borderline accurate as DataOne was scheduled to complete the first phase in September but haven’t announced anything yet. If it’s not completed in the next couple weeks they are behind schedule.