供应链分析
产业链结构、上下游与瓶颈环节研究 · 共 1315 条 · 滚到底自动加载更多
-
澄清NBIS产能售罄非瓶颈，指出新云厂商执行与第三方依赖风险。
我仅针对之前关于 $NBIS 的错误框架进行回应，因为我也看到其他人有类似说法。它并未触及产能天花板，而是新上线的产能立即售罄，这是积极信号。不过你的后续观点是正确的，新云厂商在转换和连接产能方面面临执行风险，正如你提到的，$CRWV 是第三方依赖风险的良好案例研究。
英文原文
I'm only addressing the earlier incorrect framing because I've seen it from others too when talking about $NBIS. It's not hitting capacity ceiling. It's immediately selling out of capacity they bring online, which is positive. Your follow up is correct though, it's an execution risk that Neoclouds face converting and connecting capacity and $CRWV is a good case study as you mentioned on 3rd party dependency risks.
-
@KobeissiLetter 又有400亿美元资本支出流向像 $WULF / $CIFR 这样的新云服务商(neoclouds)以及 $GOOGL 的数据中心建设。与此同时，市场正因错误的原因而无差别抛售。 https://t.co/S9wmV8d1XD
英文原文
@KobeissiLetter Another $40B in capex going to neoclouds like $WULF / $CIFR and the data center buildout by $GOOGL. Meanwhile markets are selling off indiscriminately for the wrong reasons. https://t.co/S9wmV8d1XD
-
驳斥AI板块恐慌，指出托管商受Mag7支持，是错杀买入机会。
这是$CRWV、$ORCL的OpenAI效应。市场在恐惧时是非理性的，将整个板块视为单一实体，而忽视了细微差别： GPU折旧——Burry的做空论点（硬件快速过时）： $CIFR、$WULF等完全不受影响，因为它们不购买GPU，而是做数据中心托管(Colo)。 信用风险——收紧使得所有资本密集型的AI建设成本更高且更脆弱。 $NBIS等...是孤立的（资产负债表强劲，可转换债务）。 未来收入泡沫，OpenAI没有足够的... $IREN、$NBIS、$CIFR、$WULF完全不受影响，因为它们由Mag7支持，而非OpenAI。 对某些人来说，这是令人发笑的买入机会。
英文原文
It's the $CRWV, $ORCL OpenAI Effect. Markets are irrational during fear and views the sector as a single entity vs. seeing the nuances: GPU Deprecation - Burry Short Thesis (rapid hardware obsolescence): $CIFR, $WULF, etc. Literally not affected since they don't buy GPUs and do colo. Credit Risk - tightening makes all capital-intensive AI buildouts more expensive and fragile. $NBIS and others... isolated (Strong balance sheet, convertible debt). Future Revenue Bubble with OpenAI not having enough $IREN, $NBIS, $CIFR, $WULF literally not affected since they're backed by Mag7, not OpenAI. Hilarious buying opportunity for some.
-
IREN转全栈致风险激增，NBIS/CIFR基本面稳健却被板块性做空。
谢谢，是的，我之前做分级列表时将 $IREN 列为 S 级，但他们因试图通过购买数十亿 GPU 与 $CRWV 在全栈领域竞争而被大幅降级。许多 $IREN 持有者争论称托管(Colo)最盈利，但在 $IREN 的财报电话会上，他们明确表示因“更高的收入潜力和更强的回报”而购买 GPU 并从事全栈业务。构建软件层比人们想象的要难得多，$IREN 的执行风险急剧上升。与此同时，$NBIS 和 $CRWV 已经掌握了这一环节（根据其白皮书，$NBIS 胜过 $CRWV），但 Coreweave 在利息债务和 OpenAI 风波中搞砸了。不幸的是，我们看到整个板块被做空，尽管存在许多细微差别，例如 $CIFR 不受 GPU 贬值论点的影响。或者 $NBIS 拥有超过 5 年的 Mag7 及企业客户、强劲的资产负债表，且不受 $ORCL 及信贷收紧引发的 AI 泡沫 OpenAI 风波影响。
英文原文
Thanks, yeah I had $IREN as S tier when I did my tierlist earlier but they got downgraded a lot because they ended up trying to compete vs $CRWV in the full-stack by buying billions in GPUs. A lot of IREN holders were arguing colo was most profitable, but in $IREN's earnings call, they literally said they were buying GPUs and doing the full-stack due to "higher revenue potential and stronger returns". It's much, much harder building out the software layer than people think and the execution risk for IREN skyrocketed immensely. Meanwhile $NBIS and $CRWV already has that nailed down ( $NBIS wins vs. $CRWV per their whitepaper) but Coreweave messed up with interest debt + OpenAI drama. Unfortunately we're seeing a whole sector short despite many of the nuances eg. $CIFR not being affected by GPU depreciation arguments. Or $NBIS having Mag7 + enterprise customers over 5Y periods + strong balance sheet and not affected by AI bubble OpenAI fiasco with $ORCL + credit tightening.
-
新云生态分化，NBIS领先，IREN执行存疑，市场尚未完全定价。
新云(Neocloud)并非赢家通吃，而是一个完整的生态系统，但并非所有公司都能成为AWS。 我们看到$NBIS、$IREN、$CRWV和$ORCL在真正的“新云”领域竞争，这是一场最有利可图的全栈竞赛，旨在成为Azure/AWS。它们进行垂直整合、建设容量、购买GPU并进行编排，一些孵化企业也加入了竞争。 目前$NBIS是明确的赢家，拥有超大规模云厂商+企业客户+良好的资产负债表，只是时间问题；$CRWV深陷利息债务；$ORCL建设失败，且OpenAI是其最大且风险最高的积压订单。$IREN前景看好，但执行层面存在最大问号，因为其路径不同于人们预期的$CIFR模式，其容量和软件编排比人们预期的难得多（看看$ORCL和$CRWV在软件收购上花费的数十亿美元）。 然后是$CIFR、$WULF、矿工等，它们利用现有容量进行更安全的托管机房(Colo)业务，而像Amazon这样的超大规模云厂商则插入自己的GPU。我仍将其归为新云，但也许该单独分类了，不过它们不受Burry关于2年GPU淘汰论虚假言论的影响。 其余公司正在向高性能计算(HPC)转型，承诺未来X量的容量，但在信贷收紧背景下，为此转型融资越来越难。 随着近期一系列财报的发布，我们开始看到新兴赢家，这是好事。市场尚未定价，但只是时间问题。
英文原文
Neocloud isn't winner takes all, it's a whole ecosystem but not everything can become AWS. We're seeing $NBIS, $IREN, $CRWV, and $ORCL compete in the true "Neocloud", most profitable full stack race to become Azure/AWS. Doing vertical integration, building capacity, buying GPUs, and doing orchestration, and some incubants entering the race. So far $NBIS is the clear winner with hyperscalers + enterprises + good balance sheet in a matter of time, $CRWV is swamped with interest debt, $ORCL failed buildout so far and OpenAI as their biggest + risky backlog. $IREN is promising but they have the biggest question mark for execution since it's not the original approach like $CIFR people were expecting with their capacity and software orchestration is much harder than people expect (looking at $ORCL then $CRWV spending billions on software aqusitions). Then there's $CIFR, $WULF, miners, etc and others do safer colo with their existing capacity while hyperscalers like Amazon plugs in their own GPUs. I still put them as a Neocloud but I guess it's time to make a separate category, but these aren't as affected by Burry's false comments about 2Y GPU deprecation lol. And then we have the rest... undergoing the pivot to HPC, with X amount of future capacity promised and this is becoming increasingly difficult to raise money for this change from credit tightening. It's a good thing that we're starting to see emerging winners after the series of recent earnings. Market hasn't priced this in yet but it's a matter of time.
-
XRP作为机构结算机制的优势与地址、余额等技术痛点分析。
自费用被销毁以防止通过供应销毁进行DDoS攻击以来，没有人... 这是一个过度简化的说法，但更类似于机构社区志愿者工作，以维护网络，而不是像ETH PoS那样有费用激励。只是不同的网络共识和方式。 我看到XRP被采用为机构间的结算机制（类似于Customers Bank Instant Token），并且由于即时最终性/低费用，我个人为跨境结算部分构建了一些应用程序（与美国银行代币相比），但在地址和最低余额方面有很多烦人的问题，就像Polkadot在入职/死资本方面一样，并且不像EVM那样广泛采用的脚本，你可以在Polygon到L2上部署相同的合约。
英文原文
Nobody since fees are burned to prevent ddos via supply burn. Oversimplification but it’s more like institutional community volunteer work in terms of maintaining the network rather than having fee incentive like eth pos. Just different network consensus and approach. I see xrp being adopted as a settlement mechanism in between institutions (kind of like Customers Bank Instant Token) and personally built some applications for cross country settlement part (vs. US bank tokens) just due to instant finality/low fees, but there’s a lot of annoying issues regarding addresses and minimum balances like Polkadot with onboarding/dead capital, and not widely adopted scripting like EVM where you can deploy the same contracts across polygon to l2.
-
信贷收紧利好资产负债表健康的T1新云厂商，当前抛售是买入良机。
你的分析精准捕捉到了二阶Alpha。信贷收紧削弱了弱势竞争对手融资新建产能+改善远期自由现金流/利润率的能力，从根本上制约了其增长。 这意味着资产负债表健康且无高额债务利息的T1新云厂商（如$NBIS和$CIFR）将享有更持久的利润率。这是一个巨大的长期溢价，市场目前定价错误+在此次抛售中尚未计入。 正如我们所见，Mag7的资本支出正在增加，但正流向$NBIS、$WULF等赢家。由于$CRWV和$ORCL引发的恐慌+信贷收紧，我们正目睹一场不分青红皂白的抛售，这反而是买入机会，实际上也是受冲击最小的新云厂商的竞争优势。
英文原文
Your analysis nails the second-order alpha. Credit tightening hurts the ability of weaker competitors to fund new capacity + improve forward FCF/margins, fundamentally constraining growth. This translates to more durable margins to the T 1 Neoclouds like $NBIS and $CIFR with healthy balance sheets and no high debt interest. This is a massive long term premium the market is currently mispricing + hasn't priced in yet with this sell-off. Mag7 capex as we've seen is increasing, but it's funneling into winners eg. $NBIS, $WULF, etc. We're seeing an indiscriminate sell-off due to panic from $CRWV and $ORCL + credit tightening, and this is a buying opportunity and actually a competitive advantage for the least affected Neoclouds.
-
AI股因信贷收紧恐慌抛售，建议转向有Mag7合同且低债的新云龙头。
我们正目睹 AI + 数据中心(DC) 股票实时崩盘，$IREN 本周下跌 37.88%，$NBIS 本月下跌 35.27%。 恐慌源于信贷收紧、高息债务、OpenAI 传染效应、宏观因素，以及最重要的：非理性恐惧。 以下是实时发生的情况： 1. 信贷收紧 - 像 $APLD 这样的公司未能成功发行债券（认购不足），最终发行了 23.5 亿美元的高收益债务融资——彭博社 11 月 14 日报道。债券以 97 美分的价格折价出售，收益率约为 10%。 该债券交易的风险主要与 Applied Digital 对 $CRWV 的依赖有关，后者占其合同收入的大部分。随着建设阶段机构关注度增加（例如 Burry 关于 GPU 折旧的做空报告），数据中心更难筹集资金。 融资条件收紧影响了整个通过发行债务为自身建设融资的生态系统。 2. 高息债务 - 由于信贷收紧，$CRWV 为扩张而承担的劣质债务导致每年超过 10 亿美元的损失。这侵蚀了利润率和自由现金流(FCF)。 同样，$APLD 和其他以不良收益率筹集资金的公司也面临同样的未来问题，我们看到相关公司因风险管理而出售。 3. OpenAI 承诺的 1 万亿美元+ 资金没有余额支撑 - OpenAI 承诺超过 1 万亿美元的资本支出(capex)，其中 224 亿美元+ 流向 $CRWV，3000 亿美元+ 流向 $ORCL，他们正在为此建设产能。由于 OpenAI 手头没有余额并采取避险策略，市场现在对这些承诺持怀疑态度。 现在，我们看到这种建设从 Coreweave 等公司向下传染到 $CORZ 和 $APLD，这些公司在建设期间可能依赖 $CRWV 的未来收入。 4. $CRWV 是 AI 基础设施（以及部分 $ORCL）的行业领导者。当行业领导者抛售时，通常其他公司会跟随。 5. 12 月降息概率下降： 除了高息债务外，我们看到 12 月降息概率降至约 53% 的抛硬币水平。较高的利率使得像 $CRWV 这样背负利息债务的公司再融资以增强 FCF 变得困难。 _ 细微差别与重新定位： 我们看到全面抛售。然而，Mag7（$META, $GOOGL, $AMZN, $MSFT）的资本支出和合同已确认，且盈利能力极强。 有一种简单的方法可以安全重新定位，以避免未来收入增长 + 债务周期的问题。 1. 配置与 Mag7 有合同可见性的新云(Neoclouds)： - $NBIS - 190 亿美元 $MSFT 交易，30 亿美元来自 $META，广泛的企业支持 - $CIFR - 55 亿美元与 $AMZN 的 Colo 交易，30 亿美元与 $GOOGL 的交易 - $WULF - 37 亿美元+ 与 $GOOGL 的交易等。 - $IREN - 90 亿美元与 $MSFT 的交易 我们可能会看到更多像 $SLNH, $CLSK 等投机性容量公司下跌，并整合到具有可见企业合同（未来收入下行风险低）的新云中。 2. 配置无/极低息债务且资产负债表强劲的新云： 像 $NBIS 这样的公司因拥有 47.9 亿美元的可转换票据现金及现金等价物，而与当前信贷市场波动隔离。或者像 $CIFR 这样拥有 12 亿美元现金的公司。 3. 避免与 OpenAI 有直接合同而非 Mag7 的公司。 _ 新云是 Mag7 AI 算力的骨干。这一论点没有改变，我们可能会看到一些规模达到 1000 亿美元+ 的公司成为下一个 AWS。 然而，我们看到由 $CRWV, $ORCL 和 OpenAI + 信贷收紧（部分由于 Burry 关于 GPU 折旧的错误机构做空报告）引起的广泛、不分青红皂白的抛售。 对于像 $NBIS 这样拥有 70-90 亿美元远期年度经常性收入(ARR) 和极强资产负债表的公司，这种抛售完全不合理。但对于其他投机性公司，这是一次必要的修正。 一句好话是“散户最后卖出，最后买回”。 如果你认为你可以卖出并在两周后回来，错过抛售后的一个或两个巨大反弹日可能会让你失去大部分恢复收益。 机构可能正在通过暗池/大宗交易购买更好的名字，例如 $NBIS，在这些清洗和保证金清算之后。我们还看到机构所有权从 38%-> 43-46%，尽管股价下跌。 利用这次机会购买好名字，并持有度过这些波动时期。 （该推文引用了 @aleabitoreddit 的推文，引用内容仅供理解语境）: 13F 新云申报的机构流向评估： · Nebius ( $NBIS ): 🟢 强烈正面 (7/10) · WULF ( $WULF ): 🟢 高度正面 (8.5/10) · IREN ( $IREN ): 🔴 非常负面 (3/10) · CIFR ( $CIFR ): 🟢 高度正面 (8.0/10) · Coreweave ( $CRWV ): 🟡 中性 (5.5/10) · Cleanspark ( $CLSK ): 🟢 高度正面 (9.0/10) _ TLDR 摘要更新： $NBIS · 机构所有权整体定量增长强劲，由坚实的长期机构买家以及量化基金和对冲基金的高活跃度驱动。 $WULF · 机构基础结构稳定且安全，以 Vanguard 和 BlackRock 等大型纯多头资产管理人为特征，他们对股票持有高信念。 $IREN · 所有权由高频交易员、做市商和量化基金（例如 Jane Street, Citadel）主导。这种结构被视为负面，因为它表明缺乏稳定的短期持有压力和长期机构缺乏信念。 $CIFR · 优秀的机构设置，以纳入主要指数（被动需求）和主动基金的强烈信念购买为标志，Alyeska 的大幅增持尤为突出。 $CRWV · 高交易量表明市场兴趣浓厚，但所有权目前高度集中，做市商存在显著，导致中性评分，暗示流向的风险/回报平衡。 $CLSK · 被认为是该组中最好的机构结构，结合了广泛支持的被动指数纳入和激进的、高信念的主动积累，表明定性风险最低。 _ 评论： 例如，“看多 - Jane Street + Citadel 拥有 x 的 $IREN ”（提示：这不是正面的）。尤其是当长期持有者如 FMR (Fidelity)：212 万股，大幅削减持仓时。 基金类型很重要，例如： 被动（最正面） Vanguard, BlackRock 指数, State Street, Geode 等 多头（正面） Fidelity, T. Rowe, Wellington, BIT 等 对冲基金（中性到正面） Alyeska, Coatue, Millennium 等 量化/做市商（负面） Jane Street, Citadel, Susquehanna, SIG, Two Sigma 等 这是一个主观框架，用于仅从本季度评分流向有多正面（而非整体所有权）。
英文原文
We're seeing a live crash in AI + DC stocks right now with $IREN down 37.88% this week to $NBIS down 35.27% this month. Panic stems from Credit Tightening, High Interest Debt, OpenAI contagion, Macro, and most importantly: Irrational Fear. Here's what happening real time: 1. Credit Tightening - Companies like $APLD failed to sell bonds (under subscribed) and ended up issuing $2.35B in high yield debt offerings - Bloomberg Nov 14th. The bonds were sold at 97 cents on the dollar, a significant discount, with a yield of approximately 10%. The bond deal's risk is primarily tied to Applied Digital's reliance on $CRWV, which accounts for a substantial portion of its contracted earnings. More institutional cation (eg. Burry short report on GPU depreciation) around the buildout makes it harder for DC's to raise funds. Tightening funding conditions affect the entire ecosystem that issue debt to fund their own buildout. 2. High Interest Debt - Due to credit tightening, bad debt, which $CRWV has to fund expansion, causes $1B+ in losses a year. This cuts into margins and FCF. Likewise, $APLD and others that raise funds with bad yields face the same future issues and we're seeing a selloff in related companies for risk-management. 3. $1T+ in OpenAI promised funding without balances - Over $1 trillion+ in capex was promised by OpenAI going to $22.4B+ $CRWV and $300B+ to $ORCL, which they are building capacity for. The market is now skeptical of these promises from OpenAI due to not having the balances on hand and taking a risk-off approach. And now we're seeing contagion on that buildout from companies like Coreweave spread down to $CORZ and $APLD during this buildout that might rely on $CRWV for future revenue. 4. $CRWV is a sector leaders of AI infrastructure (and partially $ORCL). When the sector leader sell-off, usually others follow suit. 5. Rate Cut odds in December Fall: On top of the high interest debt, we're seeing rate cut odds in December drop to a coinflip ~53%. Having higher interest rates make it trouble for interest-debt burdened companies like $CRWV to refinance for stronger FCF. _ THE NUANCE and Repositioning: We're seeing a broad sell-off across the board. However, Capex and contracts from Mag7 ( $META, $GOOGL, $AMZN, $MSFT ) are confirmed with their extreme profitability. There's an easy way to safely reposition to avoid issues with future revenue growth + debt cycles. 1. Position into Neoclouds with contract visibility with Mag7: - $NBIS - $19B $MSFT deal, $3B from $META, broad enterprise support - $CIFR - $5.5B Colo deal with $AMZN, $3B deal with $GOOGL - $WULF - $3.7B+ deals with $GOOGL and more. - $IREN - $9B deal with $MSFT We will likely see more capacity speculative companies like $SLNH, $CLSK, and others fall off and consolidation into Neoclouds with visible enterprise contract (low downside risk of future revenue). 2. Position into Neoclouds with no/extremely low interest debt and high balance sheets: Companies like $NBIS are isolated from current credit market volatility as they have $4.79 billion in cash and cash equivalents from convertible notes. Or companies like $CIFR with $1.2B in cash. 3. Avoid companies with direct contracts with OpenAI that's not Mag7. _ Neoclouds are the backbone of Mag7 AI compute. This thesis has not changed and we'll likely see some scale to $100B+ companies in the future as the next AWS. However, we're seeing a widespread, indiscriminate sell-off caused by $CRWV, $ORCL, and OpenAI + credit tightening (partially due to incorrect institutional short reports from Burry on GPU depreciation). These types of sell-offs are completely not warranted for companies like $NBIS with $7-9B forward ARR and extremely strong balance sheets. But for other speculative companies, it's a well-needed correction. A good quote is “retail is last to sell, last to buy back”. If you think you can sell and come back two weeks later, missing one or two huge rebound days after a selloff can cost you most of the recovery gains. Institutions are probably buying the better names right now eg. $NBIS through dark pools/block trades now after these types of washouts and margin liquidations. We've also seen institutional ownership go from 38%-> 43-46% despite share prices dropping. Use this as a buying opportunity for good names, and hold through these periods of volatility.
-
高贝塔AI股因宏观恐慌错杀，基本面强劲者终将迎来价值回归。
我们正目睹高贝塔值AI股票板块的崩盘，部分个股在1个月内下跌30-45%+，今日再跌10%+。 这包括直接受益于“七巨头”资本支出的公司： - Nebius ($NBIS)：$META 30亿美元交易 / 80亿美元远期年度经常性收入(Forward ARR)。 - TeraWulf ($WULF)：Anthropic数据中心(<500亿美元) + 与$GOOGL的合资光纤网络(FS)。 - Iren ($IREN)：$MSFT 90亿美元高性能计算(HPC)。 - Cipher ($CIFR)：与$AMZN签署15年55亿美元合同。 这些公司在过去一个月里改善了基本面。 但市场仍在恐慌性抛售从$CRDO (-10.33% 1D)到$CIFR (-7.28% 1D)的股票，尽管它们拥有创纪录的远期收入增长。 这主要是由宏观担忧（政府停摆、12月降息、公司债利差等）和虚假（但微妙）的机构观点（如Burry关于GPU折旧的说法）驱动的，而非基于个股基本面。 这清洗了AI投机性板块（如能源股$OKLO，1个月跌37%）中的泡沫，但高利润率的增长型股票也受牵连。 我之前说过，这让我想起$HOOD从$65跌至$28后反弹至$150，或$ALAB在所有财报超预期后从$100跌至$50再反弹至$245。 当市场充满恐惧，尤其是像现在这样的“AI泡沫”恐慌时，股价有时会与基本面脱节。 但当远期增长能支撑市值预期时，这只是时间和执行力的问题。
英文原文
We're seeing a sector crash in high-beta AI stocks, with names down 30-45%+ in 1M & 10%+ today. This includes direct beneficiaries of Mag7 Capex : - Nebius ( $NBIS ): $3B $META deal / $8B forward ARR. - TeraWulf ( $WULF ): <$50B Antrophic DC + JV FS/ $GOOGL - Iren ( $IREN ): $9B $MSFT HPC - Cipher ( $CIFR ): 15Y $5.5 billion w/ $AMZN. that improved their fundamentals in this past month alone. But markets are still panic selling names from $CRDO (-10.33% 1D) to $CIFR (-7.28% 1D) despite record forward revenue growth. This is driven over fears from macro (shutdown, Dec rate cut, corporate bond spreads, etc) and false (but nuanced) institutional claims eg. Burry on GPU depreciation, rather than individual fundamentals. This wipes away froth from the speculative parts of the AI sector such as $OKLO (energy, -37% 1M), but higher-margin growth names get caught together with it. I've said this before, but this reminds me of the time $HOOD sold off to $28 from $65 then pulled off a rally to $150. Or when $ALAB sold off from $100 to $50 off all earnings beats, then rallied to $245. Sometimes stock prices get detached from fundamentals when there’s fear in the markets, especially with “AI bubble” fears like what's happening now. But when the forward growth backs up any market cap projections, it’s just a matter of time and execution.
-
反驳 Burry 观点，指出旧款 GPU/TPU 仍高利用率且具残值。
关于 $ORCL、$NBIS 及其他新云厂商 GPU 折旧论点的简要总结： Burry 的一般主张：处于 2-3 年产品周期的芯片资本支出（Capex）不应导致有用寿命的延长。 这一观点错误但具有细微差别： -> 以 $GOOGL 为例，其 7 年前的张量处理单元(TPU) 仍以 100% 利用率运行。 “谷歌称 TPU 需求超过供应，声称 8 年前的硬件迭代‘利用率达 100%’”——数据中心动态 -> 以 $NVDA 为例，A100（PCIe 和 SXM4 变体）仍在运行。它们于 2020 年发布，现在是 2025 年。旧型号保持价值并仍能带来收益。 它们具有残值，5 年后仍能高价转售。他在货架寿命上错了，但在某些超大规模云服务商如何拉伸利用率概念以证明激进资本支出合理性方面，他可能是对的。
英文原文
Just a TLDR on the GPU depreciation argument for $ORCL, $NBIS and other neoclouds, Burry's general claim: capex of chips on a 2-3 yr product cycle should not result in the extension of useful lives. This is incorrect but nuanced: -> in $GOOGL's case, their TPUs from 7 years ago are still run at 100% utilization. " Google says TPU demand is outstripping supply, claims 8yr old hardware iterations have “100% utilization” -Data Center Dynamics -> in $NVDA's case A100 (PCIe and SXM4 variants), are still in operation. They were launched in 2020. It's 2025 now. Older models keep their value and still deliver equity. They have residual value and still re-sale for a lot 5 years later. He's wrong on shelf-life, but might be correct in how some hyperscalers stretch the idea of utilization to justify aggressive capex.
-
新云财报解读及反驳伯里GPU贬值论，强调新云逻辑强劲。
今日新云(Neocloud)财报 + 为何迈克尔·伯里(Michael Burry)错了： $WULF - 财报超预期(ER)表现积极，来自520兆瓦(HPC)算力的高性能计算(HPC)容量总合同价值(TCV)达170亿美元。 在德州成立240兆瓦合资企业(JV)用于数据中心(DC)扩张 + 50亿美元融资。 不像$CIFR与AWS的财报超预期那样特别，只是确认（如交易融资）和站点开发。 $CRWV - 因第四季度远期收入较低，财报超预期表现短期看空。 556亿美元积压订单（其中224亿美元来自OpenAI）。小幅盈利超预期13.6亿 vs 12.9亿（同比129%），EBITDA利润率（61%）。 股价因数据中心延迟下跌，这损害了2025财年（50.5-51.5亿 vs 共识52.9亿）。运营利润率下调，正如我预测的那样，因为债务融资产生的利息（~10亿+利息/年）侵蚀了利润率，而其他新云没有同样的问题。 话虽如此，这是一次性问题，鉴于市场是前瞻性的，我们可能会看到任何抛售在下个季度后被买回。 _ 至于迈克尔·伯里关于GPU贬值的争论：他完全错了，但在低估这一点上部分正确。 他的普遍主张：芯片在2-3年产品周期上的资本支出(Capex)不应导致使用寿命的延长 -> 在谷歌的案例中，他们7年前的TPU正以100%的利用率运行。重复一遍，是7年前。 Amin Vahdat表示，谷歌目前生产中有七代TPU硬件，且“七到八年前的TPU拥有100%的利用率”。 “谷歌称TPU需求超过供应，声称8年前的硬件迭代‘100%利用率’” - 数据中心动态 -> 在英伟达(Nvidia)的案例中，A100（PCIe和SXM4变体）仍在运行。它们于2020年推出。现在是2025年。旧型号保持其价值并仍然提供收益。它们不会仅仅因为是补充新模型的附加组件就被核销。 英伟达的A100可以划分为隔离实例以最大化利用率，旧一代加速器仍用于推理或低优先级任务进行货币化，而不是像他所声称的那样被核销。它们实际上具有残值，5年后仍能高价转售。 他在保质期问题上完全错了，但在某些超大规模云服务商如何拉伸利用率概念以证明激进资本支出方面可能是正确的。 这是一个微妙的论点，但新云逻辑依然完好且比以往任何时候都更强。
英文原文
Neocloud Earnings Today + Why Michael Burry is Wrong: $WULF - ER was positive, $17B TCV from 520 MW HPC capacity. 240 MW JV in Texas for DC expansion + $5B financing. Nothing special like $CIFR ER with AWS, just conf (eg. deal finances) and site dev. $CRWV - ER was bearish near term due to lower forward revenue Q4. $55.6B backlog ($22.4B from OpenAI). Slight earnings beat 1.36B vs. 1.29B (129% Y/Y), EBITDA margins (61%). Stock dropped on datacenter delay, this hurts FY 2025 (505-5.15B vs. 5.29B consensus). Operating margins was lowered, which I predicted due to debt financing with interest (~1B+ interest/year) that cuts into margins, while other Neoclouds don't have that same issue. That being said this was a one-off issue, and we'll likely see any sell-off bought back after next quarter since markets are forward looking. _ As for Michael Burry arguing on GPU deprecation: he's completely wrong, but partially right about understating it. His general claim: capex of chips on a 2-3 yr product cycle should not result in the extension of useful lives -> in Google's case, their TPUs from 7 years ago are being run at 100% utilization lol. Repeat 7 years ago. Amin Vahdat, said that Google currently has seven generations of its TPU hardware in production, and that the “seven- and eight-year-old TPUs have 100 % utilization.” "Google says TPU demand is outstripping supply, claims 8yr old hardware iterations have “100% utilization” -Data Center Dynamics -> in Nvidia's case A100 (PCIe and SXM4 variants), are still in operation. They were launched in 2020. It's 2025 now. Older models keep their value and still deliver equity. They don't just get nuked since it's an add-on to compliment newer models. Nvidia’s A100 can be partitioned into isolated instances to maximize utilization, and older-gen accelerators are still monetized for inference or lower-prio tasks rather than getting nuked as what he claims. They actually have residual value and still re-sale for a lot 5 years later. He's completely wrong here on shelf-life, but might be correct in how some hyperscalers stretch the idea of utilization to justify aggressive capex. It's a nuanced argument but Neocloud thesis remains in tact and stronger than ever.
-
基本面强劲支撑AI板块，短期回调因宏观因素，建议逢低加仓。
忽略噪音，持有仓位或在市场赢家身上加仓。 你拥有： - 来自 $NBIS、$IREN、$CIFR 等新型云服务商(Neoclouds)的惊人远期营收增长 - 来自 $META 到 $MSFT 等七巨头(Mag7)财报创纪录的高资本支出(CapEX)预期 - 来自 $TSM 等半导体公司创纪录的高远期营收预期+利润率增长 如果基本面完好，像“AI派对结束了”这样的新闻头条就是胡扯。 随着政府停摆结束，我们很可能会看到重大复苏。 这很可能是一次短期回调，定价了诸如没有第三次降息（概率从86%降至70%，但仍有可能）+ 政府停摆延长至16日之后（合约市场显示53%的概率，两天前为30%）等因素。
英文原文
Ignore the noise and hold positions/add in market winners. You have: - Insane forward revenue growth from Neoclouds like $NBIS, $IREN, $CIFR - Record high CapEX projections from Mag7 earnings like $META to $MSFT - Record high forward revenue projections + margin growth from semiconductors like $TSM. If fundamentals are in tact, news headlines like "The AI Party is Over" is BS. We'll likely see a major recovery when government shutdown ends. This is likely a short-term correction pricing in a few things such as no third rate cut (86% -> 70%, still likely) + government shutdown extending past the 16th (53% chance, 30% chance 2 days ago) per contract markets.
-
推测OSS为Anduril供应商，基于AI边缘计算需求匹配及AI模型概率分析。
披露：我目前无持仓，且不计划建仓。（原因：已持有过多高贝塔股票）。 我想发布一些关于 Anduril 供应链公司的投机性研究，这是在研究 $KRKNF | Kraken Robotics 之后进行的。 这是 $OSS（市值 1.1554 亿美元）。 简而言之：Anduril 为美国陆军推出了 Effects-Short Range (LE-SR) 系统。OSS 是一家边缘 AI 基础设施公司。尽管 Anduril 收购了竞争对手 Klas（一家坚固型 AI 边缘计算公司），但 $OSS 似乎是 Anduril 的供应商。 OSS 宣布获得一份 650 万美元的合同，交付 80 台“高性能服务器和现场可编程门阵列(FPGA)系统……用于支持美国军事行动的移动情报平台”。 1. OSS 的专业领域：为移动战术军事环境中的边缘 AI/ML 提供高性能、加固型 FPGA 和服务器系统。 2. Anduril 的需求：LE-SR (Altius 600) 平台需要坚固、紧凑的计算能力，以便在战术边缘自主运行其 Lattice AI 软件。 _ 650 万美元用于 80 台 3U SDS 加固服务器和 FPGA 系统。功能包括传感器数据采集、AI 生成的实时分析以及战术环境下的存储：这与 Lattice 对集群管理的需求相符： 硬件：AMD EPYC 9004 “Genoa” 处理器，最多七个 PCIe Gen4 x16 插槽用于 GPU/FPGA 集成，支持热插拔存储高达 1 PB，并针对 -40°C 至 70°C 的温度进行加固。 适合指挥控制(C2)：这些系统适合地面枢纽而非无人机载荷，支持操作员在环接口和多资产协调。 ________ 3. 买家被描述为一家“领先的国防和技术解决方案公司”。OSS 强调向国防领域转型（目前约占收入的 50%），此交易是其一系列成果的一部分，包括 600 万美元的国防部(DoD)续约和其他军事订单。 唯一缺失的是客户名称，鉴于国防承包商的性质，这通常保密，但看起来像是 Anduril，且是回头客（与该客户的额外开发和平台机会正在进行中）。 关于 OSS <-> Anduril 合同和信息的更深入内容是通过前沿模型 Deep Research 工具获取的： Gemini - 85% 的概率 Anthropic Claude - 75-80% 的概率 OpenAI ChatGPT - 70% 的概率 XAI Grok - 40-45% 的概率 认为 OSS 是 Anduril 的供应商。 如果 OSS 确实是 Anduril 的供应商并从 650 万美元的合同扩展，那么鉴于该国防承包商带来的未来收入潜力，其市值很可能从 1.2 亿美元向上重估。如果不是，鉴于其他美国军方合同，它目前的市值也是合理的。 支持 Anduril 使用 OSS 这一假设的间接证据似乎很有说服力。 我想在财报发布前发布这个看看是否正确（如果他们提到的话）。不过我不打算在此建仓，因为像 $NBIS 这样的高贝塔股票具有更高的非对称回报，但如果 $OSS 涨 5 倍，我会很遗憾。 我偶尔会研究一些小公司看看是否值得投资，但只是想出于好玩发布这个看看是否正确。
英文原文
Disclosure: I have no positions and don't plan on initiating any. (Reason: have too many high-beta stocks). I just wanted to publish some speculative research I did on Anduril supply chain companies after researching $KRKNF | Kraken Robotics. This is $OSS ($115.54M MC). TLDR: Anduril launched Effects-Short Range (LE-SR) for the US Army. OSS is an edge AI infra. $OSS seems like a Andruil supplier, despite Andruil aquiring Klas, a competing rugged AI edge company. OSS has announced a contract of $6.5 million to deliver 80 “high-performance servers and field-programmable gate array systems… for a mobile intelligence platform supporting U.S. military operations.” 1. OSS's specialization: High-performance, ruggedized FPGA and server systems for AI/ML at the edge in mobile, tactical military environments. 2. Anduril's requirement: The LE-SR (Altius 600) platform requires rugged, compact computing to run its Lattice AI software autonomously at the tactical edge. _ $6.5 million for 80 3U SDS rugged servers and FPGA systems. Functions include sensor data collection, AI-generated real-time analysis, and storage in tactical settings: mirroring Lattice's requirements for swarm management: Hardware: AMD EPYC 9004 “Genoa” processors, up to seven PCIe Gen4 x16 slots for GPU/FPGA integration, hot-swappable storage up to 1 petabyte, and ruggedization for -40°C to 70°C temperatures. Fit for C2: The systems suit ground-based hubs rather than drone payloads, supporting operator-on-the-loop interfaces and multi-asset coordination. ________ 3. The buyer is described as a "leading defense and technology solutions company". OSS has emphasized shifting to defense (now ~50% of revenue), with this deal part of a streak including $6M DoD renewals and other military wins The only missing piece is the customer name, which, given the military contractors, is often kept confidential, but it seems like its Anduril and it's a repeat-client (Additional development and platform opportunities are underway with this customer). A more in-depth information about OSS <-> Andruil contracts and info were fed through frontier model Deep Research tools: Gemini - 85% assigned probability Antropic Claude - 75-80% assigned probability OpenAI ChatGPT - 70% assigned probability XAI Grok - 40-45% assigned probability of OSS being the Andruil supplier. In the chance OSS is an Anduril supplier and scales up from the $6.5M contract, then it's likely to be re-rated upward from a $120m marketcap from high future revenue potential from this defense contractor. If not, then it deserves its current market-cap anyway due to other US Military contracts. The circumstantial evidence supporting the hypothesis of Anduril using OSS seems compelling. Thought it would be fun to post this before their ER to see if it's right or not (if they ever mention it). I don't plan on taking positions in this though since other high-beta stocks like $NBIS have higher asymmetrical return, but if $OSS 5x's, I'll be sad. I do look into small companies on the side to see if it's worth investing or not but just wanted to publish this for fun to see if it's right.
-
独立建设促专业化，全栈是护城河，苹果或向上游整合。
@DeepValueBagger 是的，观察得很准，随着各自独立建设，我们很可能会看到专业化分工的出现。 全栈能力如何构建构成了护城河，像 $NBIS 这样的模式目前尚未被复制。 不过，你也可以指出 $APPL 与 $QLCM 的情况，在熊市(🐻case)情景下，苹果正在向上游整合。
英文原文
@DeepValueBagger Yeah good observation, we’ll likely see specializations emerge from separate buildout. It is a moat in how you do full stack and some models like $NBIS haven’t been replicated yet. Then again you can point out $APPL vs $QLCM with Apple sucking back up the stack as the 🐻case
-
Neocloud获微软AWS大额订单，算力需求旺盛，板块极度看涨。
Neocloud 名单已更新： $IREN - IREN 与微软达成 97 亿美元 GPU 云合同，包含 20% 预付款，将在五年内部署 NVIDIA GB300。 $CIFR - CIFR 与亚马逊云科技 (AWS) 达成 55 亿美元、15 年的租赁协议。 随着超大规模云服务商 (Hyperscalers) 的新合同涌入，Neocloud 板块变得极度看涨，这显示了对算力 (Compute) 的无限需求。 我在原始论点中预计 $ORCL、$META 和 $AMZN 会达成交易，看来 AWS 已与 $CIFR 入场，预计与其他 Neocloud 的交易也会接踵而至。$META 最近也与 $CRWV 签署了 140 亿美元的交易，从其不断扩大的 AI 资本支出 (Capex) 来看，我们可能会看到更多与 Neocloud 的交易。 这对整个板块来说都是极度看涨的。 我附上了我的个人梯队排名 (Tierlist) 供娱乐，但随着有更多时间分析交易和利润率，数据中心/矿工板块可能会迎来全面重评。
英文原文
The Neocloud List got updated: $IREN - IREN secured a $9.7B GPU cloud contract with Microsoft, including 20% upfront prepayment, to deploy NVIDIA GB300s over five years. $CIFR - CIFR secured a $5.5 billion, 15-year lease agreement with Amazon Web Services (AWS). The Neocloud sector just got insanely bullish with new contracts coming in from hyperscalers and this just shows insatiable demand for compute. I expected $ORCL, $META, and $AMZN to make deals in the original thesis and looks like AWS entered the fray with $CIFR, with likely more deals incoming with other Neoclouds. $META also signed a $14B deal with $CRWV recently and from their ER scaling AI Capex, we'll likely see other deals with Neoclouds. This is insanely bullish across the whole sector. I attached my personal tierlist for fun, but we'll likely see a re-ratings in data centers/miners across the board when there's more time to analyze deals and margins.
-
Meta 资本支出聚焦前沿模型训练，以应对 AI 竞争。
撇开这点不谈，我同意我认识的人都不会使用 Messenger/WhatsApp AI，虽然不清楚他们如何获得 10 亿月活用户(MAU)，但这只是轶事。Facebook 正在消亡的感知(perception)与现实(reality)略有不同。 资本支出(Capex)很可能流向其前沿模型(frontier model)（例如 $14B crwv 合同和其他数据中心），以便训练更好的模型并与 Gemini/ChatGPT 竞争。
英文原文
That aside, I'd agree nobody I know would use messenger/whatsapp AI, no clue how they got 1B MAU but it's just anecdotal. Perception like Facebook dying is slightly different than reality. Capex is likely going toward their frontier model (eg. $14B crwv contract and other datacenters) so they can train a better model and compete vs. Gemini/ChatGPT.
-
对比不同数据中心租赁期限对收入爬坡及长期竞争力的影响。
我删除了之前的评论，因为 X 平台有 Bug 导致时间线混乱： 简而言之： $APLD 建设具有显著电力容量的数据中心，然后将基础设施租赁给客户。目前有 $110 亿的 $APLD 和 $CRWV 针对 Ellendale 园区的 15 年租赁协议。 Polaris Forge 2 约为 200+ MW，合同价值 $50 亿。从交易者视角来看，我不太看好较长的合同期限（例如 $CIFR | $WULF 的 10 年），相比 $NBIS | $MSFT 的 5 年合同，因为前者收入爬坡较慢且成本存在变数。但这并非真正的利空，因为如果你投资的是 1 年周期，你会希望收入爬坡尽可能快。 但如果看 FluidStack $WULF 的 15 年合同协议，它为公司在长期内提供了更多保障；反之，如果像 $MSFT 这样的超大规模云厂商最终自建容量和定制 ASIC，那么一些 Neocloud 公司在 4 年后可能会陷入困境。
英文原文
I'm deleting my earlier comment to this since X is bugged and it's messing up my timeline: But TLDR from before: $APLD builds data-centers with significant power capacity and then lease that infrastructure to customers. So there's the $11B $APLD, $CRWV lease right now over 15 years for Ellendale campus. Then there's Polaris Forge 2 is ~200+ MW with a $5B contract. Not much of a fan of longer contract terms (eg. 10 Years $CIFR | $WULF), vs. 5 Year $NBIS | $MSFT, since slower revenue ramp and changes of costs, but this is coming from a trader's perspecitve. Not as an actual negative when I talk about longer contract terms since if you’re investing for 1Y timeframe, you’d want the fastest revenue ramp possible. But if you look at fluidstack $WULF contract agreements eg. 15 years it provides a lot more guarantees for a company over time, vs. if hyperscalers like $MSFT end up building out their own capacity and custom ASICs then some Neoclouds might be in trouble after 4Y
-
分析AI基建公司融资方式、股权稀释风险及现金流依赖。
你说得对，低市值（MC）公司的风险比人们预期的要大，我会在下一篇帖子中指出其中的权衡取舍。 即使是市值较大的公司也会稀释股权，但稀释结构极其重要，就像 $CIFR 和 $NBIS 发行可转换债券（convertibles）那样。 有些公司主要通过资本支出（capex）进行融资，比如 $CORZ，但大多数公司（如 $NBIS）的目标是通过 $MSFT 的交易产生的自由现金流（fcf）来为资本支出融资。只是在初始建设阶段，像 $SLNH 这样的公司需要弄清楚如何为项目管道或 GPU 收购获取资金。
英文原文
You’re right there’s a lot more risk involved than people expect for lower MC ones and I’ll point off the tradeoffs in the next post. Even bigger ones dilute but dilution structure is extremely important like how $CIFR, $NBIS did convertibles. Some fund through primarily through capex like $CORZ, but the goal is for most like $NBIS to fund capex through fcf with the $MSFT deal. Just in initial buildout like $SLNH they need to figure out how to get funding for pipelines or GPU acquisitions.
-
AI矿工股将分化，新云厂商言论显著影响市场定价。
这几乎适用于所有矿工，而不仅仅是 $IREN 和 $CIFR 这两只股票。话虽如此，我们很可能会看到资金向表现优异的个股（如 $WULF 或 $NBIS）集中，而不是所有股票同涨同跌。这就像当 $NVDA CEO 发表关于量子计算的评论后，所有量子概念股下跌了 45%。来自领先的新云厂商（Neoclouds）关于利润率与裸金属（bare metal）等话题的评论，确实会对市场如何定价产生实质性影响。
英文原文
This is almost every single miner across the board, not just two stocks $IREN and $CIFR. That being said it’s likely we’ll see people consolidate into individual performers that do well (eg. $WULF or $NBIS) rather than everything go up and down together. Also it’s like when $NVDA CEO made a comment about quantum and all quantum stocks went down 45%. Commentary about stuff like margins vs. bare metal do have material impacts on how markets price things in when it comes from the leading Neoclouds
-
指出AI云与挖矿股出现早期分化，市场开始定价利润率差异。
当你提到6个月、12个月、3年时，你误解了帖子的细微差别。 帖子的要点是，我们现在看到通常同步交易的股票以及$NBIS等低贝塔股票开始出现早期分化。 本周：$NBIS: +7.25% | $CIFR: -7.15% $RIOT: -11.5% $BITF: -19.92% $CLSK: -11.63% $IREN: -5.54% $BTBT: -11.51%。 这仅仅是对分化的观察。我想指出，市场可能开始计入利润率差异。 当新云(Neocloud)领导者发布关于全栈利润率(full-stack margins)的此类帖子时，市场会有反应。
英文原文
You're misunderstanding the nuance of the post when you say 6 months, 12 months, 3 years. The point of the post was that we're now seeing early divergence of the names that usually trade in sync + $NBIS and others that trade lower beta. This week: $NBIS: +7.25% | $CIFR: -7.15% $RIOT: -11.5% $BITF: -19.92% $CLSK: -11.63% $IREN: -5.54% $BTBT: -11.51%. This is just an observation from about the divergence. And I wanted to point out that it's likely margin differences that the market's starting to price in. When the Neocloud leaders make posts about full-stack margins like these, markets react.
-
指出AI相关股票与矿企间因利润率差异产生的早期表现分化。
我并非声称一周就能定义趋势，而是指出我们正看到早期的分化迹象：这些通常同步交易的标的，以及 $NBIS 等低贝塔(low beta)标的表现出现背离。如果你读过我提到的帖子，其中涉及 $WYFI、$APLD、$WULF、$CIFR、$BITF、$CLSK 以及广义上的“矿企(Miners)”，我并非指向某一只特定股票。这仅仅是对表现分化的一种观察。这大概是因为市场开始定价利润率差异所致。你可以自由表示不同意，这很公平。
英文原文
I’m not claiming one week defines a trend, I’m pointing out that we're seeing early divergence where these names usually trade in sync + $NBIS and others trade lower beta. If you read the post I mentioned $WYFI, $APLD, $WULF, $CIFR, $BITF, $CLSK and "Miners" in general I'm not pointing to one specific name. This is just an observation from about the divergence in performance. And it's probably due to margin differences that the market's starting to price in. You're free to disagree and that's fair.
-
HPC矿企表现分化，资金从传统矿企转向新云，IREN相对抗跌。
这不仅仅是关于 $IREN，而是关乎高性能计算(HPC)矿企。如果非要比较，$IREN 的表现可能更为出色，尤其是与 $CLSK、$BITF 等其他公司相比。 例如： $NBIS: +7.25% $CIFR: -7.15% $RIOT: -11.5% $BITF: -19.92% $CLSK: -11.63% $IREN: -5.54% $BTBT: -11.51% 再次强调，这可能与比特币价格的多重因素有关。但最近很明显，它们之间出现了分化，而通常情况下它们会同向波动（如 $NBIS、$CRWV 具有较低的贝塔值）。 我会将叙事重心的变化——从利润率转向产能——作为矿企抛售以及资金整合进某些新云(Neoclouds)的原因之一。
英文原文
This isn't just about $IREN, it's around HPC miners. If anything IREN is probably one of the more exceptional performers, especially vs. $CLSK, $BITF, and others. EG. $NBIS: +7.25% _ $CIFR: -7.15% $RIOT: -11.5% $BITF: -19.92% $CLSK: -11.63% $IREN: -5.54% $BTBT: -11.51% Again, it's probably multi-faceted around Bitcoin prices. But it's pretty apparent as of recently, there's been divergence when normally they would move together (with $NBIS, $CRWV) lower beta. And I'd cite narrative changes around margins vs. capacity being a part of miner sell-off and consolidation into certain Neoclouds.
-
市场定价重心从产能转向利润率，导致全栈云与矿企走势分化。
所以我认为你们误解了 @SathyeshBhat @midlevelcruiser 观点的细微差别。在资本市场开始关注产能建设(capacity buildout)之前，我们曾看到全线普涨。直到**最近**（上周左右），我们才开始看到 $CRWV、$NBIS 以及来自矿企的“全栈”(full-stack) Neoclouds 出现分化。它们通常会同涨同跌，且波动率(beta)较低的 Nebius 相比 $IREN 等标的波动更小。这只是评论：目前市场可能开始更多地将规模化后的利润率(margins at scale)而非单纯的产能纳入定价，我们开始看到不同板块的标的走势出现分化。随着公司高管今天发文强调全栈的重要性，这一趋势也愈发明显。但总体而言，你说得对，短期时间框架并不重要，我也并未将其作为事实依据。
英文原文
So I think you're missing the nuance of the point @SathyeshBhat @midlevelcruiser. Before markets cared about capacity buildout and we saw the major runup on everything across the board. Only **just recently** last week or so, we're starting to see a divergence with $CRWV, $NBIS and "full-stack" Neoclouds from miners. They would normally move up/down together, and Nebius which is lower beta than the others would move less than $IREN for example. This is just commentary that as of now, markets are possibly starting to price in margins at scale more than capacity than say 1 month ago and we're starting to see more things move in different baskets. And it's becoming more apparent too with executives from the companies pointing out why full-stack matters with posts today. But generally you're right short timeframes don't matter and I'm not using it as a fact.
-
Meta AI支出利好下游供应商NBIS和CRWV
@BaronAfanas 哈哈，$VIRT 表现依然落后，但很高兴看到它略有回升。 对于 $NBIS，很难不持看多态度。 人们抱怨 $META 在人工智能资本支出(AI Capex)上花太多钱，哈哈。 如果有什么影响的话，这些资金最终只是流向了像 $CRWV 和 $NBIS 这样的公司。
英文原文
@BaronAfanas lol $VIRT is still underperforming but glad it’s recovering a bit. For $NBIS hard not to be bullish. People are complaining about $META spending too much on AI capex lol. If anything that money just funnels down to companies like $CRWV and $NBIS.
-
反驳IREN论点，指出NBIS全栈软件优势带来长期更高利润率。
观点很棒！但是…… 你的论点基于对公认会计准则(GAAP)与非公认会计准则(non-GAAP)的误解……这可能会让非会计专业人士和 X 上的其他人感到困惑。 在对比硬件部门利润与全栈 AWS 利润率时，$IREN 报告的是非公认会计准则(non-GAAP)，而 $NBIS 报告的是公认会计准则(GAAP)。 92% 是一个非公认会计准则(non-GAAP)的硬件特定利润指标，它系统地排除了折旧、数据中心 overhead、网络费用、支持等。硬件利润率通常仅通过从收入中扣除直接电力成本来计算，忽略了真实公认会计准则(GAAP) 销售成本(COGS) 的组成部分。在报告时，销售成本(COGS) 的遗漏和会计原则很重要（例如，为什么加密公司如 $MSTR 的数字与像 ASU 2023-08 这样的加密报告如此不同，因此你不能将收入与加密持仓与 Robinhood 进行比较）。 与此同时，$NBIS 71.2% 的毛利率是公认会计准则(GAAP)，并将托管费用(colocation fees) 作为 29% 收入成本的一部分。因此，当你查看 EBIT 利润率以获得全貌时，$NBIS (~30%) 要高得多，因为全栈驱动了盈利能力。当 $IREN 扩展到 5 亿美元 ARR 目标时，他们被迫纳入现实的折旧、数据中心成本、网络费用和运营。如果以一致的会计标准对两者进行标准化，$NBIS 的全栈模型在真实利润率 spread 上已经高出 15-25 个百分点。 你说超大规模客户自带 Type-1 堆栈是对的。但是……我再次对其他 $IREN 帖子说过这一点，它们一直在回避这个点。 再说一次，Nebius 的 Type-2 层不是面向客户的功能，而是优化内部运营支出(OPEX)。如果企业自带 Type 1 (K8s/Ray)，Nebius 的内部编排(Type 2) 仍然能提升其自身的利润率。这是运营支出(OPEX) 护城河，它不是客户使用的，而是关于你运行集群的效率。把它想象成 AWS 的 Nitro 虚拟机监控程序，但核心在于 AWS 如何驱动 >80% 的利用率和最小的人工运营。Nebius 的内部控制平面自动化了混合 SKU 的 GPU 调度、遥测和固件生命周期。每增加一个百分点的利用率，毛利率就扩张约 1-1.5 个百分点。Nebius 的内部编排降低了其运营成本基础并提高了利用率，无论客户是否自带堆栈，所以这不是非此即彼的关系。 长期的利用率效率和自动化复利增长快于廉价电力/容量，$NBIS 的全栈在长期驱动更高的利润率，而其他方面临利润率压缩的恶性竞争。随着 $NBIS 扩展，其软件熟练度的增量成本趋近于零，结构性地推动利润率更高。相反，$IREN 的依赖模式意味着随着扩展，许可费（如果你采用你的数字，为收入的 3-8%）加上必要的、未量化的站点可靠性工程师(SRE) 运营成本将成为盈利能力的永久拖累，阻止公司实现 $NBIS 的结构性利润率上限。外包 Type 2 解决方案的许可成本很低，如你的 3-8% 数字，但当被视为 overhead 时，“20-30%” 的数字非常相关。 回到 GAAP-non-GAAP，一旦 $IREN 在规模上被迫确认完整的公认会计准则(GAAP) 成本，与 $NBIS 相比不可避免的 15-30 个百分点的利润率差异将会显现。我同意你，短期的容量和基础设施是今天的一大优势，这就是为什么我仍然看好转型的加密货币矿工如 $IREN、$WULF、$CIFR，但在执行期间的软件护城河，在将收入转化为利润的长期来看，将大大优于裸金属基础设施（这还假设 $NBIS 无法扩展基础设施，而他们是可以的）。 $IREN 当前的优势是物理集成和廉价容量，它是资本支出(capex) 密集型的，并受制于电力市场收紧。$NBIS 的优势是运营支出(OPEX) 杠杆。$NBIS 更高的利润率是几年后基于全栈 AWS 式利润率的必然经济结果，而当其他人竞相触底时，利润率差异不会崩溃，而是随着规模扩大而扩大。 廉价电力赢得下一个季度，高效的编排成为下一个 AWS。
英文原文
Great points! But... You did base your thesis off misunderstanding gaap vs. non-gaap... which could be confusing to non-accountants and other people on X. $IREN reported non-GAAP and $NBIS GAAP when you look at hardware-segment profit to full-stack AWS margins. The 92% percent was non-GAAP, hardware-specific profit metric that systematically excludes depreciation, data-center overhead, network expenses, support, etc. Hardware Profit Margin is typically calculated solely by deducting direct electricity costs from revenue, ignoring the components of true COGS GAAP. And when you do reporting, COGS omissions and accounting principles matter (eg. why crypto firms have like $MSTR wildly different numbers from crypto reporting like ASU 2023-08, so you don’t compare revenue with crypto holdings to Robinhood). Meanwhile $NBIS 71.2% gross margins was GAAP and includes colocation fees as part of 29% cost of revenue. So when you look into EBIT margins for the full picture, $NBIS (~30%) is vastly higher because full stack drives profitability. When $IREN scales to the $500M ARR target, they’re forced to incorporate realistic deprecation, data center costs, network expenses, and operations. If you normalize both under consistent accounting, $NBIS’s full-stack model already operates 15–25 points higher in true margin spread. You’re right that hyperscale clients bring their own Type-1 stack. BUT... I’ve said this again to other $IREN posts that keep side-stepping this point. Once again, Nebius’s Type-2 layer isn't for customer-facing features, it’s optimizing internal OPEX. If enterprises bring their own Type 1 (K8s/Ray), Nebius's internal orchestration (Type 2) still drives its own margins up. That's OPEX moat, it's not what customers use, it's about how efficiently you run the fleet. Think of it like AWS’s Nitro hypervisor but central to how AWS drives >80% utilization and minimal human ops. Nebius’s in-house control plane automates GPU scheduling, telemetry, and firmware lifecycles across mixed SKUs. Every percentage point of utilization is ~1-1.5 points of gross margin expansion. Nebius's in-house orchestration lowers their operating cost base and increases utilization, regardless of whether customers bring their own stack, so it’s not quite one or the other. Long-term utilization efficiency and automation compound faster than cheap power/capacity and $NBIS full stack drives higher margins long term, while others face margin compression on a race to the ground. As $NBIS scales, the incremental cost of their software proficiency approaches zero, structurally driving margins higher. Conversely, $IREN's dependency model means that as they scale, the licensing fees (3–8% of revenue if you we go with your figure) plus the necessary, unquantified SRE operational costs will remain a permanent drag on profitability, preventing the firm from achieving $NBIS's structural margin ceiling longer on. Licensing costs for outsourced Type 2 solutions are low like your 3-8% figure, but the "20–30%" figure is highly relevant when viewed as overhead. .Going back to GAAP-non-gaap, once $IREN is forced to recognize full GAAP costs at scale, the inevitable 15–30 point margin delta against $NBIS will materialize. I’d agree with you that short-term capacity and infrastructure is a great advantage today, which is why I’m still bullish crypto miners pivoting like $IREN, $WULF, $CIFR, but the software moat during execution, would vastly outperform bare mental infrastructure in the longer term when it comes to converting revenue to profit (and this is pretending $NBIS can’t scale up infra, which they can). $IREN current advantage is physical integration and cheap capacity, it’s capex-heavy, and subject to power-market tightening. $NBIS advantage is OPEX leverage. $NBIS higher margins is an economic inevitability in a years time when rooted in full stack AWS-like margins, while margin delta isn't collapsing as others race to the bottom, it's widening as scale hits. Cheap power wins next quarter, efficient orchestration becomes the next AWS.
-
澄清加密货币挖矿利润率与HPC转型的区别。
@YungAds_ 我认为你混淆了加密货币挖矿利润率与高性能计算(HPC)转型。
英文原文
@YungAds_ I think you’re confusing crypto mining margins with HPC pivot
-
对比IREN与NBIS模式，指出矿工自建HPC堆栈难敌全栈合作优势。
$IREN 在加密货币挖矿方面拥有良好的利润率和高吉瓦(GW)容量，但这并不能直接转化为高性能计算(HPC)优势。 与此同时，他们正采取 $CIFR 和 $WULF 的路径，通过中间件编排将业务导向超大规模云服务商，但这会在长期内大幅压缩利润率。 $NBIS 直接与 $MSFT 进行全栈合作，实际上可以随着时间的推移提高其毛利率。 大家都指向 $CRWV 作为小型转型的案例，但最近我们学到的残酷现实是，如果许多矿工试图建立自己的编排和堆栈，他们很可能会失败。 $ORCL 是软件护城河的完美例子：如果一个市值 8000 亿美元的超大规模云服务商都感到困难，那么一个市值 40 亿美元的比特币矿工，或者像 $SLNH 这样市值 1 亿美元的微盘股，又该如何转型呢？ 如果他们选择直接转型，许多人在估算比特币矿工基本面时参考 $CRWV 将吉瓦(GW)转化为收入/利润的方式，但实际回报在执行过程中可能会令人失望。 Nebius 从一开始就是第一名，甚至在 GPU 利用率上超越了 $CRWV（根据其最新的白皮书），其全栈带来的长期盈利能力将在财报中胜过投机。
英文原文
$IREN has good margins for crypto mining and large GW capacity but that doesn’t translate directly into HPC. They’re going the $CIFR $WULF route in the meantime with middleware orchestration funneling to hyper scalers but that cuts into margins a lot in the long run. $NBIS is direct to $MSFT full stack and can actually increase their gross margins over time. Everyone points to $CRWV as the example to minor pivots but the harsh reality as we learned recently is that if many miners try building their own orchestration and stack, they will likely fail. $ORCL is the perfect example of the software moat where if a $800B hyperscaler has trouble how would a $4B bitcoin miner pivot, or a $100m microcap like $SLNH do it. If they go direct, a lot of people estimating bitcoin miner fundamentals use $CRWV in terms of translating GW to revenue/profit, but the actual returns will likely be underwhelming during the execution. Nebius is #1 from the very start, even beating out $CRWV (from their most recent whitepaper with gpu utilization) and long term profitability from full stack will shine when it comes to earnings vs speculation.
-
对比NBIS与转型矿企，强调软件全栈与护城河决定高毛利优势。
我在其他评论中提到 $CRWV，将其作为加密货币矿工转向高性能计算(HPC)所追求的金标准范例。 即便如此，在 $CRWV 花费数十亿美元收购软件并耗时多年构建之后，$NBIS 的 GPU 利用率依然更高。 人们对 $IREN 和其他转型中的加密货币矿工抱有过于乐观的看法，而 $ORCL 的毛利率实际上仅为 14%。$CRWV 花费数十亿美元和数年时间构建软件，人们却期望像 $SLNH 这样市值仅 1.4 亿美元的小盘股能有效转型并实现良好的毛利率。 产能带来更高的潜在收入，但如果毛利率微薄则意义不大。全栈+软件编排+客户多元化才是护城河，也是 $NBIS 在毛利率方面可能胜出的原因。 （另外我不看好 $CRWV，因为其坏账侵蚀了利润。$NBIS 的 40%+ 价外可转换票据结构才是正确路径）
英文原文
I mention $CRWV in the other comments as the gold standard example crypto miners pivoting to HPC aim for. Even then $NBIS has higher GPU utilization after CRWV spends billions on software acquisition and years to build it out. People have too much rose-tinted lens on $IREN and other crypto miners pivoting when $ORCL literally had 14% gross margins. Again CRWV spend billions and years on their software buildout and people expect some $140m micro cap like $SLNH to effectively pivot into good margins Capacity leads to higher potential revenue but it doesn’t mean much if their margins are small. Full stack + software orchestration + customer diversity is the moat and why $NBIS will likely come out on top in terms of margins. (Also didn’t like $CRWV since their bad debt eats into margins too. $NBIS convertible note structure 40%+ OTM is the way to go)
-
质疑小型矿企转型算力基建能力，对比巨头与同行效率。
$ORCL 未能匹配这些算力，而它是一家市值 8000 亿美元的公司。你是在赌那些市值 30 亿至 150 亿美元的小型加密货币矿企转型后能建成这些设施。$CRWV 花了数年时间并通过数十亿美元的收购才勉强实现运营，即便如此，根据其白皮书，$NBIS 的利用率仍更高。
英文原文
$ORCL failed to match these capacities and it's a $800B company. You're betting that a lot of these small $3-$15B crypto miners pivoting can build it out. $CRWV took years and billions of dollars in acquisitions to make it sort of work and even after that, $NBIS has higher utilization as per their whitepaper.
-
软件利用率对AI算力利润的影响远超廉价电力成本。
$NBIS 拥有廉价电力，但可能不如 $IREN 或 $CIFR 便宜。但为了让你明白为什么这没那么重要： 软件利用率/编排(Orchestration)是护城河，其重要性比廉价电力高出约 30-70 倍。（请有人核实一下这个数学计算） 一张 H100 功耗约 0.84 千瓦。在 0.05 美元/千瓦时的电价下，成本为 0.04 美元/GPU-小时；即使在 0.10 美元/千瓦时，成本也仅为 0.08 美元/GPU-小时。 鉴于租赁价格约为 4 美元/GPU-小时，电力成本仅占收入的 1-2%。但将利用率从 50% 提升到 85%，会使单 GPU 收入增加 70%，从而大幅改变利润率。 不过，如果你拥有像 $CRWV 那样的所有组件（除了他们的坏账问题），那就是另一回事了，但这巨大的投机性赌注在于所有矿工能否建成这些设施。
英文原文
$NBIS has access to cheap power, but it might not be as cheap as $IREN or $CIFR. But just to give you a sense of why it matters a lot less: Software utilization/orchestration is a moat and matters ~30–70 TIMES more than cheap power. (someone pls fact check the math) A H100 burns ~0.84 kW. At $0.05 / kWh, that’s $0.04 / GPU-hour; even at $0.10 / kWh it’s $0.08 / GPU-hour. With rental prices of ~$4 / GPU-hour, power is just 1–2 % of revenue. But going from 50% to 85% utilization changes revenue per GPU by +70%, swinging margins massively. Then again if you have all the components like $CRWV (aside from their bad debt), then that's a different story, but it's a huge speculative gamble that all the miners can build that out.