Moats in our neocloud theme: $APLD Ellendale, Macquarie financed $BTDR Bhutan power, SEALMINER ASIC $CIFR AWS anchor tenant validation $CLSK low cost power per MW $CORZ multi-billion hosting backlog $CRWV GPU operator, OpenAI $DGXX 400MW, Blackwell ready pods $HIVE BUZZ HPC GPU subsidiary $HUT vertically integrated power $IREN cleanest HPC revenue ramp $KEEL $533M liquidity, three campuses $MARA largest scale, biggest footprint $NBIS lowest cost per MW operator $RIOT Texas grid scale power assets $SLNH behind the meter, 4.3 GW $WULF Core42 deal, nuclear adjacent $WYFI $865M Nscale anchor contract The important stuff? Secured power capacity, signed anchor tenant contracts, cost per MW competitiveness, and execution speed on conversion. I think #SLNH# offers deep value. NFA. Will keep updating...

I'll make this super clear for people wondering if $DGXX or $SLNH is more asymmetric: They serve two completely different purposes, in different layers of the same supercycle. Both genuinely asymmetric in their own way. Both sit in the Neocloud ecosystem. $DGXX as a GPU-as-a-Service operator and $SLNH as the renewable powered data center beneath it. Different theses, different risks, same tailwind. $DGXX (~$600M MC) - GPU-as-a-Service operator deploying $NVDA Blackwell GPUs directly to customers. Initially shared at ~$4 (up 105%+ now). > Similar model as $CRWV (~$60B MC), $NBIS (~$45B), $IREN (~$20B). First AI revenue contract signed. $1.1B $CBRS colocation deal. Hans Vestberg / $BLK connection. > 1.9% institutional ownership leaves massive room for re-rating. Earnings tomorrow, GPU rental starts on Friday. Risks: Early stage, $750M shelf filed (dilution capacity), negative margins, execution heavy. $SLNH (~$250M MC) - Renewable powered AI data centers. Wind farm acquisition closes vertical integration loop. Initially shared at ~$1 (up 65% so far). > Same renewable power thesis as $TLN (~$17B), $CEG (~$106B), $VST (~$50B). 4.3GW development pipeline. Difference between them is instead of wind farm → grid → data center, $SLNH does wind farm → data center. > Dorothy campus operational and expanding. Nasdaq compliance just regained. Earnings May 19. Risks: Overhang from active dilution. Cash burning. Execution risk on Dorothy 3 (300MW+ campus). Both are very early stage at this point. Both have execution risk. But both have real catalysts incoming. As for dilution, that's a risk with any early stage company. Again, bears were saying the same thing about $PLTR at ~$15. Now the same bears would full-port if it ever dips to $100. Valuation gap between current MC and what their competitors are trading at is what makes both asymmetric in their own layers.

"An overnight success that took five years" @_ConorMoore at @CoinDesk Live on finding pmf: blockchain rails make loan origination and settlement atomic, neoclouds needed capital that tradfi couldn't move fast enough to provide, and the AI boom brought both together.

Right now, the majority of AI workloads are executed on GPUs, but that could soon be changing with market adoption of AI application specific integrated circuits (ASICs). @Luxor's Mike San Miguel said that hyperscalers have been utilizing these ASICs for some time, but now neoclouds are making moves to leverage them as well. "They're just starting to hit the open market now The hyperscalers have been doing versions of these for about a decade. What's different now is that, as opposed to being bespoke chips for hyperscalers, we're starting to see new manufacturers that are producing these AI ASICs, and as a result these are going to start hitting the neoclouds. "You're probably going to see them hitting the market in scale in 2027 when they're going to start entering fleets for a lot of these neoclouds." This raises a a business question for AI companies, however, because unlike GPUs which can be used for inference and training, AI ASICs can only be used for one or the other. "A GPU can be used for inferencing training. But with ASICs, they're purpose built for one or the other -- they can't do both. "But that's the advantage. You get a much higher output...what it's probably going to look like in the next 2 to 5 years -- beecause these things take time to roll out -- is a hybrid approach."

$NBIS earnings were stellar and it’s now trading $200+ premarket. Reiterated $7-9B ARR in 2026. 40% adj. EBITDA margin projections, which is vastly outperforming expectations. 4 GW contracted capacity. $6.3B capital secured by $NVDA off solid financial offering structures. Glad my high conviction Neocloud pick is performing wonders and happy management is executing so well. In the words of Jensen: “Nebius will take care of you”

As I said before $IREN is basically dogsht compared to $NBIS. $NVDA didn’t give $IREN funding yet. So IREN needs to figure out how to buy enough GPUs to monetize 5GW capacity through their 6B ATM and other means. It’s an endless dilution machine just because they secured power. I call $IREN holders 0 IQ because they just buy in it to get diluted without understanding nuances of financing. Nvidia actually gave $NBIS funds. While Nvidia got a free no-risk purchase agreement for allowing $IREN to use their logo. $IREN is basically a marketing company at this point, while the other Neoclouds actually allow equity appreciation.

HSBC Raises $NVDA PT to $325 from $295 - Buy; ER Preview Analyst comments: "At its upcoming 1QFY27 results announcement on May 20, we expect NVIDIA to report 1QFY27 revenue of USD81.1 billion, 4%/3% higher than management guidance and Visible Alpha consensus estimates of USD78.0 billion/USD78.6 billion. We also expect 2QFY27E revenue of USD91.1 billion versus consensus estimate of USD85.6 billion, implying another “beat and raise” quarter. We also raise our FY28E EPS by 27% to USD13.01, 16% above consensus of USD11.20, on higher FY28E data center revenue of USD528 billion versus consensus of USD465.3 billion, on the back of an upward revision to chip-on-wafer-on-substrate allocation from 900,000 to 1.1 million wafers. Over the past five years, all major NVIDIA stock price movements have been led by a combination of its evolving AI product roadmap — starting with significant ASP pricing power with first-generation AI GPUs, A100 and H100 — driving significant earnings upside along with consistent “beat and raise” financial results. However, since the buzz around sovereign AI and the opportunity from neoclouds, no new narrative has emerged, and NVIDIA shares have underperformed the SOX over the last six months despite having two GTC events and two sets of financial results that beat estimates and raised expectations. Hence, we believe AI GPU earnings momentum and its upcoming Vera Rubin and Rubin Ultra product roadmap have become less meaningful narratives for significant re-rating or share price upside potential. Despite the ever-increasing CAPEX trend by CSPs that shows no signs of abating, NVIDIA now has to share the CAPEX with memory makers, AI networking, and server CPU vendors. Hence, NVIDIA needs to show evidence of diversifying its non-CSP customer base to fuel its AI GPU momentum. New TAM opportunities via agentic AI server CPUs and its recent optics-related deals could also potentially be emerging narratives that could lead to more significant earnings revisions or re-rating." Analyst: Frank Lee

Why did xAI hand over a 220,000-GPU cluster to Anthropic? The technical backdrop to xAI's decision to hand Colossus 1 over to Anthropic in its entirety is more interesting than it appears. xAI deployed more than 220,000 NVIDIA GPUs at its Colossus 1 data center in Memphis. Of these, roughly 150,000 are estimated to be H100s, 50,000 H200s, and 20,000 GB200s. In other words, three different generations of silicon are mixed together inside a single cluster — a "heterogeneous architecture." For distributed training, however, this configuration is close to a disaster, according to engineers familiar with the setup. In distributed training, 100,000 GPUs must finish a single step simultaneously before the cluster can advance to the next one. Even if the GB200s finish their computation first, the remaining 99,999 chips have to wait for the slower H100s — or for any GPU that has hit a stack-related snag — to catch up. This is known as the straggler effect. The 11% GPU utilization rate (MFU: the share of theoretical FLOPs actually realized) at xAI recently reported by The Information can be read as the numerical fallout of this problem. It stands in stark contrast to the 40%-plus MFU figures achieved by Meta and Google. The problem runs deeper still. As discussed earlier, NVIDIA's NCCL has traditionally been optimized for a ring topology. It works beautifully at the 1,000–10,000 GPU scale, but once you push into the 100,000-unit range, the latency of data traversing the ring once around becomes punishingly long. GPUs need to churn through computations rapidly to keep MFU high, but while they sit waiting endlessly for data to arrive over the network fabric, more than half of the silicon falls into idle. Google sidestepped this bottleneck with its own custom topology (Google's OCS: Apollo/Palomar), but xAI, by my read, has not yet reached that stage. Layer Blackwell's (GB200) "power smoothing" issue on top, and the picture comes into focus. According to Zeeshan Patel, formerly in charge of multimodal pre-training at xAI, Blackwell GPUs draw power so aggressively that the chip itself includes a hardware feature for smoothing power delivery. xAI's existing software stack, however, was optimized for Hopper and does not understand the characteristics of the new hardware; when it imposes irregular loads on the chip, the silicon physically destructs — literally melts. That means the modeling stack must be rewritten from scratch, which in turn means scaling is far harder than most of us imagine. Pulling all of this together points to a single conclusion. xAI judged that training frontier models on Colossus 1 simply was not efficient enough to be worthwhile. It therefore moved its own training workloads wholesale onto Colossus 2, built as a 100% Blackwell homogeneous cluster. Colossus 1, on the other hand — whose mixed architecture is far less crippling for inference, which parallelizes more forgivingly — was leased in its entirety to an Anthropic that desperately needed inference capacity. Many observers point to what looks like a contradiction: Elon Musk poured enormous capital into building Colossus, only to hand the core asset over to a direct competitor in Anthropic. Others read it as xAI capitulating because it is a "middling frontier lab." But these are surface-level reads. Look at the numbers and a different picture emerges. xAI today holds roughly 550,000+ GPUs in total (on an H100-equivalent performance basis), and Colossus 1 (220,000 units) accounts for only about 40% of the total available capacity. Colossus 2 — built entirely on Blackwell — is already operational and continuing to expand. Elon kept the all-Blackwell homogeneous cluster (Colossus 2) for himself and leased out the older, mixed-generation Colossus 1. In other words, he handed the pain of rewriting the stack — the MFU-11% debacle — to Anthropic, while keeping his own focus on training the next generation of models. The real point, then, is this. Elon's objective appears to be positioning ahead of the SpaceXAI IPO at a $1.75 trillion valuation, currently floated for as early as June. The narrative SpaceXAI now needs is that xAI — long the "sore finger" — is not merely a research lab burning cash, but a business with a "neo-cloud" model in the mold of AWS, capable of leasing surplus assets at high yields. From a cost-of-capital perspective, an "AGI cash incinerator" is far less attractive to investors than a "data-center landlord generating cash." As noted above, the most important detail of the Colossus 1 lease is that it is for inference, not training. Unlike training, inference requires far less tightly synchronized inter-GPU communication. Even when the chips are heterogeneous, the workload parcels out cleanly across them in parallel. The straggler effect — the chief weakness of a mixed cluster — is essentially neutralized for inference workloads. Furthermore, with Anthropic occupying all 220,000 GPUs as a single tenant, the network-switch jitter (unanticipated latency) that arises under multi-tenancy disappears. The two sides' technical weaknesses end up complementing each other almost exactly. One insight follows. As a training cluster mixing H100/H200/GB200, Colossus 1 was an asset that could only deliver an MFU of 11%. The moment it was handed over to a single inference customer, however, that asset transformed into a cash-flow asset rented out at roughly $2.60 per GPU-hour (a weighted average of the lease rates across GPU types). For xAI, what was a "cluster from hell" for training has become a "golden goose" minting $5–6 billion in annual revenue when redeployed for inference. Elon's genius, I would argue, lies not in the model but in this asset-rotation structure. The weight of that $6 billion becomes clearer when set against xAI's income statement. Annualizing xAI's 1Q26 net loss yields roughly $6 billion in losses per year. The $5–6 billion in annual revenue generated by leasing Colossus 1 to Anthropic, in other words, almost perfectly hedges xAI's loss figure. This single deal effectively pulls xAI to break-even. Heading into the SpaceXAI IPO, this functions as a core line of financial defense. From a cost-of-capital standpoint, if the image shifts from "research lab burning cash" to "infrastructure tollgate stably printing $6 billion a year," the entire tone of the offering can change. (May 8, 2026, Mirae Asset Securities)