there's a lot of FUD on the timeline this week, so i went back to my favorite story about how fear actually spreads. three schoolgirls on a train once started a run on a perfectly healthy bank. Japan, December 1973. one of the girls had just landed a job at Toyokawa Shinkin, a small local bank. her friends teased her that banks were "dangerous" (they meant robbers). it was a joke. she didn't take it as one. that night she asked a relative if it was true. The relative asked someone else, and the question passed from family to family, through a beauty salon, picking up weight as it went. Then a man borrowed a shop phone and told someone to withdraw 1.2 million yen for a routine payment. a listener overheard and assumed he knew something. now the rumor had a witness. a taxi driver later traced how the story changed in a single day: - afternoon: the bank "might be in danger" - 2:30: "it is in danger" - 4:30: "it will go into bankruptcy" - night: "its shutter will never be lifted tomorrow" nobody added a single fact, each retelling just dropped a hedge. by the end, more than a billion yen had walked out the door. The Bank of Japan rushed in cash and stacked it a meter high in front of the vault so people could see it with their own eyes. Police went looking for the crime behind the rumor and found nothing, the bank had done nothing wrong. so why does the scary version always win? because a loss feels about twice as strong as an equal gain. Psychologists have measured it for decades. Standing in line costs you an afternoon, but being the one who stayed could cost your savings. with that trade in front of you, running is the rational move, even when the bank is fine. which brings me to this week. the timeline is busy with stranded compute and debt stories. Those are fair questions, and some builders will deserve them, but the rumor doesn't check who, it hits everyone standing closest. so do what the Bank of Japan did, skip the gossip and go look at the vault. is the power actually connected? are the customers actually signed? is the money already there? those answers don't change because a post went viral. They change slowly, in substations and contracts and concrete. and that work didn't stop this week. If the cranes are still up, the racks are still getting installed and the demand that started all of this is still there. a run doesn't make a bank unsound, it just tests whether it was. Toyokawa survived, the ones building on solid ground usually do.

Avride has been named Autonomous Vehicle Solution of the Year in the 2026 AutoTech Breakthrough Awards! From robotaxis on Dallas roads to delivery robots on sidewalks and campuses, we’re building autonomy for the real world. Multiple applications. A growing impact.

$NBIS @realroseceline Good framework, but it fits any capital-intensive company. Every question in it has a number for Nebius, and the essay doesn't use one. Most are in the Q2 letter. Returns vs cost of capital. The scoreboard you ask for, cash returned per dollar invested, is close to one Nebius already discloses. Payback on Q2 deals is 22 months, capex and operating costs included, down from 2-3 years historically. That's ~55% of the capex back each year. If the GPUs were worth zero after a 3-year contract, that's still a ~29% unlevered pre-tax IRR. The money funding it: converts at 0.50% and 4.50%, Citi at ~6.6%, an asset-backed loan at SOFR + 2.50%. That's well above your 12% example, funded far more cheaply. Timing. In H1 2026 customers prepaid $4.4B, funding 54% of Nebius's $8.1B of capex ahead of delivery. ~70% of Q2 deals had prepayments covering 50-60% of capex. The cash-before-revenue problem you describe is largely being paid for by customers. Obsolescence. Servers are depreciated over 5 years, and the payback figure is cash with capex included, so it doesn't lean on EBITDA. The market also says older chips are holding value: Nebius's older-generation GPU pricing rose 30%+ QoQ in Q2, and Crusoe's CEO said last week A100s rent above 2020-21 levels and H100s above their 2023 launch. With a 22-month payback, the capex comes back well inside the depreciation life and inside any contract of 2+ years. Whatever the GPU earns after that is margin. That also answers your $2B vs $10B FCF question. A ~55% annual cash return on hardware that costs ~20% a year to replace is the $10B world. Pricing. The 10x demand / -80% price example mixes price per token with price per MW. Tokens get cheaper because each MW produces more of them, which cuts the provider's cost per token too. Nebius rents MW, and revenue per MW is rising: $20-25M on Q2 deals, $40-50M+ short-term, a first auction 15% above its best Blackwell price and the latest one higher again. Rubin is ~$40B per GW of NVIDIA hardware vs ~$25B for Grace Blackwell, so it has to rent for more. Operating leverage. During a contract, utilization risk sits with the customer. The contracts are take-or-pay, and management says the short-term deals and auctions are too. Contracts. The trade-off you describe is how they already run it: long-term investment-grade contracts for the floor, short-term deals and auctions for upside, and part of 2027 held back on purpose though management says it could sell all of it on current terms. Competition. A glut needs cheap capital and power. Lenders are rationing capital to the strongest operators right now, and new power takes years to connect. The telecom bust happened with most fiber dark. This month SpaceX said on its call it couldn't rent GB300s, GB200s, TPUs or even H100s from other clouds. SemiAnalysis rated 77 GPU clouds last month and only CoreWeave and Nebius made Platinum, above Google Cloud, Oracle and Azure. Value capture. Microsoft and Meta are the kind of companies your essay expects to capture the value, and both signed multi-year deals with Nebius rather than wait to build it all themselves. Dilution. Nebius sold its 12.7M ATM shares at ~$224, and its August converts convert at ~$313 and ~$325, about 2x today. My per-share targets take the count from ~309M shares today to ~432M by end-2028 with every convert treated as stock, and still get ~$1,190. Where I agree: none of this is proven over a full cycle. Execution is the real risk, getting 1 GW+ a year online on time, and post-contract pricing has to hold. But "great business or great story" doesn't need to stay philosophical. A 49.7% core EBITDA margin, a 22-month payback and customers prepaying half the capex is more evidence than the essay allows for.

This recent sell off in Neocloud & Data Centre stocks is one of the biggest bear traps I have witnessed. It is so obvious that this is a bear trap given the huge demand for their services, the increasing prices of GPU and the timing of the sell off. This sector will be the next bottle neck of the AI build out and this buying opportunity will likely prove to be one of the best.

The Wall Street playbook: 1. Say AI is a bubble 2. Cause mass panic and selloffs 3. Buy sold off stocks for cheap 4. Say AI had a big breakthrough and pump your stocks 5. Sell your stocks 6. Say AI is a bubble again Rinse and repeat…

🚨 NOT ALL NEOCLOUDS ARE CREATED EQUAL. I’ve been saying this for weeks. $IREN and $NBIS should NOT be trading like the rest of the sector. Look at the debt-to-equity ratios 👇 🟢 $NBIS: 1x 🟢 $IREN: 2x 🔴 $APLD: 4x 🔴 $HUT: 5x 🔴 $CIFR: 9x 🔴 $CRWV: 10x 🔴 $WULF: 36x Here’s my prediction: The market is eventually going to separate the financially stronger players from those drowning in debt. When financing gets tighter, leverage becomes a LIABILITY, not an advantage. I believe $IREN and $NBIS could be among the first to recover and significantly outperform, while heavily leveraged competitors struggle to regain their highs. The entire sector has been punished. But the balance sheets tell VERY different stories. The next big trade might not be buying neoclouds. It might be buying the RIGHT neoclouds. Time will tell. 👀

$NBIS: A Simple ARR Valuation Model and Why Nebius Could Be Undervalued Here There are different ways to calculate valuation for a high growth company that is still pre-profit. As for balance sheet, it’s currently strong as Nebius has about the same amount of cash on hand as outstanding debt (roughly ~$8B cash vs. ~$8.5–10B debt as of the latest reported quarter, primarily longer-dated convertibles), but also owns stakes in subsidiaries (including Avride, TripleTen, and interests in ClickHouse and Toloka) that various sum-of-parts analyses value conservatively in the $5B+ range, and their Adj EBITDA has hit around 40% (Q2 2026 printed ~41% group adjusted EBITDA margin, with AI cloud higher, and full-year 2026 guidance reaffirmed at approximately 40%). Here I will Use a Basic ARR Model: Nebius is now trading at ~$160 a share which is ~$43B market cap WHAT ARR MULTIPLE TO USE : At the worst (lowest) trough of the bear market in winter of 2022, no major Cloud or SaaS company growing ARR at greater than 100% traded below 6X ARR. But we’ve seen as high as 14-20X ARR with some AI cloud companies in raging bull markets like for example Coreweave in 2025. My multiple range then is 6X ARR to 14X ARR. To be more conservative I’ll take the high end bull metric out completely and use a simplified 6X for BEAR, 8X for NEUTRAL and 10X for BULL. Nebius is growing ARR at over 100% YoY. So at this point due to macro fears rooted in higher oil prices and potential for FED rate hikes and higher cost of capital I will primarily be looking to apply the bear market multiple of 6X ARR. It's important to understand that the market is forward looking and already anticipating what Arkady will guide for on Q3 for end 2027 ARR. Comments from Nebius CRO Marc Boroditsky in August indicated potentially tens of billions in revenue at that time next year (that would be by August 2027). And this week at GTC, Marc indicated that Nebius would be looking to acquire 1GW worth of Vera Rubin GPUs in 2027 (note he didn’t say all would be online). Other management commentary at GTC indicates that Nebius will achieve roughly 50% of their revenue from the short term and auction contracts (estimated now to be roughly in the range of $40-$50M per MW). So with a CEO who has never missed guidance, GPU prices continuing to rise, Arkady saying publicly that financing is no longer a problem (only a task), the fundamental trajectory starts to look increasingly bullish at this valuation. Let’s hypothesize what end 2027 ARR might look like based on the above commentary. SHARE COUNT : I will start with all the current total shares outstanding possible if fully diluted ~378M (that’s assuming all convertibles and warrants are exercised). I will then assume Nebius opts to use the remaining ~12M share ATM for 2027 Capex bringing the total to around 390M shares. I would then add another 10M shares for flexibility to do another convertible raise in 2027 (which may or may not be needed) which brings my total estimate to approximately 400M shares outstanding. 2027 ARR Hypothesis (illustrative ranges grounded in the CRO’s “tens of billions,” the 1GW Vera Rubin commentary, short-term/auction pricing of $40–50M per MW, and continued high growth): In a more conservative bear case we might see the street's current estimate of ~$15B ARR if deployment and utilization of new capacity come in much lower than hoped. Imo this would be a bear case based on the above commentary. A case for $25B ARR would reflect solid progress bringing most of the above-reference 1GW of Vera Rubin capacity online with a mix of contract lengths. A bull case made by @daniel_koss would bring Nebius closer to $35B ARR and align more closely with some of the higher published estimates that factor in a substantial contribution from the Vera Rubin capacity plus the existing 2026 ARR base. Applying the hypothesized exit run-rates to 400M shares, the bear-case $15B ARR would imply roughly $225 at 6x, $300 at 8x, and $375 at 10x. I think this is roughly where the street thinks we will be which explains the average institutional PRICE TARGET of ~$304 right now. Taking the commentary seriously and we could see more like $25B ARR by end of 2027 which would point to about $375 at 6x, $500 at 8x, and $625 at 10x. The bull-case $35B ARR would suggest approximately $525 at 6x, $700 at 8x, and $875 at 10x. These are of course simple ARR multiples applied to hypothesized 2027 exit run-rates and are not forecasts of course so take all this with a grain of proverbial salt. Actual outcomes will depend on capacity online, utilization, realized pricing (what kind of mix of short-term premium vs. longer contracts Nebius can achieve), dilution, financing execution, and of course the ever-volatile macro conditions. If Nebius achieves what I think is a base case of $20B end-2027 ARR and we continue using the same simplified multiples from the earlier model (6x bear, 8x neutral, 10x bull) with 400M shares outstanding, my implied end-2027 PRICE TARGET range would be as follows: • 6x ARR (bear): $20B × 6 = $120B market cap → ~$300 per share • 8x ARR (neutral): $20B × 8 = $160B market cap → ~$400 per share • 10x ARR (bull): $20B × 10 = $200B market cap → ~$500 per share Would welcome your comments! Thx for reading and of course these are just my back of the napkin projections but they are why I remain heavily invested in Nebius... ...NFA : Always Do Your Own Research