Founder of Northwise. Trying to understand technology, incentives, and where the value goes.
Joined September 2026
- Tweets97
- Following17
- Followers829
- Likes359
Pinned Tweet
Our $NBIS 2027 model and forecast might be our least viewed model for Nebius yet, but I think it's the most well done by the team and the most significant and relevant timing wise.
We are planning a massive build after Nebius earnings, where we are going to leverage this new data and model to fill in the gaps and model out 2028, 2029, and then a new 2030 model using this year by year approach in hand with our ground-up data center capacity data.
It's never been more important to understand the fundamentals of this sector and Nebius' specific circumstances.
If you follow our work and haven't yet pulled the trigger on a membership, we are running our first year anniversary sale until early Monday morning.
I highly recommend seeing if its a fit for your individual research workflow.
Members get our full price targets for 30+ names, both flagship and active growth portfolios, data center tracker, imbedded capacity built financial forecasting tools for $NBIS $IREN $CRWV $CIFR $KEEL and $WULF and support our goal of bringing a full time research team to Northwise.
Regardless, cheers to an incredible full first year of Northwise memberships and a special thanks to those who continue to support our journey to fill the gap between institutional and retail research.
northwiseproject.com/checkou…
Nebius $NBIS is nearly 50% below its all-time high, and the fear spreading through AI infrastructure is hard to ignore: what if demand keeps growing, but the companies building for it run out of money?
The pressure on $CRWV and $IREN shows how widely that fear has spread. For businesses buying expensive GPUs and renting out computing capacity, access to financing can decide whether a full order book ever becomes a good outcome for shareholders.
We think the market is underestimating how much revenue Nebius can bring online in 2027. Our base case forecasts $21.9B of revenue and $12.5B of EBITDA as new capacity enters service.
But getting there requires $58.8B of cash capex. Even with $21.6B of operating cash flow, including customer prepayments, that leaves roughly $37.2B to fund. The timing of those advances, the availability of secured debt and the price of new equity all matter.
That is where our conviction has to earn its keep. We follow the capacity through delivery, contracts and cash, then account for the debt and dilution needed to build it. Slower commissioning or worse financing terms change the outcome.
We've made the full operating analysis and the bear, base and bull results free to read. The image below puts the capacity, earnings, cash requirements and share count together so you can see what our view depends on.
The new Nebius 2027 report is in the first reply.
ALT Northwise's public Nebius 2027 operating forecast, with Bear, Base and Bull scenarios and a quarterly Base case. Revenue ranges from $19.07B to $24.61B, with $21.86B in Base. EBITDA ranges from $10.26B to $14.82B, with $12.49B in Base. Base operating cash flow is $21.60B and cash capex is $58.77B, leaving a $37.18B funding gap. The table also shows capacity, customer advances, new debt, equity and share counts. Figures are Northwise forecasts, not company guidance.
Remember $NBIS is trading for less than one time our projected 2027 ARR.
If we are too high by 50%, nebius is still trading less than two times next year’s exit.
Not many opportunities to intersect that deep of value, elite management, and an insatiable demand environment.
I am bullish on NVIDIA $NVDA, but I do not treat better AI efficiency as automatic proof that equipment demand will rise.
If the computing needed for a fixed-quality task falls to one-quarter, task volume has to quadruple just to keep aggregate computing unchanged. Double the tasks and computing falls by half. Eight times the tasks would double it.
Those are hypothetical examples, but they put a real question in front of the thesis: how much useful new work becomes economical as the price falls?
For Nebius $NBIS and IREN $IREN, utilization and the price customers will pay matter alongside task growth. I want to see new applications absorb the savings, and enough demand left over to justify more machines.
When I follow a launch for AST SpaceMobile $ASTS, I try to connect it to four separate business clocks: when cash leaves, when hardware is ready, when it enters service and when a carrier can be billed.
Those dates can be far apart. Paying for a satellite does not put it in orbit. Reaching orbit still leaves deployment, testing and commissioning. Carrier integration and commercial readiness matter before the hardware can support recurring revenue.
That timing is why a delay can hurt even if the eventual constellation works. Construction spending may continue while the associated receipts move further out.
I also separate an insurance recovery from service restoration. Cash compensation can help fund a replacement, but it cannot instantly replace the missing coverage or recover every lost month of sales.
The attached six-satellite loss case is a hypothetical stress test. I use it to examine how operational setbacks travel through the business, rather than treating a successful launch count as a complete financial forecast.
Nebius $NBIS could earn more per megawatt while customers pay less per unit of AI work.
Better hardware can make that happen. If the amount of useful work a system delivers grows faster than the price of renting it, the customer gets better economics even as the provider earns more.
The Rubin illustration in our 2027 Nebius Model shows the relationship. Using a selected NVIDIA benchmark and reference rack-power figures, we get about 59% more useful work per megawatt than GB300. Our assumed gross rental revenue per megawatt rises by about 22%.
Under those assumptions, the customer's price per unit of work falls roughly 23%, while Nebius earns more from the same power budget.
The figures are a workload-specific illustration. Actual customer workloads and negotiated prices will determine how much of the gain Nebius captures.
For Nebius, Rubin could raise the revenue potential of each secured megawatt while making it more economical for customers to expand their AI usage. Its software platform can help turn that hardware capability into billable revenue by keeping the systems busy and delivering reliable performance.
ALT Northwise exhibit, Who captures the gains from Rubin. A selected Rubin versus GB300 benchmark shows up to 3.7 times throughput. Scaling by reference rack power of 142 versus 330 kW produces an illustrative 1.592 times work per rack IT megawatt. Northwise's assumed gross rental yields of $55M versus $45M per MW give a 1.222 times revenue ratio and a 0.768 times implied price per unit of work. Actual value capture depends on customer workloads, provider execution and contract negotiations. This is conditional arithmetic using reference power, not measured benchmark energy use or an observed rental price.
Tyler | Northwise retweeted
$NBIS's Independence campus could become a foundation for years of US growth, giving Nebius more control over where and how it adds compute.
The site is planned for up to 1.2 GW across four buildings. Owning that footprint lets Nebius plan power, cooling and later expansions together. The buildings and electrical infrastructure can support several generations of GPUs, giving it a place to expand customer capacity over many years.
For shareholders, the opportunity is a larger business built on infrastructure Nebius controls. Ownership also puts more of the construction and financing burden on Nebius. The return depends on getting power, computing equipment and customers into place together. Delays leave capital tied up longer before it earns revenue.
That makes the power arrangements central to the story. The utility supply, off-site Blue Valley generation and future on-site generation are separate parts of the plan. A power-station permit is progress, but it does not establish operating compute.
The 1.2 GW is ultimate planned capacity. Our review did not verify current active IT MW. Nebius's May 2026 guidance pointed to Missouri operations in 2027, while its four-building plan spans five to seven years. Confirmed power delivery, commissioned capacity and customer activation will show when that long-term opportunity starts contributing to the business.
Our refreshed Independence report explains the full site, its power arrangements and the path ahead, with new diagrams and direct tracker and model connections.
🤖 Made with AI
ALT Conceptual diagram of planned power pathways for Nebius's Independence campus. The off-site Blue Valley / Independence Power Partners generation project connects through the SPP grid and Independence Power & Light utility supply. An arrow labeled 800 MW utility plan leads to the campus, labeled up to 1.2 GW planned. A separate dashed arrow shows a future 400 MW on-site generation option under evaluation. Planned power is not current operating capacity. The illustration is conceptual, not a photograph or as-built site plan.
I think the Street is underestimating the size of the business Nebius $NBIS could be operating by the end of 2027.
Our forecast is $21.9B of 2027 revenue, roughly 79% above the $12.2B consensus used in our report. We also see $42.6B of owned-cloud exit ARR, rising to a $44.2B platform run rate when net partner revenue is included.
Here is what gives me conviction in numbers that far apart.
In its Q2 shareholder letter, Nebius said: "We could sell our entire 2027 capacity on these terms today." It was deliberately retaining capacity for customers willing to pay more for immediate access. The same letter reported landmark deals at $20M-$25M of annual revenue per MW and a $40M-$50M opportunity for short-duration capacity, with the first such deal already signed.
At GTC Berlin, Arkady Volozh said demand was not the constraint and that he wanted at least half of capacity serving those shorter-duration needs. That commercial mix matters enormously. The revenue potential of a megawatt changes with the hardware inside it and the contract attached to it.
Then there is the physical expansion. Management plans to deploy more than 1 GW a year starting in 2027. The Berlin discussion specifically addressed 1 GW of Vera Rubin, with Volozh describing a ramp weighted toward Q3 and Q4. Pennsylvania provides a concrete site milestone: an initial 260 MW phase targeted for October 2027.
We translate that evidence into 1 GW of new Rubin installations, another 250 MW of Blackwell, and an existing fleet that continues earning. Those are our assumptions. We allow time for commissioning, price each hardware generation separately, honor signed contracts and allocate roughly half of incremental accepted capacity to shorter-duration customers.
That leaves about 1.61 GW accepted at year end, but only 938 MW billing on average during 2027. This timing is why the exit ARR looks so large relative to annual revenue. A cluster arriving late in the year contributes little to that year's sales while contributing fully to December's annualized pace.
My reading is that consensus implies a smaller productive fleet, lower realized pricing, later delivery, or some combination of the three. I think the capacity plans and signed deal economics support a much larger outcome.
Nebius $NBIS is nearly 50% below its all-time high, and the fear spreading through AI infrastructure is hard to ignore: what if demand keeps growing, but the companies building for it run out of money?
The pressure on $CRWV and $IREN shows how widely that fear has spread. For businesses buying expensive GPUs and renting out computing capacity, access to financing can decide whether a full order book ever becomes a good outcome for shareholders.
We think the market is underestimating how much revenue Nebius can bring online in 2027. Our base case forecasts $21.9B of revenue and $12.5B of EBITDA as new capacity enters service.
But getting there requires $58.8B of cash capex. Even with $21.6B of operating cash flow, including customer prepayments, that leaves roughly $37.2B to fund. The timing of those advances, the availability of secured debt and the price of new equity all matter.
That is where our conviction has to earn its keep. We follow the capacity through delivery, contracts and cash, then account for the debt and dilution needed to build it. Slower commissioning or worse financing terms change the outcome.
We've made the full operating analysis and the bear, base and bull results free to read. The image below puts the capacity, earnings, cash requirements and share count together so you can see what our view depends on.
The new Nebius 2027 report is in the first reply.
ALT Northwise's public Nebius 2027 operating forecast, with Bear, Base and Bull scenarios and a quarterly Base case. Revenue ranges from $19.07B to $24.61B, with $21.86B in Base. EBITDA ranges from $10.26B to $14.82B, with $12.49B in Base. Base operating cash flow is $21.60B and cash capex is $58.77B, leaving a $37.18B funding gap. The table also shows capacity, customer advances, new debt, equity and share counts. Figures are Northwise forecasts, not company guidance.
I think it's very important to start distinguishing vocabulary between what's known as the neocloud sector.
The mistake many are still making is mixing in together every company that buys, builds, or hosts GPUs and datacenters.
There are significant differences between players like $NBIS that are pulling together full stack software and hardware cloud and managed inference services and companies that are simply providing bare metal, or even some that are just accumulating energy.
There are a number of subcategories within this space that need to start being broken down; to be transparent the moats are completely different, as will be there access to financing through any squeeze in capital markets.
Will be posting more on this soon. @SMASIMHO has done a great job of pushing more individual language when it comes to Nebius in particular.
As we are seeing with the basket selling across the sector and even confusion surrounding posts by @elonmusk , it's never been more important to educate both retail and institutional investors (many of which are STILL just taking revenue and guessing on a multiplier to model out this buildout).
We have always believed, at Northwise, that not all of these players will make it and that there will be a period of consolidation. This can be seen in every sector to have ever existed. There's nothing wrong or unusual about it.
During every sectorial boom there is VC like period where investors bet on a quantity of horses rather than the pack leaders. Every time there is a shift, once investors begin to realize that there are clear quality leaders.
This Short term period of anxiety surrounding the markets is a phenomenal time to reevaluate our approach as independent viewers of the sector.
You'll see shortly how massive the capex is going to be in 2027-2028 in our eyes.
It's our responsibility to better educate why companies like Nebius are not going to struggle to raise tens of billions, but also acknowledge this is the exception not the indefinite norm.
Tyler | Northwise retweeted
$NBIS has reached a major milestone at Vineland, reporting delivery of all capacity tranches under its Microsoft agreement. The focus now turns to how the wider campus develops around that customer commitment.
The latest developments put power at the centre of that question. New Jersey's enforcement action and the subsequent notice of intent to sue bring permitting into sharper focus. DataOne, the facility's owner and operator, disputes the generator-permitting determination and says it plans to seek permits while transitioning to fuel cells.
For us, the next meaningful signs of progress are resolution of the permitting issues, confirmation of permanent power and additional capacity entering service. Those milestones will help us assess the path toward Nebius's wider 2027 campus target.
We've refreshed our Vineland report to explain what has changed and what comes next, with direct connections to the Data Center Tracker, Nebius hub and existing model.
ALT Northwise conceptual illustration for its Vineland report, with the Nebius mark over stylized energy and data center infrastructure. This is an illustration rather than a photograph of current site conditions.
A single price target hides much of what matters about $IREN.
Our report's $108.78 probability-weighted December 2030 value combines different delivery, customer and financing outcomes. Some paths work well on one dimension and poorly on another. We also include funding failures that can leave shareholders with nothing.
I find that range more useful than assuming every part of a bull case arrives together. It shows what the investment needs to go right and where a good operating story can still disappoint an owner.
The probabilities are our judgments, not measured odds. The full report, including the scenarios, is free to read:
northwiseproject.com/researc…
ALT A wide outcome set, including failure. 27 continuing paths plus their funding-failure branches. Failure has 2.83% total weight and can wipe out equity. Probabilities are analyst judgments.
When I think about NVIDIA $NVDA's inference opportunity, I start with the work someone wants an AI system to finish.
A single task can require the model to read context, generate a response, call a tool, inspect the result and try again. The tool's own execution is separate from the model calls around it. Caching can also reduce repeated work.
That makes a count of user prompts a weak measure of the computing behind an AI product. A more capable service can use more inference per task while still saving the customer time or money.
I find that a more useful way to think about demand for NVIDIA's systems and the capacity sold by Nebius $NBIS and IREN $IREN. The economic question is whether the completed task is valuable enough to pay for all the work needed to finish it.
ALT A finite AI task can involve repeated model calls around external tool use. Context processing, reusable attention state and response generation have different roles. Tool execution is separate from another model inference call. The diagram is conceptual, not a hardware wiring plan.
Our initial results from our $NBIS 2027 modeling are pretty shocking. We really think the street is getting this wrong by one of the largest margins we've ever seen.
Pulling in a lot of recently announced colocation and greenfield capacity.
2027 and 2028 demand environment is looking very certain with the larger owned sites beginning to come online in chunks.
A lot of commentary from GTC Berlin is changing how we look at and model near term and long term capacity allocation.
We think there are real cases where prepayments could pay almost 100% in GPU cost in exchange for lower end of the spectrum revenue per MW pricing (still very high in 6-12 month contracts).
Don't underestimate the importance of this pricing power as capital needs are set to explode higher over the coming years.
Additionally, in light of recent price action and to celebrate the first full year of Northwise Premium and memberships we are considering a sale for new signups.
Members have full access to all of our price targets across the AI infra buildout, our full downloadable models, our full data center tracker features and integrated modeling, as well as both of our portfolios (flagship and newer active growth strategy)
Let me know if there's an interest in this and we will make it happen.
Expect an update and new report soon on Nebius in 2027 before Q3 earnings.
A higher earnings-per-share forecast can come with less cash left in the company. I want both numbers in front of me when judging a buyback.
For Celsius $CELH, Northwise's 2030 operating forecast produces $3.69 in normalized common EPS with the assumed repurchases, or $3.50 if additional buybacks stop from 2027. The second path finishes with about $1.46 billion more cash. The underlying operating forecast is the same.
That makes the lower EPS much less troubling to me than it looks on its own. The company has retained money it otherwise would have spent buying shares. Whether repurchases are the better choice depends on the price paid and what else that cash could earn.
These future buybacks still require board authorizations. I would judge the operating progress first, then ask whether management is using the resulting cash well.
Full Northwise analysis: northwiseproject.com/researc…
ALT Two Northwise 2030 forecasts with the same operating business. Modeled repurchases give $3.69 normalized common EPS, 229.0 million closing common shares and $1.81 billion unrestricted cash. No additional buybacks from 2027 gives $3.50 EPS, 253.6 million shares and $3.27 billion cash. Both retain forecast second-half 2026 repurchases and a $441.8 million analyst operating reserve, which is not legally restricted cash. Future authorizations are assumptions.
Customer prepayments help $IREN build sooner. They also give customers a claim on future service.
When that service is delivered, part of the invoice is settled with credits instead of fresh cash. The money arrived earlier and helped pay for equipment.
That is why I subtract the remaining prepaid-service obligation when moving from business value to shareholder value. Debt has to be accounted for too, before dividing the equity claim across the shares.
The chart shows that bridge for the ARR method in our Base 2030 case. It is one valuation method within the report, rather than the final blended target. The business can become enormous while the claim belonging to each share remains much smaller.
ALT From enterprise value to the equity claim. Base 2030 ARR method, $B. $88.87B / 1,183.53M shares = $75.09 per share. Convertibles are tested at each method value; blended value is $138.02.
The legal-settlement section of the Meta $META report is a good example of how a few large numbers can give the wrong impression when added together.
The August agreement described roughly $12.7B of guaranteed payments, $5.3B of conditional payments and an expected $10B accounting charge. The charge is not another $10B cash payment on top of the settlement.
I would follow three schedules: when the expense enters earnings, when cash is due and when the required product changes take effect. Each answers a different question about the business.
A known settlement makes part of the cost easier to forecast. It still consumes cash, and it does not remove exposure to future legal events. That is how the October 9 model treats it.
ALT Meta's August 2026 settlement, as described in the October 9 Northwise report. Earnings: approximately $10 billion expected Q3 accounting charge. Cash: a ten-year schedule with $12.7 billion guaranteed and $5.3 billion conditional payments. Product: ongoing teen-account obligations. The accounting charge is not an additional cash-payment category.
I have been doing some thinking on $NBIS management commentary coming out of GTC Berlin this week.
I think many are hyper focused on the announcement that Nebius plans to bring online a GW of Vera Rubin capacity in 2027. (this is obviously very exciting news for the company.
However, I see it as highly unlikely that this is the only capacity Nebius will be bringing online with the pricing power that other chips are still retaining and with demand remaining insatiable across the board.
Nebius not only appears ready to shatter 2027 ARR and capex expectations, but also potentially capacity expectations as well.
If AST SpaceMobile $ASTS eventually delivers 99% availability in a market, I would still want to know what happens during the missing 1%.
Consider two hypothetical networks. One loses service for a single 14-minute stretch each day. The other drops out 48 times for 18 seconds. Both have approximately 99% daily availability, but they can feel very different when someone is trying to finish a call or load directions.
This is why I separate the existence of satellite coverage from the continuity of a usable session. Satellite geometry, signal quality and handoffs can all affect what the customer experiences. The 99% example is an illustration, not an AST performance claim.
I see real value in filling coverage gaps with an ordinary phone. My confidence in recurring customer demand will rise as evidence shows which everyday tasks the service can reliably complete, across whole sessions and under realistic conditions.
Could $IREN lease its powered buildings to another operator if owning the compute became less attractive?
It is a useful option, but its value depends on when the decision is made.
In our comparison of one 100 MW Sweetwater slice at 2029 costs, owned compute earns a 15.6% project return. A landlord lease at the assumed rent earns 7%. Once the shell is already paid for, leasing looks better as a decision about the remaining capital.
That distinction matters. A sensible way to use an existing building does not establish that the original construction investment earned enough, or create a guaranteed floor under the shares.
ALT Own the compute or lease the capacity?. One 100 gross MW Sweetwater slice on 2029 costs. Landlord: $2M rent per IT MW, 80% NOI margin, 10-year lease. Not a company valuation floor.
The Pepsi partnership helps Celsius $CELH get its energy drinks into stores across the country. When I think about what that reach is worth to a common shareholder, I have to follow the earnings through the partnership too.
Pepsi owns preferred shares that receive a regular dividend and also participate in earnings before conversion into common stock. A larger consolidated profit therefore has to be allocated before calculating what belongs to ordinary shareholders.
In Northwise's 2030 forecast, about $1.05 billion of normalized net income leaves roughly $866 million for common shareholders after those preferred claims. The participation amount is an earnings allocation, not a second cash dividend.
I remain interested in the growth opportunity after allowing for those claims. This is also why I want the ownership terms beside the revenue forecast: they determine how much of a successful business reaches each share.
ALT Northwise central 2030 normalized earnings forecast. Consolidated net income of $1,045.83 million less $56.75 million regular preferred dividend and $122.97 million participating preferred earnings leaves $866.11 million for common holders. Dividing by 234.58 million common and employee claims gives $3.69 EPS. Preferred participation is an earnings allocation, not an additional cash dividend. It must not also be deducted a second time through the denominator.
Tyler | Northwise retweeted
Remember that we have continued to reiterate our view that $NBIS ARR will likely come in double current street estimates (maybe more)
Price action over these weeks has nothing to do with the current demand environment or continued rapid execution from management.
Due to popular request to bring our 2030 models in closer before earnings, we will be dropping our full year 2027 Nebius model and targets soon.
We were the first to view hyperscaler deals as anchor contracts for financing and that Nebius should be modelled through a capacity first ground up view point.
If you follow street estimates and models, the vast majority now use that same framework to model Nebius and they continue to adjust their near and long term capacity forecasts to the numbers we were publishing as early last year.
We may not have the distribution of some of the big banks, but we have thousands of pages of reports and thousands of hours of research that is now aiding the street to catch up.
Most of these analysts reactively adjust their models to replicate short term price action, meanwhile our forecasts have only been revised up as Arkady and the team continue to execute at a higher level that we could have imagined.
We aren’t bothered by corrections, we research and display our conviction publicly, even as our targets grow farther apart from near term valuations.
Despite recent market volatility and the AI selloff, $NBIS remains my highest-conviction investment and is once again the largest holding in our portfolios.
It’s rare to find a generational opportunity at the intersection of seemingly insatiable demand, low-cost financing, and management execution that continues accelerating despite the company’s growing scale.
The Street expects roughly $16B in exit ARR for 2027. Following Q2, our full-company model projected approximately $31B. I increasingly believe even that may prove too low.
Management has discussed short-term capacity deals approaching $50M in annualized revenue per MW. If Nebius can capture more of this demand through short-term auctions rather than locking capacity into lower-priced multiyear agreements, revenue per MW could meaningfully exceed our assumptions.
Combined with accelerating capacity deployment, $NBIS could potentially exit 2027 at more than 2x current Street expectations.
I’ve been deploying significant cash from our flagship and active Growth portfolios during this selloff.
Looking beyond 2027, our models suggest $NBIS could surpass $100B in annual revenue by 2030. I increasingly believe even that extraordinary forecast could be exceeded.
Two developments are changing my expectations.
First, Vera Rubin could materially increase the revenue potential of newly deployed capacity, particularly in short- and medium-term contracts. Alongside Nebius’s expanding software stack, including Token Factory and zero-day model support, the company is moving further beyond commodity GPU rentals.
Second, capacity acquisition is accelerating. Much of our original forecast depended on greenfield developments across the US and Europe coming online in 2028 and 2029. Management’s hybrid approach, incorporating brownfield acquisitions, colocation, and faster near-term deployments, could pull substantial revenue forward.
Higher revenue per MW combined with faster capacity deployment creates enormous upside to our existing scenarios. New management announcements continue strengthening the probability of our bull case.
Of course, none of this comes cheaply. Our CapEx forecasts are also substantially above Street expectations. But unlike the financing concerns recently surrounding $CRWV , Nebius benefits from a strong cash position, attractive debt terms, and significant flexibility across debt, convertibles, equity, and potential subsidiary monetizations.
Management has demonstrated an impressive ability to balance these financing sources without compromising growth.
I suspect the upcoming Q3 earnings could be a major catalyst if management provides formal 2027 ARR and CapEx guidance. Both figures could catch Wall Street badly off guard.
The market continues to underestimate not only the scale of infrastructure Nebius is building, but also the increasingly valuable software platform sitting on top of it.
$NBIS remains my highest-conviction long-term holding, and recent volatility has only strengthened that conviction.