全部推文
共 6775 条 · 第 186 / 226 页 · 滚到底自动加载更多
-
作者基于技术背景看空ETH，认为其核心机制恶化，同时提及BMNR。
诉诸权威的论证是最糟糕的，因为没有推理过程就毫无价值。以太坊尤其如此，当你看到其核心方面发生负面变化时更是如此（我们从分片(L1 Scaling)转向L2，$ETH代币也从因使用而增值的Gas费代币转变为治理/抵押品代币）。如果你想了解我的背景，我发明了如今大多数钱包使用可信执行环境(TEE)进行密钥签名的方法。我曾在加州大学伯克利分校研究过权益证明(DPoS)和秘密共享。所以我知道得足够多，可以发表评论。我们看看价格会怎么走，这是我个人看空的原因，人们可以自由地用他们自己的钱买入$BMNR并持有$ETH。
英文原文
Arguments from authority are the worst ones because they hold no merit without any reasoning. With Ethereum especially this is true when you see core aspects of it changing in a negative way (we went from L1 scaling with sharding -> L2s, and $ETH token going from a gas token that appreciated from usage to governance/collat). If you want my background though, I invented how most wallets nowadays do key signatures with TEEs. And I was a researcher in dpos + secret sharing back at UC Berkeley. So I do know a tiny bit enough to comment. We'll see where the price goes, it's my individual thesis on why I'm bearish, people are free to buy $BMNR and hold $ETH with their own money.
-
以太坊L2降低费用削弱ETH通缩，作者看空ETH及BMNR。
这进一步印证了我的观点。L2（二层网络）的数据可用性（Data Availability）成本更低，虽有利于扩展性，但无法解决价值向 ETH 回流的问题。我一直认为以太坊实现了扩展，且是极佳的开发网络。但这以牺牲 $ETH 升值潜力为代价，因此我对 $BMNR 也持看跌态度。ETH 的 L2 构想最初对币价有利，因为稀缺的区块空间加上使用量上升形成了费用/销毁飞轮。但：Fusaka 升级 -> 更多 Blob -> 更便宜的数据可用性 -> 更低的 L2 费用。这对 ETH 的费用/销毁机制不利。此前 L1 使用带来的代币销毁是价格升值的首要机械驱动力（如 EIP-1559、合并 -> 通缩），但这已消失。除非情况改变，否则我会继续使用以太坊网络，但绝不会将其代币作为投资标的。
英文原文
So, that just reinforced my point more. L2s get even cheaper data availability, which is good for scaling but doesn't address value accrual to ETH. I've maintained Ethereum managed to scale. And it's a great network to develop on. But all that at the cost of $ETH appreciation, which is why I'm bearish also on $BMNR. ETH's L2 idea was originally good for ETH price because scarce blockspace + rising usage = fee/burn flywheel. But: Fusaka -> more blobs -> cheaper DA -> lower L2 fees. Not good for ETH fee/burn. Token burn from L1 usage was the #1 mechanical driver of price appreciation (eg. EIP-1559, Merge - > deflationary) before but that disappeared. Until that changes, I'll keep using Ethereum as a network but I will absolutely stay away from the token as an investment.
-
ETH网络成功但代币逻辑弱化，对ETH及BMNR持看空观点。
$ETH 最初且简单的投资逻辑是：更多网络使用 -> 更多 Gas 需求 -> 更高的 ETH 价格。 正如你所说，在 L2 + Blobs/EIP-4844 之后，价值体现方式已不同。生态系统使用量 ↑↑↑，L1 Gas + 销毁 ↓。 这也是我对 ETH 作为代币/ $BMNR 作为投资载体持负面看法的原因。以太坊作为网络获胜（它确实赢了）与 $ETH 代币价格上涨是有区别的。 与现已消失的使用量带来的机械性上涨相比，抵押品等经济挂钩极其微弱。
英文原文
The original and simple investment thesis for $ETH was: more network usage -> more gas demand -> higher ETH price . As you mentioned this is no longer the case since value shows up differently after L2s + blobs/EIP-4844. Ecosystem usage ↑↑↑, L1 gas + burn ↓. And this is where I have a negative view of ETH as a token/ $BMNR as a holding vehicle for investment. There's a difference of Ethereum winning as a network (it did), and $ETH token appreciating in price. Economic hooks like collateral, etc. are extremely weak compared to mechanical upside from usage that is now gone.
-
博主回应网友，承认此前观点有误并感谢对方的深入分析。
@Agrippa_Inv 是的，我并没有反对你，因为我已经说过我的观点有误。我只是很荣幸你在事后花时间做了深入分析 lol
英文原文
@Agrippa_Inv Yeah I'm not disagreeing with you since I went out and said it was off. I'm just flattered you spent the time to do a deep dive after the fact lol
-
博主讽刺对方未读更正而发长文回复。
@Agrippa_Inv 很高兴你花时间发了一个40段的回复，而不是去读8小时前发布的更正。https://t.co/0oJT77s1jx
英文原文
@Agrippa_Inv I'm glad you found the time to post a 40 paragraph response instead of reading the correction posted 8 hours ago. https://t.co/0oJT77s1jx
-
ETH网络增长但价值外流，L2导致销毁量低，高价下投资逻辑失效。
以太坊网络本身正在增长。但 Vitalik 扩展以太坊的方式将经济价值从资产本身推开了。 这就是让 $ETH / $BMNR 在 $4K 以上成为糟糕投资（尽管是好交易）的残酷经济现实。 L2 削弱了持有者的价值捕获，因为数据可用性(DA)和区块空间(blockspace)并非如此运作（10倍使用量 -> 10倍 DA 费用 -> 10倍销毁）。生态系统使用量创历史新高，但 $ETH 销毁量却处于历史低位(ATL)，这正是展示这种差异的关键指标。 压缩的低成本 Blob 意味着数十亿级的 L2 活动仅转化为极少量的主网 ETH 支出。 主网每天仅销毁数十枚 ETH，远低于 EIP-4844 之前的数百/数千枚。你说得对，L2 确实改变了 ETH 的价值捕获机制……通过将其移出 ETH。 你可以争论作为抵押品资产的投资选项和价值捕获的转移。但最初证明 ETH 作为投资合理性的主要代币经济学已不复存在： ETH 作为资产的价格不再与网络使用量成比例增长。
英文原文
Ethereum's network itself is growing. But the way Vitalik scaled Ethereum pushed economic value away from the asset. That's the sad economic reality that makes $ETH / $BMNR a terrible investment above $4K (it's a good trade). L2's weakened value accrual for holders because it doesn't work this way with DA and blockspace (10x usage -> 10x DA fees -> 10x burn). Ecosystem usage hits new highs, yet $ETH burn ATLs is exactly what you look at to show this disparity. Compressed, low-cost blobs mean that billions in L2 activity translate into only a tiny amount of ETH spent on mainnet Mainnet is burning tens of eth a day, down from hundreds/thousands a day before EIP-4844. You're right by saying L2s have genuinely changed ETH's value accrual mechanics... by moving it off of ETH. You can argue collateral asset as a investment option and a shift of value accrual. But the main token economics that originally justified ETH as an investment doesn't exist anymore: ETH price as an asset no longer scales proportionately with network usage.
-
ETH因L2抽离价值致主网收入崩塌，代币缺乏投资吸引力。
最后，@VitalikButerin 完美指出了我对 $ETH 看空以及 @fundstrat 的 $BMNR 的原因。 $ETH 作为一种代币是极差的投资，尤其是在 $4K 以上时。 这是 Tom Lee、Cathie Wood 的 ARK 创新基金、Vitalik 等人没有理解的一点： 以太坊网络(Ethereum network)与作为实用代币的 ETH 之间存在脱节。 以太坊网络（正在增长）的增长并不转化为代币价格的上涨。 如果你在 2021 年底附近买入 $ETH，你会亏钱。与此同时，$BTC 的价格自那时以来上涨了接近 300%。为什么？ 二层网络(Layer 2s)。 1. Vitalik 在过去几年鼓励用户转向 L2（如 Base）。结果现在这些网络处理数十亿的交易量，但所有费用都流向了像 Coinbase（排序器 sequencer）这样的公司，而不是以太坊的验证者。 2. L2 支付存储数据的费用，但这与它们从以太坊主网(Ethereum mainnet)抽离的经济价值相比微乎其微。 就是这样。L2 抽走了利润和活动，而以太坊网络几乎收不到任何数据存储费。价值积累(Value accrual)流向了像 Coinbase 这样的公司，而不是 ETH 代币持有者。 以太坊网络使用量创历史新高，但主网收入崩溃，持有代币作为投资几乎没有经济激励。 除非有协议变更能从 L2 活动中为 ETH 代币产生合理的价值积累，否则不要将 ETH 作为投资。
英文原文
Finally, @VitalikButerin perfectly pointed out why I’m bearish on ETH, and @fundstrat’s $BMNR. $ETH as a token is a terrible investment, especially above $4K. This is what Tom Lee, Cathie Wood’s ARK innovation, Vitalik and others didn’t understand: There is a disconnect between Ethereum as a network and ETH as a utility token. Ethereum network (that’s growing) does not translate to the token price appreciating. If you bought $ETH near the end of 2021, you would have lost money. Meanwhile $BTC price has appreciated close to 300% since then. Why? Layer 2s. 1. Vitalik encouraged users to move to L2s (like Base) over the past few years. As a result now these networks move billions in volume but all fees go to companies like Coinbase (sequencer) not validators on Ethereum. 2. L2s pay fees to store their data but this is just incredibly minimal compared to the economic value they siphoned off of Ethereum mainnet. That’s it. Layer 2s siphoned off profit, activity, while the Ethereum’s network receives barely anything for data storage. Value accrual went to companies like Coinbase, not ETH token holders. Ethereum’s Network usage is at an all-time high, but Mainnet revenue has collapsed and there’s little economic incentive to hold the token as an investment. Unless there’s protocol changes that generate reasonable value accrual to the ETH token from L2 activity, don’t bother with ETH as an investment.
-
批评Vitalik将用户导向L2导致主网贡献不足。
@VitalikButerin 你把所有人都转移到了 Layer 2 (L2)……现在却惊讶于 Base 等网络吸走了所有用户，却对以太坊主网贡献甚微。
英文原文
@VitalikButerin You moved everyone to L2s… and are now surprised that networks like Base siphoned off all the users while contributing little back to Ethereum mainnnet.
-
调侃为建数据中心而清理葡萄园。
@RJCcapital @sam_badawi @NighthawkTradez @amitisinvesting @mvcinvesting 看起来 Vineland 已经没多少葡萄藤了。想必是为了给新的 $NBIS 数据中心腾出空间，把丛林都清理掉了。
英文原文
@RJCcapital @sam_badawi @NighthawkTradez @amitisinvesting @mvcinvesting Doesn’t look like there’s many vines anymore in Vineland. Must have cleared out the jungle to make room for the new $NBIS datacenter
-
解释因X平台降权外链而克制发帖，建议截图分享图表。
@_stockResearch 不知道，尽力而为，截图图表之类在 X 上很难排版的内容。不幸的是，X 会降权任何包含外部链接的用户，所以我一直克制着不发。
英文原文
@_stockResearch Idk just try your best and screenshot graphs and stuff that’s hard to format on x. Unfortunately X deboosts any user that links outside so I’ve refrained from posting
-
分析NBIS、IREN和CRWV商业模式优劣及新云厂商时间窗口。
好问题。麦肯锡曾就此话题发文（我觉得写得极差，因为他们以 $CRWV 为主要锚点）。 但其中部分观点成立，并对 $IREN 等公司发出警示。他们的观点： - 当前的裸金属租赁商业模式薄弱且脆弱 - 避免过度依赖少数大客户 - 开辟可防御的利基市场（如主权计算、专用工作负载） - 通过收购整合或成为超大规模云服务商 这些确实正确，但未能捕捉到一些细微差别。 关于 $NBIS： - 极度多元化（这构成了利用率的强大护城河，对利润率计算至关重要） - 全栈式（可防御的利基市场） - 通过收购整合（旨在成为超大规模云服务商，拥有4家同步增长的子公司） 这就是我说它具有最高非对称上行潜力的原因。 关于 $IREN： - 当前的裸金属租赁业务目前是护城河。文章指出长期来看它很脆弱，这是正确的。因此 $IREN 正通过与 $MSFT 合作开展 GPU 基础设施即服务(IaaS) 向上攀登全栈阶梯，并可能尝试构建上层软件层（尽管这很难） - 我们将拭目以待，这需要极高的执行力。 关于 $CRWV - 老实说，我不知道他们如何摆脱债务陷阱 - 他们试图用 $NVDA 作为后盾，但这充其量也很脆弱（例如 OpenAI 拥有 1 万亿美元以上的资本支出，试图争取政府 + 科技七巨头提供资金担保） 新云厂商是一场与时间的赛跑，我同意文章的观点（这就是我说高确信度持有2年，而非5年以上的原因）。 他们拥有从科技七巨头(Mag7)弱势中获取收入的绝佳窗口期 -> 将收入转化 -> 建立长期差异化和护城河。 我不知道最终结果如何，但我们将拭目以待。
英文原文
Hi great question. So there was an article by Mckinsey on this topic (which I think is terribly written since they use $CRWV as the main anchor). But some points holds true, and gives warnings to $IREN and others. Their claims: - current bare-metal rental business model is weak and fragile - avoids overreliance on a few giant customers - carve defensible niches (sovereign compute, specialized workloads) - consolidate through acquisitions or be a hyperscaler Are definitely correct, but fail to capture some nuances. So for $NBIS: - Extremely diversified (so this is more as a powerful moat for utilization, which is huge for margin calculations) - Full-stack (defensible niche) - consolidate through acquisitions (it's aiming to become a hyperscaler, has 4 subsidiary companies growing alongside it) That's kind of why I've said it has the highest asymmetrical upside of the bunch. For $IREN: - current bare-metal rental business is a moat as of today. The article is correct in stating long term it's fragile. That's why $IREN is moving up the full-stack ladder doing GPU iaas with $MSFT, and will likely try and build software layers on top (though it's hard) - We will see what comes out of this, it's high execution. For $CRWV - idk how they're going to get out of the debt trap tbh - they're using $NVDA to backstop it, but it's shaky at best (eg. openai with $1t+ in capex trying to get gov + mag7 to backstop funding) Neoclouds are a race against time, I agree with the article (which is why I said 2 year high conviction hold, not 5 years + ). They have this brilliant window of opportunity of weakness from mag7 -> funnel revenue down -> build long term differentiation and moats. I don't know what will happen, but we'll see
-
AI云股被算法归篮联动，CRWV财务堪忧，NBIS/IREN长期价值或超CRWV。
是的，完全同意。我认为算法/市场目前将 $NBIS、$IREN 归入 $CRWV、$NVDA 的篮子中。因此，任何关于 Coreweave 的负面报道都会对其他公司产生负面影响。我们终于看到 $WULF、$CIFR 等被归入数据中心(colocation)篮子（相对不受 GPU 贬值论点影响），并表现优异。但坦率地说，$CRWV 是一个财务噩梦，只要它被视为新云(neocloud)行业领导者，就会影响其他公司。至于 $NVDA，$GOOGL 的 TPU 论据是我目前看到的针对 $NVDA GPU 云业务最强的看空理由，但这些公司已经从 $META、$MSFT 锁定了 5 年的超大规模云服务商(hyperscaler)合约。我认为市场最终会正确定价，我相信 $NBIS 和 $IREN 有一天会比 $CRWV 更有价值，但其中只有一家拥有自动驾驶 Robotaxi lol。
英文原文
Yep absolutely. I think algos/market put $NBIS, $IREN in the $CRWV, $NVDA basket right now. So any negative hit piece about Coreweave does negatively affects the others. We've finally seen $WULF, $CIFR and others get put into the colo basket (which are relatively unaffected to GPU depreciation arguments), and outperform. But $CRWV is a financial nightmare to put it bluntly, so it does affect the others as long as it's treated as the neocloud sector leader. As for $NVDA, TPU arguments from $GOOGL is the strongest bear case I've seen to date though on $NVDA GPU clouds, but these companies already have have 5 year hyperscaler deals locked in from $META, $MSFT. imo markets will price things in correctly in due time, I do think both $NBIS and $IREN will be worth more than $CRWV one day. but only one of those has self-driving robotaxis lol
-
感谢博主并期待其后续研究文章。
@_stockResearch 谢谢！期待看到你未来的深度分析文章。
英文原文
@_stockResearch Thanks! Looking forward to seeing your future writeups too.
-
IREN单季签约容量大增，展现强大执行力与护城河优势。
哦，看来 @babyfolio 已经替我回答了！ 从 1 GW 到 2.5 GW 的指导性签约容量（Guided Contracted Capacity），仅一个季度就实现这样的增长已经非常惊人。稍微调整一下大家的预期，GW 级容量是一道护城河（Moat），而 $IREN 在这道护城河上坚守了最长时间。 因此，能在短短一个季度内实现与 $IREN 的中长期对等，这表现堪称惊艳。 话虽如此，正如 babyfolio 所说，我认为 2.5 GW 这一数据已经大大超出了我的预期哈哈，但如果他们能在一个季度内实现 250%+ 的增长，他们可能再次做到。所以在这些情况下，你只需信任管理层的执行力（Management Execution）。
英文原文
oh looks like @babyfolio answered that for me! Going from 1 GW to 2.5 GW guided contracted capacity is already amazing from just 1 quarter. Just to tamper your expectations a bit, GW capacity is a moat and one that $IREN held for the longest time. So to achieve medium-long term parity with $IREN in just 1 quarter is spectacular. That being said, I think the 2.5 GW already kinda blew away my expections like babyfolio said haha, but if they're able to increase 250%+ in 1 quarter, they could do it again. So in these cases, you just trust in management execution.
-
NBIS产能指引大增追平IREN，长期竞争力趋同
事实上，截至第三季度财报，$NBIS 将其签约产能指引提高了2.5倍，并在长期层面几乎追平了 $IREN。然而，$IREN 在前置并网产能（现在至2026年上半年）方面拥有巨大价值。但从2026年下半年起，$NBIS 在这方面似乎已迎头赶上（截至Q3，签约电力指引从3 GW调整为2.5 GW）。
英文原文
So actually as of Q3 earnings, $NBIS 2.5x'd its contracted capacity guidance and achieved near long-term parity with $IREN. However, $IREN has a tremendous amount of value in front loaded grid-connected capacity (now -> H1 2026). But H2 2026 onwards, $NBIS looks like it caught up in that front (3 GW to 2.5 GW contracted power guided as of Q3)
-
回应反馈，重申对NBIS作为十年一遇机会的高置信度持仓逻辑。
感谢反馈，我会铭记于心，并照常发布其他主题的内容！ 我昨天发布了关于 $NVDA 等标的的深度研究(DD)，但正如你提到的，近期内容主要集中在 NeoCloud 板块。我之所以聚焦于此，是因为近期新闻密集，且巨大的抛售引发了大量关注！ 具体到 $NBIS，这是我持仓两年中信心最高的一笔投资。我认为这是继 $NVDA 之后十年来难得的机会，你很少看到一家核心公司净利润同比增长 700%，同时旗下 4 家投资组合公司净利润均实现 100% 的同比增长。 其中一家恰好是一家非常酷的自动驾驶 Robotaxi 公司（我是 Waymo 和 $TSLA Robotaxi 的粉丝）。
英文原文
Thanks for the feedback, I'll take it to heart and do posts on other stuff as usual! I posted DD about stuff like $NVDA yesterday but recently as you mentioned it's been mainly about the neocloud sector. I've been focused on it because of how much news there was recently + the huge sell off attracted a lot of attention! For $NBIS in specific, it was my highest conviction 2 year hold. I just see it as the opportunity of a decade since $NVDA, you don't really see any core company growing its bottom line 700% Y/Y with 4 separate portfolio companies growing 100% Y/Y. One of which happens to be a super cool self-driving robotaxi company (I'm a fan of Waymo, $TSLA robotaxis).
-
重申看好IREN及新云板块，特别看好其高增长副业。
@babyfolio 自从我发布最初的论点帖子以来，我一直看好 $IREN 以及新云(neocloud)板块！ 我只是恰好对那家作为副业(side quest)拥有4家投资组合公司且同比增长100%+的企业极度看好。
英文原文
@babyfolio I've always been bullish on $IREN and the neocloud sector since my original thesis post! I just happen to be mega bullish on the one that has 4 portfolio companies growing 100%+ Y/Y alongside it as a side quest.
-
博主更正$MSFT数据错误，澄清计算逻辑并建议忽略特定图表。
是的，$MSFT 的交易数据确实有误。这是最后时刻加进去的，但我本应该仔细校对一下。 不幸的是无法编辑 X 帖子，所以我就留在这里说明： - 计算时使用了混合估算值，例如芬兰设施（Mantsala）。你说得对，$MSFT 的交易应该仅按美国本土交易计算。市场费率可能在 200 万美元以上。 - 我根据 X 上的多个 GPU 场景做了拆解，但这只是其中之一。$NBIS 先提升 H200 产能，随后部署 GB300，但这并未展示全貌，所以请忽略该图表。 标准化表格仍然有效，可作为粗略的相对比较。
英文原文
Yep, the $MSFT deal in specific was off. Threw that in last minute but should have proof-read it. Unfortunately can't edit X posts but I'll just leave this here: - Used a blended estimate eg. Finland facility (Mantsala) for caluations. You're correct in saying $MSFT deal should be US-only deal calculations. Market rates would likely be around $2.0M+ - Did a breakdown of multiple GPU scenarios off X but this was just one of them. $NBIS was ramping up with h200 first then deploying gb300's down the road, but doesn't show the full picture so ignore the chart. Normalization table still stands as a rough relative comparison.
-
澄清 $MSFT 交易估算差异及图表错误，确认 H100 数据有效。
你在 $MSFT 这笔交易上的看法是对的。我在托管设施(Colocation)方面使用了混合估算值，$NBIS 在其他地点拥有设施，但对于 $MSFT 这笔特定的独立交易，托管费会有所不同。此外，我在进行 GPU 标准化(GPU normalization)的不同对比时，用错了图表。右侧的 H100 标准化数据仍然有效。
英文原文
You're right on the $MSFT deal. I used blended estimates for colocation, $NBIS owns facilities in other location but for specific $MSFT deal standalone, the colocation fee would be different. Also was doing different comparisons for GPU normalization, and used the wrong chart. H100 normalization on the right should still stand.
-
澄清NBIS成本估算逻辑及与微软合作路线图
与 $MSFT 的对比有些偏差，但 H100 的标准化处理虽然带有推测性，应该更稳健。 我使用了整个投资组合的混合估算值，例如芬兰设施（Mantsala），那里的数据中心租赁费实际上为 $0（仅包含折旧摊销+运营支出）。当他们从 DataOne 租赁美国设施（新泽西州 Vineland）时，我对其进行了平均处理。 正如其他人指出的那样，仅就 $MSFT 的交易而言，租赁成本可能要高得多，接近每兆瓦 $180万-$220万的市场价格。 $NBIS / $MSFT 的合作不仅限于 H200。GB200/B300 也在路线图之中。我当时正在对比 H200、H200 和其他 GPU，最后顺手加上了这个。可惜帖子发晚了没法编辑。 感谢大家的提问。
英文原文
The $MSFT comparison is off but the h100 normalization, while speculative, should be more robust. I used a blended estimate of entire portfolio eg. Finland facility (Mantsala), where colo rent is effectively $0 (just D&A + OpEx). When they lease the US facility (Vineland, NJ) from DataOne and I averaged it. For the $MSFT deal only, the lease cost is likely much higher closer to market rates of $1.8M-$2.2M per MW as someone else pointed out. $NBIS / $MSFT is not limited to H200. GB200/B300 is on the roadmap too. I was doing comparisons with H200, H200, and other GPUs and threw that in at the end. Too late to edit the post though. Appreciate the questions.
-
看好NBIS为行业龙头，静待市值超越CRWV。
@babyfolio 在我看来，它已经是明确的领导者！只等市场将 $NBIS 的未来增长/利润率计入定价，使其市值超越 $CRWV。
英文原文
@babyfolio It’s already the clear leader in my view! Just waiting for market to price in $NBIS forward growth/margins and the marketcap overtaking $CRWV.
-
对比NBIS与IREN的微软交易，指出NBIS收入溢价及IREN利润率被高估。
是的！关于收入溢价的观点确实很有帮助。但我认为细微差别体现在 $MSFT 的交易中（这显示了 $NBIS 和 $IREN 每兆瓦的利润率差异）。 Nebius 的微软交易使其每兆瓦年的收入比 IREN 的微软交易高出约 19-20%。许多毛利率数据因资产负债表会计处理而被夸大，因此我发此帖以标准化利润率。 $IREN 的实际杠杆内部收益率可能更接近 20%，鉴于其与戴尔的数十亿美元支出，使用这一指标可能优于 85% 的项目息税折旧摊销前利润（EBITDA）数据。 $MSFT 基于合理推测（结合靠近 Azure 服务器的地理位置和软件优势）更看重 $NBIS 的完整人工智能云平台。如果 $IREN 在顶层软件和基础设施即服务（IaaS）层面补齐短板，其未来利润率和合同有望缩小这一差距。
英文原文
Yep! Definitely some helpful points about the revenue premium. But I think the nuance did show up in the $MSFT deal (which shows the margin difference between $NBIS and $IREN per MW). Nebius’s MSFT deal gives it ~19–20% higher revenue per MW-year than IREN’s MSFT deal. A lot of the gross margin figures are inflated by balance sheet accounting, hence why I made this post to normalize margins. $IREN 's realized levered irr is probably closer to 20%, it's probably better to use that over the 85% project EBITDA figures since they're spending billions with Dell. $MSFT values $NBIS full AI cloud platform more from an educated guess (mix of location closer to azure servers and software). If $IREN closed the software on top level and iaas level, its future margins/contracts could close that gap.
-
TPU虽具竞争力，但NVDA积压订单提供中期保障，难构成看跌。
是的，话题是关于TPU v7 Ironwood作为首个对$NVDA构成竞争替代的产品，在特定推理工作负载上表现更优。目前$NVDA的积压订单是中期的安全垫。关键问题在于TPU能否持续优于Nvidia旗舰GPU（我的答案是否定的，随着下一代发布）。然而，如果以下情况发生，则构成看跌逻辑：- TPU持续优于$NVDA旗舰GPU；- 人们更倾向于选择TPU而非NVDA芯片；- AI算力需求减弱，且$NVDA的GPU未完全售罄。
英文原文
Yeah, the topic was about how the TPU v7 Ironwood is the first competitive alternative to $NVDA and performs better for specific inference workloads. The current $NVDA backlog is a security guarantee for medium term. It's just a question of whether TPUs continue to outperform Nvidia's flagship GPUs (my answer is probably not, with the next gen release). It is a bear case, however, if - TPUs continues to outperform $NVDA flagship GPUs - people want TPUs over NVDA chips. - AI compute demand lessens, and $NVDA is not fully sold out of GPUs.
-
对比GOOGL与NVDA，认为前者风险更低，后者潜在回报更高。
$GOOGL 具有非对称上行潜力。张量处理单元(TPU) 只是搜索、Waymo、YouTube 及其他业务之外的一个“支线任务”。 $NVDA 若 AI 随机器人/智能体指数级爆发，且其新模型在训练/推理性能上大幅超越其他供应商，则上行空间更大。 由于无法预知结果，同时持有两者无妨，但 Google 显然是两者中风险较低的选择。
英文原文
$GOOGL for asymmetrical upside. TPUs is just a side quest among search, Waymo, YouTube, and less other things. $NVDA for much higher upside in case AI continues taking off exponentially with robotics/agents and they end up leapfrogging all other providers in both trading/inference performance in their newer models. Doesn’t hurt to own both since it’s impossible to know, but Google is clearly the less risky of the two
-
AI算力需求指数级增长抵消GPU迭代贬值，NVDA客户优质，非泡沫崩盘。
答案很微妙。 主要看两个因素： 1. GPU 变得更节能。 2. 大语言模型(LLM) 在容量/能效上更高效。 在 LLM 方面，我们看到像 DeepSeek 这类模型在处理不需要高精度的任务（如回答烹饪食谱或知识库查询）时极其高效。 然而……随着计算力的增加，准确率（尤其是复杂研究问题）也在提升。Elon 和 Magnificent Seven 意识到了这一点，所以他们正在扫货市场上的所有 GPU 以创造超级智能。这也是为什么 Anthropic 和 Google 正在建设耗资 400 多亿美元的数据中心，用于运行需要更多算力进行批判性思维（如 Genesis 任务）的更高级 Opus 和 Gemini 模型。 在 GPU 方面，每一代新 GPU（例如 H100 -> B200）在能效和每美元性能上都有显著提升。例如，Blackwell B200 是 Hopper H100 的 30 倍。 如果基于这个假设，那么到 2027/2028 年，市场上将出现大量过时的低效 H100 和 B200，导致二手 GPU 市场崩盘。 但是：这是假设我们没有看到对新 AI 能力的指数级需求（我们很可能会看到，且正在发生）。正因为这种指数级需求，今天旧模型（如 7 年前的 TPU 和 2020 年的 GPU）仍被用于低优先级的推理任务。 $NVDA 的订单已积压数年，人们正在购买 $AMD 的 GPU 和 $GOOGL 的 TPU 来构建任何新增产能。 至于思科类比，思科的客户是互联网泡沫时期无盈利能力的公司。$NVDA 的客户是 $META、$AMZN、$GOOGL、$MSFT，这些是世界上最盈利的公司。所以最坏的情况我们可能看到回调，而不是互联网泡沫式的崩盘。
英文原文
Answer is nuanced. So two factors: 1. GPUs get more power efficient. 2. LLMs get more capacity/power efficient. For the LLMs case, we're seeing that on deepseek type models be extremely efficient on stuff that don't require much accuracy. Basic stuff like responding to questions about cooking recipes, or knowledge-base stuff. However... accuracy increases, especially with complex research questions, scaled with compute. And people like Elon + mag7 realize this, which is why they're just buying up all the GPUs on the market to create superintelligence. And why antrhopic/google is building $40B+ datacenters for more advanced opus and gemini models that require more compute for critical thinking (eg. Genesis Mission) For the GPUs case, every new generation of GPU (e.g., H100 -> B200) offers dramatic improvements in power efficiency and performance per dollar. eg. Blackwell B200 is 30x than the Hopper H100. If we go off that assumption, then there would be a massive useless supply of less-efficient H100s and B200s in 2027/2028 creating a used GPU market crash. HOWEVER: This is if we don't see an exponential demand for new AI capabilities (which we likely will, and what we're seeing now). Because of this exponential demand, TODAY, older models are still used (eg. TPUs from 7 years ago and GPUs from 2020), for lower inference task in lower priority inference tasks. $NVDA is backlogged for years and people are buying GPUs from $AMD /TPUs from $GOOGL to build out any new capacity. As for Cisco analogy, Cisco's customers were .com bubble companies with no profitability. $NVDA's customers are $META, $AMZN, $GOOGL, $MSFT the most profitable companies in the world. So worst case scenario we might see a correction, not a .com bubble crash.
-
澄清NVDA与TPU性能对比误区，重申NVDA短期订单安全及长期技术优势。
是的，$NVDA 是强力买入标的。 然而，其他关于 $NVDA GPU 与 TPU 性能对比的病毒式传播的研究帖子，错误地拿旧一代 TPU v6e 来对比，以展示当前 $GOOGL TPU 与 $NVDA GPU 的情况。 我的帖子是关于新一代 TPU v7 Ironwood 与 $NVDA Blackwell B200 的对比，并展示了架构上的对等性。 TPU v7 Ironwood 的性能飞跃非常巨大（计算能力提升高达 10 倍），以至于 X 上流传的许多关于 TPU 推理能力和成本节约的研究具有误导性。 就买入逻辑而言，$NVDA 很可能在未来几代芯片中超越 TPU v7，且其巨大的积压订单为其提供了短期（2年）的安全性。
英文原文
Yes $NVDA is a strong buy. However to the other viral research posts about $NVDA GPU vs TPU performance, they were falsely comparing older generation TPU v6e to show the current situation on $GOOGL TPU vs $NVDA GPU. My post was on newer TPU v7 Ironwood vs $NVDA Blackwell B200 and showing the architectural parity. The TPU v7 Ironwood performance leap is so substantial (up to 10x compute jump) that a lot of research going around X is misleading on the inference capabilities and cost savings of TPU. In terms of buy thesis, on $NVDA, they will likely leapfrog TPU v7 in newer gen chips in due time and their immense backlog provides them near term (2Y) security.
-
指出基于旧版TPU的分析已过时，强调TPU v7性能飞跃使旧成本分析失效。
你发布的 @ArtificialAnlys 分析对于评估当前 $GOOGL TPU 与 $NVDA GPU 的局势已经过时/错误。他们基于 TPU v6e 进行了分析。 我的帖子是关于更新型号（TPU v7 Ironwood 与 $NVDA Blackwell B200）在架构上的对等性。 TPU v7 Ironwood 的性能飞跃如此巨大（计算能力提升高达 10 倍），以至于他们对 v6e 的成本分析已不再相关。
英文原文
The @ArtificialAnlys analysis you posted is outdated/ wrong for analyzing the current $GOOGL TPU vs. $NVDA GPU situation. They did theirs on TPU v6e. My post was on architectural parity on newer models (TPU v7 Ironwood vs $NVDA Blackwell B200). The TPU v7 Ironwood performance leap is so substantial (up to 10x compute jump) that their v6e cost analysis is irrelevant.
-
NVDA短期强烈买入，下一代芯片优势明显，市场未放缓。
哈哈，我之前在 $NVDA 的公关稿下当“神评论” trolling（嘲讽/玩梗）火了。所以想帮我的哥们 Jensen 一把，写写关于 TPU vs GPU 争论的看法，以及股价下跌是否反应过度。给快速浏览的朋友划重点： - 基于订单积压（Backlog）的可见性，短期强烈买入。 - $NVDA 下一代芯片很可能在技术上超越 TPU。 - AI 市场正在扩张，我们没有看到放缓迹象。 - 唯一需要担心的是，如果人们刻意去买 TPU/AMD，即下一代超大规模云服务商（Hyperscaler）芯片，那也只有在 $NVDA 生产的任何 GPU 都卖断货的情况下才会发生。
英文原文
lol I was top comment trolling $NVDA's press release that went viral. so wanted to do my buddy Jensen a solid by writing up thoughts on TPU vs GPU arguments + whether stock drop was an overreaction. So TLDR for people skimming the post: - strong buy near term from backlog visibility - nvda next-gen chips will likely leapfrog tpu - ai expanding market, we're not seeing it slow down. - only thing to worry about is if people go out of their way to buy TPUs/AMD, next gen hyperscaler chips if and only if NVDA stops selling out of any GPUs they make.
-
英伟达虽面临谷歌TPU等定制芯片竞争，但中期统治力稳固，逢低买入。
英伟达($NVDA)公布的2026财年Q3营收为570.1亿美元(+62.5% YoY)，表现强劲。并指引Q4营收超650亿美元(超预期30亿美元+)，以及通过CY 2026年Blackwell/Rubin系列营收超5000亿美元。尽管如此，股价仍下跌12%。现在$NVDA是强力买入吗？答案如下： $GOOGL的TPU项目成为首个对$NVDA GPU构成竞争替代的方案，Anthropic承诺采购超100万颗TPU芯片，Meta据报道正在就数十亿美元的TPU采购进行高级别谈判。沃伦·巴菲特近期也向$GOOGL投资超40亿美元，鉴于伯克希尔对科技股保守的投资立场，这极为罕见。 尽管创下盈利新高，英伟达股价在过去10个交易日中有6天下跌，较10月29日触及5.03万亿美元市值时的历史高点$207.04下跌约12-15%。 分析师反应普遍看多，普遍上调目标价： - Evercore ISI从$261上调至$352 - 美银从$235上调至$275 - 花旗从$220上调至$270 - 高盛从$240上调至$250 - 摩根士丹利从$220上调至$235 但这里有个价值万亿美元的问题：超大规模客户日益增长的定制硅片威胁是否会削弱英伟达在AI领域的统治地位？ 与主要捕捉英伟达GPU缺货时的溢出需求的$AMD不同，谷歌的TPU项目代表了根本不同的竞争动态。 TPU v7 Ironwood是首款在性能上与Blackwell持平的非英伟达加速器，提供4.6 petaflops的FP8性能（对比B200的4.5 petaflops），配备192GB HBM3e内存。 Ironwood的架构差异化显著。虽然英伟达最大的集群配置为72个GPU(NVL72)，TPU Ironwood可扩展至9,216个芯片组。 客户斩获显著且不断增长： - Anthropic承诺采购超100万颗TPU芯片，价值“数百亿美元”，1GW算力即将上线。 - Meta正在就2026年从谷歌云租赁TPU容量进行高级别谈判，并计划2027年直接采购硬件用于自有数据中心。 - 苹果透露Apple Intelligence基础模型完全在TPU上训练，使用8,192颗TPUv4芯片用于服务器模型。 - Midjourney从GPU转向TPU，推理成本降低65%（从每月200万美元降至70万美元）。 定位微妙。TPU在超大规模推理方面表现出色，在生产级大规模服务中成本性能最高提升4倍（目前）。在训练方面，英伟达优势明显。 对于高度优化的推理任务，TPU架构可能比$NVDA的通用GPU更高效。 然而，我预计下一代英伟达GPU将在许多场景下在推理性能上超越TPU。（类似于LLM之间GPT 5 -> Gemini 3 -> Opus 4.5的迭代超越） 我们看到： 谷歌TPU、AWS Trainium、Meta MTIA、微软Maia和定制超大规模芯片都在扩展，但仍依赖$NVDA。但超大规模客户集体减少对英伟达的依赖，其累积效应是否会削弱英伟达的统治地位？ 答案：不会。至少未来两年统治地位稳固。之后如何纯属猜测。 仅看数据：Q3结果证实公司仍是AI不可或缺的基础设施提供商，5000亿美元的订单积压为2026年提供了极高的可见性。 但市场似乎正在定价3年+后的这种微妙现实：长期面临超大规模客户定制芯片的竞争不确定性。 英伟达产能完全售罄，且很可能在下一代芯片中超越TPU性能（推理性能更高且保持通用性）。但定制硅片威胁和44倍市盈率的估值担忧仍是逆风。 无论如何，鉴于出色的超预期财报和未来两年的订单积压，$NVDA目前因恐惧而下跌，似乎是中期强力逢低买入的机会。 你只需要记住这一点： 只要英伟达仍是AI工作负载的行业首选，且TPU和AMD GPU仅在需求超过英伟达供应时填补空白，它就是强力买入标的。 英伟达订单积压已满，AI需求并未放缓。
英文原文
Nvidia ( $NVDA ) reported a blowout Q3 FY2026 revenue of $57.01 billion (+62.5% YoY). And guided $65B+ Q4 ($3B+ beat), and $500B+ USD in Blackwell/Rubin rev through CY 2026. Despite that, the stock dropped 12%. Is $NVDA a strong buy now? Here's the answer: $GOOGL TPU's program emerged as the first competitive alternative to $NVDA GPUs, with Anthropic committing to over 1 million TPU chips and Meta reportedly in advanced negotiations for billions in TPU purchases. Warren Buffet also recently invested $4B+ into $GOOGL, which is extraordinarily rare given Berkshire's conservative stance to tech investments. Despite a record earnings beat, Nvidia's stock has declined in six of the last ten trading sessions and sits roughly 12-15% below its October 29 all-time high of $207.04, when it briefly touched a $5.03 trillion market cap. Analyst reaction was overwhelmingly bullish, with price targets raised across the board: - Evercore ISI raised to $352 from $261 - Bank of America raised to $275 from $235 - Citigroup raised to $270 from $220 - Goldman Sachs raised to $250 from $240 - Morgan Stanley raised to $235 from $220 But here's the trillion dollar question: will the emerging custom silicon threat from hyperscalers reduce NVIDIA dominance in AI? Unlike $AMD, which primarily captures overflow demand when NVIDIA GPUs are unavailable, Google's TPU program represents a fundamentally different competitive dynamic. TPU v7 Ironwood is the first non-NVIDIA accelerator that achieves performance parity with Blackwell, delivering 4.6 petaflops of FP8 performance (versus B200's 4.5 petaflops) with 192GB HBM3e memory. Ironwood's architectural differentiation is substantial. While NVIDIA's largest cluster configuration is 72 GPUs (NVL72), TPU Ironwood scleaes to 9,216 chip pods. The customer wins are significant and growing: - Anthropic committed to over 1 million TPU chips worth "tens of billions of dollars," with 1 gw of compute capacity coming online. - Meta is in advanced negotiations to rent TPU capacity from Google Cloud in 2026, with direct hardware purchases for its own data centers planned for 2027. - Apple revealed that Apple Intelligence foundation models were trained entirely on TPUs using 8,192 TPUv4 chips for server models - Midjourney switched from GPUs to TPUs and reduced inference costs by 65% (from $2M to $700K monthly) The positioning is nuanced. TPUs excel at hyperscale inference with up to 4x better cost-performance for production serving at scale (for now). For training, NVIDIA is the clear advantage. For highly optimized inference tasks, TPU architecture might remain more efficient than $NVDA's general-purpose GPU. However, I'm expecting next-gen Nvidia GPUs to leapfrog TPUs for inference in many scenarios. (similar to how LLMs leapfrog each other GPT 5 -> Gemini 3 -> Opus 4.5) We're seeing: Google TPU, AWS Trainium, Meta MTIA, Microsoft Maia, and custom hyperscaler chips scale up to reliance on $NVDA. But will the cumulative effect on hyperscalers collectively reducing Nvidia's dominance? The answer: No. Not yet. Dominance is secured at least for the next two years. What happens after is only speculation. Just looking at the numbers: Q3 results confirm the company remains the essential infrastructure provider for AI, with a $500 billion order backlog providing exceptional visibility through 2026. But the market appears to be pricing in this nuanced reality 3 years+ from now: long-term competitive uncertainty with custom hyperscaler chips. Nvidia is completely sold out of capacity, and are likely to leapfrog TPU performance in their next generation chips (higher performance for inference while being general purpose). But the custom silicon threat and valuation concerns at 44x earnings remains a headwind. Regardless, $NVDA seems to be a strong mid term dip-buy now on fears, given the exceptional blowout earnings and backlog for the next 2 years. This is the only thing you need to remember: NVIDIA is a strong buy as long as it remains the industry’s first choice for AI workloads, with TPUs and AMD GPUs filling gaps when demand exceeds NVIDIA’s supply. Nvidia is maxed out on backlog, and AI demand is not slowing down.
-
推测可能是ATM增发，具体抛售时间需待下季披露。
@yianisz 也可能是2500万股的自动股权融资(ATM)机制在运行。我们直到下个季度才知道这些股份何时在公开市场售出。
英文原文
@yianisz Could also be the 25 million share ATM running. We don’t know when shares are sold on the open market until next quarter