· 个股论点

作者认为NBIS全栈软件优势带来高毛利,优于IREN等矿企,故重仓NBIS。

涉及标的:

中文翻译

为了解决关于新云(Neocloud)的争论: $NBIS > $IREN 及其他。 基于 $ORCL 的报告、NBIS 白皮书、$CRWV 的收购案以及其他因素, 我决定将数百万资金整合进 Nebius,并清仓像 $CIFR 这样的矿企。 📏 毛利率 > 吉瓦(GW)产能 以下是数学逻辑及原因: 此前,由远期收入支撑的新云板块(例如 $NBIS 与 $MSFT 的 190 亿美元交易)经历了巨大的投机性上涨。 然而,由于原始 GW 产能(如 $IREN)和廉价能源,加密货币矿企近期上涨了 500%+。 然而,人们忽略的是,产能虽能带来更高的收入,但如果不可盈利(例如 $ORCL 14% 的毛利率),这些收入毫无意义。 随着更多关于 $ORCL 建设失败、$NBIS 白皮书等的信息公布,我们得知如果经济模型无法转化,电力和产能意义不大。 而 $NBIS 拥有一切优势。 1. 毛利率差距:全栈 vs 中间件 像 $IREN 和 $CIFR 这样的矿企必须依赖编排合作伙伴(Fluidstack、Poolside 等)来变现其 GPU。这意味着要放弃 20-30% 的收入(~来自深度研究的私人估算),并承担 GPU 折旧、电力和运维成本。结果是? 在 $3-4/小时 的 GPU 定价和 ~80% 利用率下,IREN/CIFR 的毛利率 = ~44-52% 在 $5-6/小时 和 90% 利用率下,毛利率可能达到 ~55-60%,但这已是天花板。 与此同时,Nebius 通过拥有编排层并分摊 4 年成本,目前获得 70-75% 的毛利率(上一季度为 71.2%,可能为 60-70%,近期白皮书声称更高的 GPU 利用率可能使其更高)。 随着规模和利用率的提高,这一差距会复利扩大。 矿企每赚一美元都要分出一部分,而 Nebius 的毛利率随时间增加。 2. 软件是护城河,Oracle 证明了这一点 即使 $ORCL 这样市值 8000 亿美元的超大规模云服务商,也无法盈利地构建 GPU 编排,其 AI 租赁平台报告约 14% 的毛利率,并在此过程中亏损约 1 亿美元。 如果 Oracle 都无法执行,指望市值 1 亿美元的小型矿企如 $SLNH 从头构建超大规模云级别的编排系统只是痴人说梦。 像 Fluidstack 这样的平台是必要的桥梁,但代价高昂:毛利率压缩、收入流失和平台依赖。 Nebius?它已经完成了最难的部分。其内部编排软件、GPU 利用率,没有中间商赚差价。 3. 电力 vs 毛利率计算 单张 H100 功耗约 0.7-0.84 kW。即使在 $0.10/kWh 的电价下,每张 GPU 每小时的电费仅为 $0.07-0.08。当 GPU 以 $4-6/小时 出租时,这仅占收入的 1-2%。 真正影响大局的是什么?利用率: 50% -> 85% 的正常运行时间 = 每张 GPU 收入增加 70% 这是数百个基点的毛利率波动,远超与矿企相比的“廉价电力”优势。 软件利用率/编排是护城河,其重要性比廉价电力高 30-70 倍。而且 $NBIS 也不缺廉价电力 lol。 4. 矿企中的 CRWV 例外 大家都指向 $CRWV 作为矿企转型的案例,但 Coreweave 实际上花了数年时间,并进行了数十亿美元的软件收购。 即使现在,报道显示其利用率仍落后于 Nebius。如果矿企认为他们能复制 CRWV,祝他们好运。 我预计 $NBIS 在明年的财报中,当执行与投机面临检验时,将大幅跑赢。 🧩 非对称性 Nebius 毛利率:70-75%(全栈,4 年折旧) IREN/CIFR 现实毛利率:40-60%(中间件,2-3 年折旧) 毛利率差额:15-30 个百分点 执行风险:Nebius = 0(已在执行);IREN/CIFR = 高 Nebius 不是电力博弈。它是软件毛利率博弈,带有硬件上行空间。对于 $IREN 等,你是在猜测它能否像 $CRWV 那样成功。 人们不断说“软件不是护城河,看看 $CRWV 对 $IREN 未来 HPC 毛利率的影响”,但如果 $ORCL 这样最大的超大规模云服务商之一都卡在 14% 的毛利率。那么我们怎么指望这些小加密货币矿企能搞定 Coreweave。 就非对称性更新而言,$NBIS 是明确的赢家。就原始潜在上行空间而言,如果 $IREN 能像变魔术一样搞定 $CRWV 并在此方面击败 $ORCL,那我确实会认输。 软件全栈是一个巨大的护城河,人们严重低估了它。对于 $NBIS,他们可以只是即插即用,随着时间推移扩大规模,拥有行业内最高的毛利率。 $NBIS 就是真正的非对称回报的样子。

英文原文

To settle the Neocloud debate: $NBIS > $IREN + others. Based on the $ORCL report, NBIS whitepaper, $CRWV acquisitions, and other factors, I decided to consolidate millions into Nebius and sell off miners like $CIFR. 📏 Gross Margins > GW capacity Here's the math + why: We've had a huge speculative run across the board on Neoclouds backed by forward revenue (eg. 19B $MSFT deal with $NBIS). However, crypto miners have recently gone up 500%+ due to raw GW capacity like $IREN and cheap energy. However, what people miss out on is capacity leads to much higher revenue but that revenue means nothing if it's not profitable (eg. $ORCL 14% gross margins). Now that more information has come out regarding $ORCL's buildout failure, $NBIS's whitepaper, and others, we know power and capacity mean little if the economics don’t translate. And $NBIS has everything. 1. The Margin Gap: Full Stack vs. Middleware Miners like $IREN and $CIFR must rely on orchestration partners (Fluidstack, Poolside, etc.) to monetize their GPUs. That means giving up 20–30% of revenue (~private estimates from deep research), plus absorbing GPU depreciation, power, and O&M. The result? At $3–4/hr GPU pricing and ~80% utilization, IREN/CIFR margin = ~44–52% At $5–6/hr and 90% utilization, margin could reach ~55–60%, but that’s the ceiling Meanwhile, Nebius earns 70–75% today by owning the orchestration layer and amortizing over 4 years (71.2% from previous Q, likely 60-70%, possibly higher from recent whitepaper claiming higher GPU utilizations). This gap compounds as scale and utilization rise. Miners give away a piece of every dollar they earn while Nebius INCREASES gross margins over time. 2. Software Is the Moat and Oracle Proved It Even $ORCL, an $800B hyperscaler, failed at building GPU orchestration profitably, reporting ~14% gross margins on their AI rental platform and losing ~$100M in the process. If Oracle can’t execute, expecting $100M marketcap small miners like $SLNH to build hyperscaler-grade orchestration from scratch is wishful thinking. Platforms like Fluidstack are essential bridges, but they come at a cost: margin compression, revenue leakage, and platform dependency. Nebius? It already did the hard part. Its in-house orchestration software, GPU utilization, with no middlemen taking a cut. 3. Power vs. Margin Calculations A single H100 uses ~0.7–0.84 kW. Even at $0.10/kWh, power is just $0.07–0.08 per GPU-hour. When GPUs rent at $4–6/hr, that’s 1–2% of revenue. What actually moves the needle? Utilization: 50% -> 85% uptime = +70% revenue per GPU That’s a multi-hundred bps margin swing, far outweighing any "cheap power" comparison with miners. Software utilization/orchestration is a moat and matters ~30–70 TIMES more than cheap power. And it's not like $NBIS doesn't have cheap power either lol. 4. The CRWV Exception to Miners Everyone points to $CRWV as miners pivot but Coreweave literally spent YEARS to do this, along with billions in software acquisitions. And even now, reports suggest their utilization trails Nebius. If miners think they’ll replicate CRWV, good luck. I'd expect in NBIS to strong outperform in next year's earnings reports when it comes time for execution vs. speculation. 🧩 The Asymmetry Nebius GM: 70–75% (full-stack, 4yr depreciation) IREN/CIFR realistic GM: 40–60% (middleware, 2–3yr depreciation) Gross margin delta: 15–30 points Execution risk: Nebius = 0 (already doing it); IREN/CIFR = high Nebius isn’t a power play. It’s a software margins play, with hardware upside. With $IREN and others, you're guessing if it can pull off a $CRWV. People keep saying "software it not a moat, and look at $CRWV for $IREN future HPC margins", but if $ORCL one of the largest hyperscalers is stuck at 14% gross margins. Then how do we expect these small crypto miners to pull off a Coreweave. In terms of ASYMMETRICAL UPDATE, $NBIS is the clear favorite out of anything. In terms of raw potential upside if $IREN manages to pull a $CRWV out a hat and beats out $ORCL in this, sure I'd concede. Software full stack is a HUGE moat that people vastly, vastly underestimate. With $NBIS, they can just let it plug and play scale up over time with the highest gross margins in industry. $NBIS is what true asymmetric return looks like.

在 X 上查看原推 ↗