AlexKarpsHair

Just to be clear. I LOVE $NBIS. I think the management is incredible. World class tech engineers. But do not discount $IREN. They're team is incredible in other areas. The acquisition of Mirantis changes things dramatically for the company. I love both. I own both.

@brad_D6 @Mario20253035 They’re adding Kubernetes orchestration, enterprise cloud infra software, k0rdent AI, support capability, and 1,500+ enterprise customers. If power is the actual bottleneck, IREN may be starting from the more valuable side of the stack.

@brad_D6 @Mario20253035 Nebius does have a deeper native AI-cloud engineering pedigree today. But this frames IREN as if it is still just power/real estate. That changed with the Mirantis acquisiton.

$NBIS execution risk isn't priced in. let me break down what most equity research is missing. structural setup: nebius does not own vineland. dataone (BSO subsidiary, ardian-backed) owns the site. nebius confirmed this themselves when announcing independence, kansas as "the first fully owned by nebius." that one sentence reframes the entire thesis. their flagship US site the one backing $17.4B microsoft, the meta partnership, and the $2.6B bloom commitment is a colocation arrangement. not infrastructure ownership. what this means in practice: 1) the $2.6B bloom contract installs fuel cells onto someone else's real estate. nbis funds capital improvements to dataone's asset. all appreciation accrues to the landlord. 2) nbis does not control the permit timeline. dataone is the applicant before NJDEP. dataone requested the 6/25 board reschedule. nbis can tell investors "we're working on it" they have zero standing to refile, accelerate, or negotiate with the township. 3) the microsoft and meta delivery clocks run against a site nbis doesn't own. execution risk tiers: near-term (6–12 months): • 6/25 vineland board meeting is "tentative." another slip = bloom's 90-day install clock can't even start. • bloom backlog is ~$20B (oracle 2.8GW, AEP 1GW, brookfield $5B, project jupiter 2.45GW). nbis 328MW is queued behind larger contracts signed earlier. "phased delivery across 3 phases" means you're not getting it all in 2026. • microsoft start dates "later this year." most hyperscale contracts have service credits or step-down clauses for sustained delays. nbis hasn't disclosed terms. structural (12–24 months): • SOFC + battery buffer architecture is unproven at scale for GPU workloads. nebius's own engineering lead said bloom didn't fit their spec on load-following grounds. now they're deploying exactly that architecture. if GPU loads cause voltage sag or frequency drift, they're either accepting degraded compute or writing more checks (the oracle-abilene playbook). • dataone is a 2024 startup with no public track record at 300MW scale. if dataone hits financial or operational stress, nbis has no recourse. dataone is learning on the job at nbis's expense. • microsoft + meta is likely 60–70% of forward revenue. both have alternatives. both can read the structure. cascading scenario: permit slips → fuel cell deployment slips → microsoft invokes service credits → revenue guide cut → stock compresses → capex guide-up creates financing stress → equity raise → dilution → multiple compresses → covenant pressure on the converts. nbis guided capex higher while lengthening time to first revenue. that's the exact profile that turns into a 60% drawdown if the market loses patience. bottom line: nbis is running a hyperscale-customer business out of someone else's data center, with a power architecture they originally rejected, against contractual milestone dates they don't control. the market is paying them an infrastructure-owner multiple for execution risk that compounds in three independent failure modes. physical infrastructure ownership is the moat. $IREN owns sweetwater dirt, owns the 345kV interconnect, energized may 2026. nbis doesn't own vineland and can't control when it goes live. these are not the same business.

This isn’t a picture of 3 CEOs. It isn’t a picture of 3 Founders. It isn’t a picture of 3 billionaires. It’s a picture of 3 guys who are all of those things but also, most importantly, working collaboratively to build the future of AI in the United States and globally.

That Mirantis acquisition is incredible. The market is OBSESSING on “when deal?” But IREN is quietly doing something more structural: building the stack required to command premium pricing on scarce AI capacity. IREN is moving to a full-stack AI cloud operator. $IREN #IREN

I LOVE $NBIS. One of my best picks. BUT: $NBIS is a cloud company becoming an infrastructure company. $IREN is an infrastructure company becoming a cloud company. And if power is the real constraint, $IREN may be starting from the more valuable side of the chessboard.

@Sam_Badawi Sam, take a look. x.com/AlexKarpsHair/…

I’ve discussed the energy bottleneck for the last 3 weeks. The story continues to play out as neocloud platforms including $NBIS begin signing major agreements directly with energy providers. The $2.6B deal with $BE is likely just the start.

@BitcoinAIGuy It's the true bottleneck. What's even more hilarious is that the folks who are actually involved *IN* the buildout itself are SCREAMING that LAND / POWER is scarce. There's not enough. They need 1000x more than what's available. Yet the market is HYPER focused on deals only.

@AlexKarpsHair few understand this

seems costly for $NBIS, and others without secured power... an extra ~$10B/GW expense in fuel cells alone. On top of obvious, third-party execution risks. $IREN does not have this expense or these additional risks... and also saves a ton of money NOT paying rent/colocation fees because it owns and fully controls their sites. got $IREN?

@chinoalemano @christo84019133 I'll share this here, too. x.com/AlexKarpsHair/…

@chinoalemano @christo84019133 You guys have to realize one important thing. IREN own an insanely valuable *REQUIREMENT* in the AI build out. IREN do not need to rush out and get deals. They're the ones in the driving seat. They can and SHOULD wait for the best deal for the company.

$IREN - Today. I don't want any BS afterward. I said there will be a deal involving IREN today or I'll stfu until there is one. Most comments are laughable, even people bullish on IREN doubt it. If there's a deal involving IREN today, it's not fair to come back later saying it was obvious. Clearly it isn't. Yesterday, right before close, I put 70% (rest sits with DGXX) of my portfolio into IREN sub $48. I tripled my % in IREN, since it was at 25%. But I didn't just triple it, I multiplied it by more, since I added IRE before market close. My thesis is simple. My money where my mouth is. NVIDIA didn't go to IREN without a customer. NVIDIA already had a potential customer, which could very well be Anthropic. AMD not long ago closed 6GW directly with OpenAI. I expect something similar, no less, for the real king: Anthropic. Anthropic must have demanded something similar from NVIDIA. That's why NVIDIA went to IREN. Obviously NVIDIA made a deal that benefits itself. And IREN accepted because obviously, they're getting a strong customer out of it. That's why Dan is looking for investors. He needs money. He needs to use the ATM. He needs money for the SweetWater buildout. Maybe even to pay for the GPUs too. But the deal would be so big that I think IREN is going to fly. On the other hand, I'm also seeing the bulls being less bullish, even bearish. Bad sentiment. IREN has barely grown 10% YTD, while its peers are up 150%+ (let's go Digi!). That's where my spider sense kicks in, telling me it has to rerate. Maybe not 150%. Maybe not even 100%. But at least 50% from here. The worst case scenario, waiting until July. Though I don't think we'll need to wait that long.

@franklee6924T Yep. Agree fully with this. My thesis: x.com/AlexKarpsHair/…

AI Compute Is Approaching a Turning Point $NBIS’s current premium valuation is essentially a market prepayment on the narrative of a “future hyperscale AI cloud platform.” Wall Street is willing to assign it extremely high multiples mainly because of its explosive revenue growth and the credibility backing it receives from Meta, Microsoft, and NVIDIA. From that perspective, it arguably deserves a higher valuation than $IREN. But what the market is pricing today is not a risk-adjusted path — it is the theoretical best-case outcome. Recent target price increases from Citi and Goldman Sachs have only reinforced expectations around this idealized success scenario. NBIS’s current business model follows a classic project-finance structure: first pre-sell long-term capacity, then finance against those contracts, then lease and build data centers, deploy GPUs, and finally deliver to customers. In the short term, this model provides financing capability, revenue visibility, and attractive backlog figures. But the tradeoff is that future profit elasticity becomes locked in, while strategic flexibility declines. As electricity prices rise and compute infrastructure architectures evolve, NBIS risks missing significant upside from both trends. Although NBIS emphasizes that “75% of contracted power has shifted to owned contracted capacity,” contracted does not mean energized, built, delivered, or operational. Much of this so-called capacity still exists only at the stage of land agreements, power intentions, or development contracts. The real risk in the neocloud industry is not lack of demand — it is execution and delivery. GPUs arriving does not mean they can immediately go online. Approved power does not mean actual grid connection. A completed data center does not mean liquid cooling has been fully commissioned. Signed customers do not automatically translate into recognized revenue. NBIS’s ramp guidance is extremely aggressive, targeting growth from roughly 170MW to 800MW–1GW within about a year. In the real infrastructure world, that pace is nearly anti-physical. Substations, cooling systems, high-voltage distribution, grid approvals, liquid-cooling stability, and supply-chain coordination can each become bottlenecks. Any delay could result in project slippage, contract penalties, idle GPUs, and deteriorating capital efficiency. If all of these systems were fully vertically controlled internally, the execution credibility would be much higher. But once multiple third parties are involved, it becomes difficult to ensure flawless coordination across the board. Yet NBIS’s valuation today largely assumes these problems will be solved smoothly. The risk is that the market is valuing NBIS like a future hyperscaler, while in reality it remains a highly leveraged AI infrastructure contractor with substantial construction risk and long-term margins that are still unproven. That gap represents an enormous valuation leap. NBIS’s rapid expansion, strong EBITDA, and growing backlog exist because it is not truly solving infrastructure constraints — it is financially routing around them. It resembles an asset assembler for the AI era, rapidly piecing together capacity through leased campuses, power agreements, third-party data centers, structured financing, and NVIDIA ecosystem support. The heavy work of self-built, self-controlled, long-cycle infrastructure simply takes time. It is understandable why capital markets reward this type of short-term efficiency, but genuinely scarce resources are difficult to sustain through asset assembly alone. The bottleneck is gradually shifting away from GPUs and toward power, time, grid interconnection, scalable campuses, and high-density liquid-cooling infrastructure. Whoever controls these assets will ultimately possess the real pricing power. This is precisely the fundamental strategic difference between IREN and companies like NBIS and CRWV. CRWV pioneered the model of using contracts to drive financing and financing to drive expansion, and NBIS followed that path. Neocloud is still a newly formed category, and the market’s valuation framework has largely been shaped around CRWV’s approach: whoever can monetize GPUs into revenue the fastest wins. That is why the market capitalizations of CRWV, NBIS, and IREN have correlated almost perfectly with total contracted capacity. IREN effectively stopped after the Microsoft deal and no longer appeared eager to aggressively pre-sell capacity, despite arguably being one of the most qualified companies to do so. Instead, it placed “bringing compute online” as the top priority and shifted its strategic focus toward transforming scarce secured resources into fully operational AI factories as quickly as possible. The market is rewarding whoever monetizes GPUs the fastest — not whoever brings real compute infrastructure online the fastest. Capital’s choice is practical: the benefits of the former appear immediately, while the latter takes time and includes many variables that are difficult to measure today. Whether investors can recognize that difference is becoming an important test. If you believe the future will continue to be dominated primarily by standalone GPUs, then the current market valuation framework is probably correct. But if you believe the future will instead be defined by the integration of GPUs with power infrastructure, grid interconnection, high-voltage distribution, liquid-cooling stability, cluster orchestration, network topology, and uptime, then the market’s current valuation direction may ultimately prove to be a trap. IREN’s agreement with Microsoft was not just about validating capability. The larger purpose was to prepare and validate the large-scale deployment of high-density racks and highly efficient liquid cooling systems, positioning itself ahead of the industry’s biggest upcoming transition. Why has IREN seemingly pursued “No deal” afterward? I believe the answer is gradually becoming clear: AI compute is transitioning away from a world where the GPU chip alone is the central character and toward an integrated AI factory model. In this new paradigm, the moat is no longer simply GPUs — it becomes GPUs combined with power, land, grid access, high-voltage distribution, liquid-cooling stability, cluster orchestration, network topology, and uptime. The barrier becomes much higher and the moat becomes much wider. NVIDIA is positioning itself to dominate the foundational layer of the AI industry through this transition. If this direction becomes dominant, then the valuation foundation currently supporting CRWV and NBIS — maximizing near-term GPU monetization — will fundamentally change. The entire underlying logic of AI compute development will shift. Dell’s integration capabilities, IREN’s more advanced AI factory capabilities, and the DSX intelligent factory systems they are building together with NVIDIA could become the center of AI industry development for years to come. NVIDIA has now delivered two consecutive earnings reports far above expectations while the stock price reaction remained relatively muted. My guess is that Wall Street still does not fully see the future form of compute infrastructure. Standalone GPU performance may be approaching its practical ceiling around the VERA RUBIN generation, so the next layer of compute evolution likely shifts toward large-scale integration. If NVIDIA confirms this direction in future earnings commentary, then the valuation frameworks of all companies involved in this transition could be completely rewritten. CRWV and NBIS were key enablers during the era when standalone NVIDIA GPUs were the core protagonist of AI compute. In the coming era of integrated AI factories and intelligent infrastructure, they will still play important roles — but the logic behind how they are valued may change dramatically.

@Craaazy1231 My thesis in a nutshell: x.com/AlexKarpsHair/…

$IREN Someone change my mind. I'm willing to learn about this, since I'm so bearish on this company. What makes Iren so relevant? What makes it a good company? Why should I even invest in it?

The market wants “Deal first, then value.” I'm betting on “Scarce capacity first, then customers compete, then value gets marked up.” I don’t need $IREN to announce a deal every week. I need the world to become more power constrained while $IREN sits on increasingly capacity.

CoreWeave and Nebius proved that AI-cloud demand is enormous. IREN may be entering the same market from a better physical starting point: already owning or controlling the land/power layer that everyone else is scrambling to secure. $IREN #IREN

Substations and transformers are constrained. Large powered campuses are rare. Good land near high-capacity grid nodes is finite.

The case for $IREN: Whilst CRWV/NBIS started with the GPU-cloud, software, and the AI infrastructure talent, they then had to secure land, power, leases, data-center capacity. This is a HUGE amount of capex.