Daniel Koss (@daniel_koss) on Surfaced

Daniel Koss (@daniel_koss) on Surfaced

Recent calls

29 calls

@daniel_kossJun 4, 2026 · 5:31 AM

Great job highlighting this @zen_tropy This is what I wrote multiple times. A good project manager plans to be ahead of schedule to be on schedule. Just like $NBIS plans to build so many data centers that even with delays and even with things going wrong, they are still ahead of guidance. That's the difference between companies who turn every tiny and normal step into a press release and Nebius management who casually adds another 1.2GW like it's a normal day of the week. As a shareholder it's generally good to invest in stocks where management pushes back against high expectations and then crushes them anyway. Great performance often comes down to reality > expectations

@daniel_kossJun 3, 2026 · 4:55 PM

Another indirect Claude x $NBIS connection! Nebius is the primary AI cloud infrastructure provider and co-engineering collaborator for Higgsfield AI. Higgsfield relies heavily on Nebius's specialized hardware stacks to train and run its massive generative video models. Nebius <-> Higgsfield <-> Claude / Anthropic Nebius <-> ClickHouse <-> Claude / Anthropic Again, I know of no deal and I'm 100% speculating here. I don't want investors to get the wrong expecatations. But I do want to think out loud... If Anthropic likes all these companies that are incubated inside of Nebius or run on Nebius, isn't this a very clear indicator that the two engineering teams certainly know each other, talk to each other and maaaaybe like working together? At the very least the first 2 things should be true... SF is a small bubble :)

@daniel_kossJun 1, 2026 · 10:56 AM

I know many people say the risk/reward for $NBIS here is not as good as it used to be, but I think that is a clear case of playing Captain Hindsight. Everyone now knows Nebius is a massive AI cloud winner long term. The only real question now is how big AI will be, how much market share Nebius can get, and how profitable the business will become. Do you really think that is a worse risk/reward than when people were wondering whether the business would ever be profitable, or whether Nebius could even get capacity? I strongly disagree with the idea that either the ceiling is close or the risk is now higher. It is the exact opposite, and that take just shows a lack of understanding of both the TAM of AI and what the real risks used to be. The only thing I will say is: yes, Nebius will not 10x this year. But a 2-3x in 12 months is totally possible. One thing @DeepValueBagger often mentions that I really like, even though I see it slightly differently, is that he does not look at price targets because they create anchoring bias. This is clearly happening on X. People remember when the stock was $90, and now they just turn off all their brain cells and wait for it to get back there. That is a bit like stopping your life because you are waiting for your high school crush to come back, while somehow assuming you would not fuck it up this time ;)

@daniel_kossMay 31, 2026 · 2:00 PM

Why @GavinSBaker thinks there might not be an AI bubble. $NVDA sees the demand. $TSM ok buddy, you get +5% hehe I think: Demand > Supply (for at least a decade) Image from my coming $NBIS deep dive. Will be live on Substack tomorrow (Monday). https://t.co/ERQSjFWpbW

@daniel_kossMay 30, 2026 · 12:28 PM

Preview of my coming "Should You Buy $NBIS" article on Substack. What the Company Does (Part 1/13) Today, Nebius is building a vertically integrated AI cloud. But I don’t like this description, because I believe it does not properly explain the insane potential I see in Nebius. Here’s how I see it: Nebius gives customers compute at the highest useful level of abstraction they can handle. Too much abstraction and the customer cannot use it properly. Too little abstraction and the customer is forced to do unnecessary work. The more a customer can do in-house, the less Nebius has to do. The less a customer can do in-house, the more Nebius abstracts away, and the more value Nebius can capture. Win-win. For highly technical customers like hyperscalers, such as Meta and Microsoft, that can mean bare metal GPUs. For the long-term majority of customers, especially enterprises, it can mean managed training, inference, or tokens through Token Factory. Over time, I think this abstraction keeps moving higher. GPUs as a service → tokens as a service → end-to-end agents → results as a service. Eventually, when agents in the digital world and robots in the physical world are reliable, scalable, and can deliver high-quality work end to end, Nebius might sell what I think of as “results as a service.” This rhymes with Palantir’s view of the future of software. Palantir’s view is that AI should not just be a tool the customer uses. It should turn into a measurable stream of high-value finished goods and services. That is exactly how I think about Nebius long term. The customer should not care about GPUs. The customer should not care about tokens. Eventually, the customer may not even care about agents. The customer pays money and gets the solution to the work that needs to be done, without touching infrastructure, hardware, APIs, or any of the technical complexity underneath. This is not just my long-term view on Nebius. It is my long-term worldview on AI investing in general. In the short term, of course, GPUs and tokens remain highly relevant. That is where the revenue is today. But the long-term direction is abstraction.

@daniel_kossMay 29, 2026 · 8:44 AM

The main bottleneck in AI is the team! A supercycle of this magnitude will create infinite new bottlenecks. The biggest long term winners will not be the companies that own the most MW today. It will be the companies with the best "machine that creates the machines". that machine = the people Do NOT ask "how much power do you have"? Ask: How much power can you add? Who can build fastest? Who can finance cheapest? Who can deploy GPUs most efficiently? Who can turn capex into ARR at the highest rate? Who has the best strategy? Who can increase $ / MW by adding value? That is why I bet on strategy + leadership + track record of execution + a constant pattern of beat-and-raise expectations. For a long time I've been the loudest and most bullish person on X talking about $NBIS with very high ARR targets. My ARR expectation of $5B - $7B for 2026 was seen as "pumping" and "delusional" in 2025. Then they said $7B - $10B in their guidance 😂 This company will always deliver long-term. All setbacks ar hiccups and every dip is a buy. Not financial advice. As always very biased ;) Long $NBIS

@daniel_kossMay 27, 2026 · 7:55 PM

@ImStereotypic honestly, yes. but people don't understand that you can be long-term bullish a stock and trade it on the way up, expecting pullbacks. they think bullish posts + sell = pump and dump. even though I bought and sold my favorite stock $NBIS like 100 times. never does this imply i'm not bullish, always does this simply represent a local top expectation and a view on timing. anyway. i won't even try to explain it again. it simply doesn't land.

@daniel_kossMay 26, 2026 · 5:56 PM

In case anyone wonders why I record so many interviews with $NBIS executives (spoiler: more coming)... It's really simple. I'm 100% convinced $NBIS is the AWS of AI and the next hyperscaler. If you could interview the leaders of AWS 10 years ago, wouldn't you want to learn as much as you could from them? They're brilliant, really kind people and what might be underrated: they work freaking hard! All these headlines we get to enjoy as investors, all of that is only because they spend months grinding and making shit happen. I promise you guys: $NBIS x Physical AI will be massive. Soon... People once again sleeping on what they are cooking. While I do publish some interviews (with their permission), I also talk to many leaders and smart people that work in AI at many of the world's leading companies in my free time (honestly where's the difference). I'm insanely happy with how well investing in AI pays, but at the end of the day, enough money is enough money. My true addiction is and has always been learning. Knowing what's going on in the world and where we're going is so freaking exciting. Everything I do is really just a lazy excuse to learn every day and somehow have people not realize I'm still just a little kid following his curiousity every day.

@daniel_kossMay 22, 2026 · 1:38 PM

$FCEL trades at only 9.1% of $BE Bloom Energy’s valuation per gigawatt of guided 2027 annualized capacity. If FuelCell got Bloom’s multiple, it would be ~$15.3B, or an ~11x from today. To be clear: this will almost certainly not happen. Bloom is the better company and deserves a higher valuation. Period. Better product. More efficient. More proven. Founder-led. Better revenue visibility. Better margins. Better execution history. But does that justify Bloom trading at ~11x FuelCell’s valuation per GW of capacity? That’s the question. Even at 25% of Bloom’s multiple, FCEL would be ~$3.8B, more than a 2.5x from today. At 50%, it’s ~$7.7B, more than a 5x. None of this requires FCEL to be as good as Bloom. It only requires the market to stop pricing FuelCell like it’s doomed. “Market cap per guided capacity” isn’t weird. For companies like this, long-term value is about annual capacity, revenue per MW, margin at scale, and dilution. Demand isn’t theoretical: Nebius $NBIS just raised GPU pricing ~30% overnight. That’s the compute supply crunch. And power is the constraint behind every GPU. FuelCell guides to 350 MW by end of 2027, aiming for 1 GW/year. In the AI power crunch, 350 MW is tiny. Every competitive, scalable fuel cell solution will sell out. FuelCell doesn’t need to be Bloom. It needs to be good enough and available 😂👌🏻 If FCEL sells out capacity, we could see extreme revenue growth, margin expansion, operating leverage, and a re-rating. Maybe FCEL deserves a discount. But 90%+? That’s too much. The market prices FCEL like it’s doomed. If they sell out capacity and become a real AI power supplier, the upside/downside is huge. That’s the bet.

@daniel_kossMay 20, 2026 · 2:39 PM

Let's goooo!!! Bloom Energy x Nebius deal done! Worth up to $2.6B Freaking huge for Bloom $BE Bloom’s 2026 revenue guidance is $3.1B to $3.3B This one single Nebius deal is worth roughly 79% to 84% of Bloom’s entire expected 2026 revenue😂 Nebius will pay monthly service fees over 10 years, while Bloom installs, operates, and maintains the systems. 250MW to 328MW -> I think the Vineland NJ data center is roughly ~300MW, coincidence? I noticed during the $NBIS earnings call that the team kinda avoided the Vineland question. They only said the tranches so far were delivered on schedule, but I don't think they ever confirmed it's on target for end of 2026? This is a major milestone to unblock the biggest bottleneck and social pushback - the air permits and social pushback! Funniest part is, DataOne CEO CAB had this in his bio the ENTIRE TIME😆

@daniel_kossMay 14, 2026 · 1:24 PM

I sold $CGEH I wanted to be 100% in $NBIS into earnings. But now after earnings, I will still not buy $CGEH back. Here's why & what I found while researching them. I want to be very clear upfront: This is not a claim that Capstone is a bad company, fraudulent, or doomed. It is simply my own investment process, based on the information I had at the time, followed by additional diligence that changed my risk/reward view significantly. Here is the sequence: I became increasingly focused on the AI data center power bottleneck. The headline that “30% to 50% of US data centers are delayed or cancelled” made me think much harder about whether companies like $NBIS could run into capacity issues. In that context, I started looking for companies that could help solve the “power now” problem. The obvious name is $BE. I wanted to find alternatives. If there is another credible, scalable, modular, fast-deploying on-site power solution for data centers, the upside could be very large. That search led me to $CGEH. I read the available bullish writeups on Substack and X and thought the story was interesting enough to take a position while continuing to research it. After buying, I found several things that made me uncomfortable. First, some of the strongest bullish content came from anonymous accounts. I have no issue with anonymity itself, but when I tried to pressure-test the thesis directly, I did not get the level of transparency I would want. I'll leave it at that. Second, I received feedback from another investor who had spoken with management. Their impression was negative. I will not share private details or name anyone, but the takeaway reduced my confidence significantly. Third, I spoke with people closer to actual data center development, and the feedback I received was that Capstone’s current solution may not be as scalable or efficient for large AI data centers as the bullish thesis assumes. None of this proves that the thesis is wrong. But it changed the risk/reward for me. My original upside case was simple: If Capstone can become a real scalable alternative to $BE for AI data center power, then the stock could have massive upside. I still think that is theoretically true. But after doing more diligence, I no longer had enough confidence in the probability of that outcome. So I sold. I was conflicted about sharing my view publicly, because I generally want to avoid talking negatively about companies. When I sell it often just means I found a better opportunity / better risk-reward elsewhere. But since so many people asked, I changed my mind and think I should share these thoughts. Again, please make your own opinion. Don't just buy or sell based on a random guy on X thinks. Maybe I'm totally wrong and just paperhanded a true 10x opportunity.

@daniel_kossMay 13, 2026 · 1:47 PM

VERY important bit from the $NBIS call and the MAIN reason why they increase capex, don't ignore this! "we are building for 2027 demand, where we have customer commitments ALREADY increase" so the additional ~$5B in capex in 2026 is DIRECTLY tied to additional ARR in 2027! This is NOT the same as costs increasing. It's insanely bullish and translates to high ROI making them want to invest more aggressively and at low risk, because demand is there. If these are big partners like $MSFT or $META then this will also mean MORE pre-payments, because of the smart financing structure they are using. Will break that down one of the next posts when we get there.

@daniel_kossMay 12, 2026 · 9:59 AM

Huge day for all $NBIS investors! Tomorrow, Nebius reports earnings. Because they report at 2pm CET, all Nebius investors need to decide today whether they want to hold, sell, or buy more into earnings. Unless you want to trade pre-market, obviously. The two most important data points we will learn: Monetization How well are they monetizing their capacity? In simple terms: How many million dollars of revenue are they generating per MW of monetized capacity? Capacity expansion This is currently the much more important variable in my opinion. How is their capacity expansion going? For the second point, I’ve got you. Today before market open, I’ll post a fresh interview with @CABeyney , CEO of DataOne, the company responsible for building the 300 MW data center in Vineland, New Jersey for Nebius. I think this one will be interesting not just for $NBIS investors, but also for anyone investing in data centers and AI infrastructure.

@daniel_kossMay 11, 2026 · 4:17 PM

@Yeah_Dave 100% spot on The issue is not Cerebras it's $NBIS being undervalued.

@daniel_kossMay 8, 2026 · 10:44 AM

Recorded the podcast with @CABeyney, CEO of DataOne, and @itsalasdairmann. They are building a 300MW+ data center for $NBIS, potentially 1GW+ in the future! Will share my thoughts, insights, and ideas with subscribers today. Overall, my impression is that building data centers in 2026 is insanely hard and there are a million physical-world challenges. Charles: "You can have all the GPUs in the world, but if you miss one genset, the data center does not work. If helium is missing in the GPU process, you cannot print GPUs. One broken brick breaks the whole chain." Challenges include: - community opposition - water usage - electricity bills - pollution - noise - environmental impact - general fear around AI infra But here's the thing: the best builders of data centers actually have very strong answers and solutions for those issues. They innovate quickly and confront these issues to find sustainable solutions. It honestly sounds like the real challenge is that many of these "protesters" are intentionally spreading misinformation to get projects delayed, because they are paid by people on Wall Street who have massive short positions. It's literally paid FUD with the goal of creating real economic damage. Charles was incredibly high energy, smart, and charming. I think you guys will get a ton of insights from him!

@daniel_kossMay 7, 2026 · 6:20 AM

Accurate 💯 If you didn't outperform just holding $NBIS (adjusted for taxes!) you should just hold and only time your entries, not your selling. I know this sounds like a double standard coming from me cause I do trade Nebius, but I've strongly outperformed just holding the stock so far and feel like that tells me I should keep doing it. But I constantly monitor risk, catalyst etc. behind the scenes. I don't just randomly buy and sell cause the stock is up or down!

@daniel_kossMay 5, 2026 · 11:24 AM

@tomsirois I have a private portfolio where I go full degen and trade. Then there's the Autopilot public profile where I'm way more risk-averse (because people copy it) and just buy, hold and slightly adjust positions to optimize. I always say my recommendation for most people is to just hold $NBIS and that's what I'm doing in the Autopilot portfolio.

@daniel_kossMay 4, 2026 · 3:46 PM

Added a new 30% position today! Been obsessed with this company the entire weekend. Listened to earnings calls and industry reports. Think it can 10x easily. Haven't been this excited about the risk reward of a stock since I found $NBIS at $40. Dropped the ticker for Subs, but will share the full thesis for free tomorrow. I think it's another name that you'll soon see EVERYONE on X talking about. Just like nobody owned $OUST and now everyone loves it (rightfully so!).

@daniel_kossMay 1, 2026 · 4:43 PM

1 day in AI really does feel like 1 year. Just 4 days ago I published this $NBIS valuation article. Read the comments. People said my $ / MW assumptions are unrealistic. I thought it would take months or quarters for people to realize this is indeed where we're going. It took 4 days and I think it's already becoming obvious? Anthropic growing from $30B to $44B and the only limit is compute capacity. Please read the article (or at least the fitting chapters on capacity growth and $ / MW) again. I keep repeating how humans don't understand exponential growth. The most ironic outcome would be, if I understood the concept, but then guessed way too low in a few years, because I too, underestimated exponential growth. See you guys in 2030 where you'll either call me a visionary or an absolute retard 😂 Either way, I'm happy to transparently think out loud, to me that's what makes this so fun.

@daniel_kossApr 29, 2026 · 6:38 PM

Moved my last 10% cash into $NBIS today at $136 and have now reached degen levels of concentration again. Nice! 77% $NBIS 8% Towa (6315) 8% $HIMS 7% $OUST If you go back and read my portfolio update where I trimmed Nebius from 75% to 20% (at the exact top) I told you that this was my plan. Wait for the stock to cool down. Slowly add it back and keep enjoying the ride up. The word needs more compute and Nebius delivers a world class cloud experience and crushes benchmarks. I believe it is by far the best positioned company to profit long-term from this environment, because they optimize the value of every MW of capacity through an increasingly distinguished and complete compute product that meets customers at the level of complexity they want and need! In a world where many companies have no moat and either grow like crazy or make a little money, I believe Nebius is building a cash printing business, while growing 600-800% yoy and all of that while establishing strong moats and a defensible, recurring revenue business that is loved by many and soon will be loved by even more customers. Customers and partners love working with Nebius and literally can't get enough compute. They also appreciate how Nebius helps them maximize performance and optimize efficiency. Nebius will take care of you.

@daniel_kossApr 28, 2026 · 4:46 PM

this guy explained it perfectly. also why does anthropic now suck? NOT ENOUGH COMPUTE entering a world where compute is scarce is good for $NBIS so obviously the stock is *checks notes* down -8% 😂

@daniel_kossApr 27, 2026 · 2:48 PM

My $NBIS 2030 valuation for those who haven't seen it yet! Thank you for all the great feedback so far! Reactions are funny. "I find all of your estimates widely unrealistic and optimistic." but also things like "20M/MW is also pretty conservative. Personally I’d do 30M/MW" These are my personal assumptions. Feel free to disagree, as I said many times. Just one thing I want to add: this entire model prices in exactly 0$ in additional value add products and services on top of their infrastructure offering. The people who think my model is absolutely crazy don't really have enough imagination to consider the insane upselling potential here. Just for context, AWS makes roughly 50% of their revenue from services ON TOP of their infra offering. Guess where the margin is biggest. Just putting that out there.

@daniel_kossApr 26, 2026 · 11:10 AM

$NBIS with another amazing hire, this time from $CRWV! I believe a lot of the big CoreWeave customers want to a) compare CoreWeave with other providers b) become less dependent on CoreWeave (for reliability but also negotiating power) c) simply maximize how much capacity they can get in total So I wouldn't be surprise if many CoreWeave clients will announce deals with Nebius in the near future. If we could compare how much Nebius vs. CoreWeave will get in terms of $ / MW from the SAME clients on NEW deals, that would be incredibly interesting! But which ones are the most interesting clients in general? Anthropic (Claude): Massive CoreWeave client. Growing insanely fast. Probably going public at $1T valuation. Insanely compute constrained. Product getting unusable because compute is so limited. Service has become slow and expensive. They NEED compute, ideally yesterday. Nebius would be a perfect partner for their Europe expansion. IBM: Big, boring, business, but insanely attractive, reliable partner. Nebius could certainly help them innovate more together. I imagine for customers like IBM, getting access to Nebius' world class AI team is super valuable to identify the biggest areas to unlock value. Adobe: Similar to IBM. Boring "old" business. 7B profit per Year! But insanely threatened by AI. They need to innovate and keep up, NOW! Nebius will take care of you ;) OpenAI: Still the big dog today. Less attractive business. Slower growth, but still the top dog for now. They also will launch a couple of really interesting agent products which I believe will 100x their compute needs, at the very least. I imagine having OpenAI as a client is seen as a bigger risk and will worsen any company's credit terms. So if OpenAI will be a client, Nebius would probably get more $ / MW from them, but have to pay higher interest rates to lenders, for the increased perception of risk. Perplexity & https://t.co/WUxQnOp4Qt: These are inference-heavy companies. Nebius has been winning on inference pricing and latency lately. If the AE can prove that Nebius is cheaper for day-to-day operations, these high-traffic startups will jump ship to save on their massive monthly burn. Then there's a couple of noteworthy mentions like Mistral (very likely as they and Nebius are very strong in Europe), Cohere + Aleph Alpha, etc. Any of those wins would be great! Personally, I would love Anthropic and OpenAI deals, because those would obviously be huge. But more importantly they would test Nebius, be super happy with the quality and then increase deals aggressively. Since they need compute so desperately, I can imagine them paying upfront, at least partially! Great for capex heavy businesses like Nebius!

@daniel_kossApr 24, 2026 · 3:00 PM

$NBIS: 300% CAGR since its 2024 relisting. $IREN: 15% CAGR since 2021. -> a 20x difference in annualized returns. A part of the equation is timing, but certainly not the full 20x difference. $NBIS clearly started from an inferior physical infra position. But they have a superior team, business model, and as a result invest capital at a much higher ROI. $IREN bulls on X claim $IREN will "flip" $NBIS. That's funny, because what's happening is exactly the opposite. Nebius started from an inferior infra position, but is executing way better across the entire stack, launches great prodcuts, crushes benchmarks across the board, is hiring impressive talent and is aggressively scaling across the US, Europe and now also Asia. If you pay attention you can see a night and day difference in capabilities and execution. I'm done arguing. If you don't see it by now, cool. You do you. $IREN has a really good position today and I get the swing trade setup, as the market might still re-price their physical infra higher. But $IREN only has the valuable assets it has because it diluted shareholders to death for years. Judging their asset base today without looking at how they got here is like judging a companies income without looking at the costs it took to generate that income. $IREN's infra advantage is short-term. $NBIS's superior leadership, talent and business model is present long-term moats. If you can't tell the difference between a valuable asset and a sustainable business moat (which protects your business in the long-run and will lead to superior margins), that's on you. If you want to profit from a swing trade, $IREN actually looks really interesting at this price: capacity bottlenecks in the market, imminent catalysts like new deals (Anthropic, OpenAI, Microsoft), and a technical setup moving from mid-term downtrend into long-term uptrend. I see all of that and it looks bullish. Do you know who's funding all this new capacity? You are. Their $6B ATM is NOT good for shareholders, no matter how it gets framed. Cherry-picking the last 1 out of 5 years is dishonest. History shows this company has not done well for shareholders over the long run. So if you ask which business will add new capacity faster and monetize it more efficiently, in other words which is the better long-term compounder, it's obviously Nebius.

@daniel_kossApr 22, 2026 · 12:15 PM

@sunny21704 Are you trolling? For $NBIS the right metric is ARR in my view because they are in hypergrowth. Full year revenue doesn't make sense for a company growing 600-800% year over year. And their business will be highly profitable. 30% profit margin long term is my guess. But we're in hypergrowth mode for now. 10x ARR at least. 15x makes more sense, once the market realizes Nebius is the clear and dominant winner in the space.

@daniel_kossApr 21, 2026 · 5:26 PM

@mvcinvesting Nice! Very happy to see you didn't trim $NBIS. I think the fundamentals are outperforming the stock price, so despite the massive run, I feel like there's zero need to trim even at $200... AI demand is accelerating RIGHT NOW. Agents are here. Still stupid. I really hope that you will be able to manage your psychological drive to realize the gains and just stay in this name! For your gains of course, but also selfishly, because I hope you keep covering the stock lol If they hit mid-point of guided capacity ~900mw and can hit $12M / mw, which currently seems reasonable with GPU price hikes (just starting) and new chips, they will hit $10.8B ARR, significantly above guided 7-9B ARR. This also means their adj. EBITDA could be something like 50%. Everyone who uses Claude Code or Cowork knows how insanely slow, expensive, etc. it has become. Same for OpenAI and probably Gemini at some point. Obviously insanely bullish for $NBIS as all these customers will very happily switch to by now insanely good open models that they can run much cheaper via $NBIS I think you know all of this, but I'm trying to repeat some key parts of the thesis to avoid you getting shaken out 😂 And before anyone points out that I did trim recently: very different. Manuel is a long-term investor. I add very frequent trading. Me selling a stock means nothing and I could add it again a week later.

@daniel_kossApr 20, 2026 · 5:05 PM

Honestly? If you're a $NBIS or $IREN investor I think you should take 10min to read both sides. It will help you a ton to understand which thesis you want to underwrite and what the different views even are! I really like that Agrippa lays out his thoughts in detail and shares his full perspective. Such comments are rare and in my view what makes X so exciting! We agreed to take this to DMs for the deeper back-and-forth, which I'm looking forward to. But while reading the full comment I realized his comment does publicly characterize me as poorly informed or spreading bad information. Since Agrippa is very different from Mike Alfred or cult-like X trolls and I do respect him, I do want to address the substance of his post in 1 last post though, as I can't let such claims stand unanswered. Anyone reading should see both sides. First, one concession. He's right that "build speed" and "cloud scaling speed" are distinct KPIs. My original wording was not clear enough, so I want to sharpen it in this post. What I measured is the rate at which each company converts power into AI-operational, revenue-generating capacity. That's the KPI that matters for investors in my view, because MW sitting idle is a cost center, not an asset. Now the main points: 1. A BTC mining facility is not an AI data center. Agrippa argues IREN was building AI data centers in 2019/2021. But then concedes IREN had to add backup generators, redundancy infrastructure, and the rest to run cloud workloads. That is the work. With newer GPUs, interconnectivity, etc. this work will only get more complicated. I don't think the Bitcoin mining data center skills translate this simply to building world-class AI factories. So my view is still that IREN has built exactly 0 data centers you can count as world-class AI factories to date. There is no data I see that indicates this. AI data centers require Tier 3+ uptime, high-bandwidth networking, liquid cooling readiness, enterprise compliance, and SLAs that customers underwrite. A shed optimized for hash output does not meet that bar until converted. Calling the pre-conversion asset an "AI data center" because it was later retrofitted is like calling a warehouse a hospital because you could eventually install operating rooms. The conversion is the work. It's not "just" expensive, it's also complicated and takes increasingly rare and expensive talent. 2. The hardware cycle timing thesis is post-hoc. If IREN was strategically waiting for the right GPU cycle, why deploy H100s in Feb 2024 and H200s after? You cannot simultaneously argue "we timed Blackwell perfectly by waiting" and "we also deployed two prior generations." Those claims are incompatible. The simpler explanation is that the ramp was slower than planned, and the "cycle timing" narrative was built around it after the fact. 3. Independent benchmarks back my view. SemiAnalysis ClusterMAX 2.0 (Nov 2025) rates 84 GPU cloud providers, covering 95% of the market by volume. $CRWV: Platinum (only one) $NBIS: Gold $IREN: not rated SemiAnalysis explicitly lists IREN in the Bitcoin miner cohort and writes: "crypto miners are not yet competing at the Platinum or Gold tier for ClusterMAX." Their wording, not mine. Could IREN get there? Sure. Crusoe started as a miner and made it. But "not yet" is the operative phrase, and it's exactly my original point. Bottom line. My original claim was narrow: it is not YET proven that IREN can build and operate AI data centers at Nebius's level. Nothing in this reply disproves that. The entire reply argues IREN will, or could, or should. Those are forward-looking claims, not facts. I'm positioned based on demonstrated execution. IREN investors are positioned based on future cycles converting as planned. Both views are defensible. Neither is misinformation. IREN has tremendous upside IF they perform. I don't doubt that, but that doesn't make my view dishonest, negative or FUD.

@daniel_kossApr 19, 2026 · 7:49 AM

I’ve bought and sold $NBIS countless times. Here’s why that’s actually the most bullish thing I can do. I model and track many stocks. $NBIS is still by far my best risk/reward. But I’m a perfectionist. I invest full time. Just holding isn’t enough for me. My personal rule: → “If I can’t outperform $NBIS’s yearly return, I should just hold and do nothing.” So I actively try to: •Sell when sentiment turns (even if unjustified, like the Michael Burry AI bear thesis) •Buy back when I think the next leg up is coming His bear case was that GPUs would depreciate within 2-3 years. That has been completely falsified. We now see that even 5-6 years might be too pessimistic. Even the oldest GPUs are seeing prices rally. Supply shortage is that strong, and it’s only getting worse. Has it always worked? No. Will I make mistakes again? Absolutely. Has it worked overall? Yes. I’ve outperformed buy-and-hold $NBIS both last year and this year. People comment under every trade: “Trading doesn’t work.” They’re right. For most people. My actual advice is simple: buy quality, hold until the thesis breaks. That’s it. But that advice isn’t for me. I’m optimizing for every percent. I have the time, the models, and the track record to justify active trading. One more thing on taxes: If you’re in a country with capital gains tax, please don’t do what I do. The tax drag alone makes it not worth it. I invest through a fund in a zero-tax jurisdiction. Different rules apply. How do you guys think about trading? Would love to hear some smarter opinions than "trading never works".

@daniel_kossApr 18, 2026 · 4:22 PM

Update on my $IREN view Changed: - I'm way more certain that a pure Infrastructure business "baremetall" can actually become insanely attractive for 5+ years. I do think that the view that pure infrastructure will become a commodity is WRONG. Talent, GPUs, capital, energy, etc. arr just too scarce for this bottleneck to go away in the near term future. The resulting pricing power means calling the physical infrastructure a commodity is just wrong if you define commodity as sth that purely competes on price. Same as before: I totally get why $IREN investors are so excited about its position. The assets and strategic position in market look freaking perfect. Here are my 2 concerns: 1) I don't think they can execute nearly at the level of $NBIS meaning I think their AI factories have worse uptime, lower efficiency etc. How can we KNOW instead of debate if this will turn out to be true or not? Simply measure those metrics. 2) I also think they have significantly less qualified talent across all the 4 other layers of the "AI cake". Meaning they will offer a less vertically integrated service and have much lower margins. Now the important part: We will see if that's true or not. They should deliver AI factories at meaningful scale this year. If they prove me wrong and get comparable margins to $NBIS on the same amount of capacity I will publicly share that I was wrong and even happily take a position at a higher price! Even if they can NOT match Nebius, the raw compute business alone is turning out to be so much more interesting than anyone thought that IREN could still crush it. I just see so much upside in Nebius over next 3-5 years that I don't want to invest in IREN until they falsify all my skeptical points very clearly so that I KNOW my money is invested with the idea risk reward, not just high upside IF they perform.