全部推文
共 6773 条 · 第 157 / 226 页 · 滚到底自动加载更多
-
LEU作为美国唯一HALEU供应商，受核能政策利好但股价反应平淡。
♻️ $LEU 是一家市值 58.3 亿美元的公司。 Centrus 是原始瓶颈清单上的另一家公司。 他们是目前美国唯一一家能够将铀浓缩为用于小型模块化反应堆(SMR)核电站的高丰度低浓铀(HALEU)的公司。 能源部(DOE)表示，一个园区将从浓缩到回收吸引“500 亿美元的投资”。 消息公布后股价仅上涨 2%，这令人惊讶。 “特朗普政府已将核能作为其国家能源政策的基石”
英文原文
♻️ $LEU is a $5.83B MC company. And Centrus was the another company on the original bottleneck list. They’re the only company in the US that can currently enrich uranium into HALEU (High-Assay Low-Enriched Uranium) for SMR nuclear power plants. DOE said one campus would attract “50 Billion in investments” from enrichment to recycling. It’s only up 2% on the news, which was surprising. “The Trump administration has made nuclear power a cornerstone of its national energy policy”
-
调侃$DJT因TAE成核聚变概念股，疑自谋联邦拨款。
@TheWingman4Life 不错！令人啼笑皆非的是，由于 TAE，$DJT 公司现在成了核聚变概念(Proxy)股。 如果他们给自己申请一大笔联邦拨款(Federal Grant)，我也不会感到惊讶。
英文原文
@TheWingman4Life Nice! Hilariously enough $DJT company is now a nuclear proxy because of TAE. I wouldn’t be surprised if they’re giving themselves a large fgrant too.
-
揭秘拥有AlSiC供应链优势、受益于AI散热与国防的低风险公司。
我惊讶于评论区里大多数 X 侦探都猜对了。 我很快会发布关于这家神秘公司的深度文章。 根据我的研究，一家美国公司拥有铝硅碳化物(AlSiC)供应链的很大一部分份额是独特的，因为这种材料可用于 $NVDA Rubin 代产品的散热。 更妙的是，AlSiC 在太空和国防垂直领域的应用使得该公司下行风险较低，因为国防部/雷神/诺斯罗普/NASA 都依赖这种复合材料。 但谁知道，这种用于高超音速导弹、F-35 战斗机和火星太空探索的散热材料，也能用于解决 2000W+ AI 部署中的翘曲问题？ 市场似乎尚未意识到这一点。
英文原文
I'm surprised majority of the X detectives in the comments guessed this right. I'll do a write-up on this mystery company soon. From my research, it was unique to have US company own a large % of the AlSiC supply chain, as this material could be used for $NVDA Rubin gen for thermal. Better yet, Space + Defense vertical usage of AlSiC gives this company low downside risk bc of DoD/Raytheon/Northrop/NASA reliance of the material composite. But, who knew the same thermal material used for hypersonic missiles, f-35 fighter jets, and Mars space exploration could also be used to handle warpage for 2000W+ AI deployments? Doesn’t seem like markets have yet.
-
IBKR因国际半导体投资受益，HOOD应优先拓展台日市场。
@hippygecko 有趣的是，$IBKR 是国际散户投资存储/半导体超级周期的二阶受益者。$HOOD 无法接入这类台湾或日本股票。我确信 @vladtenev 应该将此作为他的第一优先级。https://t.co/ne4ZdKaZYy
英文原文
@hippygecko It’s interesting that $IBKR is second order beneficiary of US retail investing internationally in the memory/semi supercycles. $HOOD doesn’t have access to these types of TW or JP stocks. Pretty sure @vladtenev should make this his #1 priority. https://t.co/ne4ZdKaZYy
-
作者分享CRWV交易经历，指出其巨额债务偿付压力。
@Veganhippo21 我去年在 $34 买入 $CRWV，今年在 $70 做了波段交易。话虽如此，“你酸吗”这种论调无助于解决 Coreweave 每年 $12.5 亿+ 债务偿付的问题。 https://t.co/DrP4LIKpYm
英文原文
@Veganhippo21 I was buying $CRWV at $34 last year, and I swing traded it at $70 this year. That being said, "u mad" doesn't help Coreweave's issue with $1.25B+ yearly debt payments. https://t.co/DrP4LIKpYm
-
对比沃尔玛与美光基本面，建议关注价值而非择时。
我喜欢举这个例子： $WMT，这家袜子零售商的前向市盈率约为40倍，其营收同比增长约5.8%，与通胀水平一致。 $MU，这家领先的存储超级周期公司，前向市盈率约为8.4-8.7倍，营收同比增长133%，每股收益(EPS)同比增长319%。 我认为人们不应该在这里试图择时交易，只需关注基本面即可。
英文原文
I like to give this example: $WMT, the sock reseller has a forward P/E is ~40 and it's growing ~5.8% Y/Y in line with inflation. $MU, the leading memory supercycle company, has a forward P/E is 8.4-8.7 ish, off 133% revenue growth Y/Y, and EPS growth 319% Y/Y. I don't think people should be timing the market here, just look at fundamentals.
-
对比美光与Nayna的风险收益，因市值差异调整持仓权重。
@ShortsHoward 风险特征和上行空间。鉴于地缘政治因素，$MU 的波动性低于 $NAYA，因此我对两者的投资组合权重配置不同。市值越小（例如 $NAYA 为 290 亿美元 vs $MU 为 4610 亿美元），通常上行空间越大。
英文原文
@ShortsHoward Both risk profile and upside. $MU is less volatile than Nanyna just given geopolitics so I do portfolio weightings differently. The smaller the MC (eg. $29B Nayna vs $461B Micron), the higher upside there is typically .
-
告知用户需在盈透证券开启境外交易权限。
@WaterFish2000 它在 $IBKR 上。你只需要开启境外交易权限即可。https://t.co/FY3s1GYrpn
英文原文
@WaterFish2000 It's on $IBKR. You just have to enable foreign trading permissions. https://t.co/FY3s1GYrpn
-
内存涨价致MU等前向市盈率降低，厂商竞相提价。
取决于具体日期。根据近期涨价的估算预测，如果计入25日的NAND（闪存）价格涨幅，$MU 的前向市盈率（Forward P/E）可能在8.4-8.7左右。南亚科（Nanya）可能在9.5左右，但其增长速度更快。再次强调，这些公司不断上调内存价格，因此即使股价上涨，前向市盈率也会降低。之前的预测也难以跟上变化。目前它们只是在比拼谁涨价幅度更大。
英文原文
Depends on the day. Using est. projections from recent price hikes, $MU is probably sitting around 8.4-8.7-ish forward p/e if you account for NAND price hikes on the 25th. Nanya is maybe 9.5-ish but it's growing a lot faster. Again these companies keep raising memory prices, so forward p/e gets lower even as the stock prices increases. And previous projections can't keep up as well. They're just one-upping each other on who can hike prices the most at this point.
-
南亚科因巨头退出成熟存储市场成唯一供应商，享稀缺溢价，股价低估。
南亚科(TPE: 2408)是我在存储领域的多头持仓之一，与$MU、SK海力士和三星并列。 但理由不同。 你可能在想，这种抛物线趋势（1年+972%）是否会停止？可能不会。 而且它每天还在上涨7%。 以下是简单的总结： 其远期市盈率仅为~13.62（随着涨价，现在可能更低）。 营收同比增长+357.7%/净利润同比增长+804.27%，其利润天文数字般巨大。 类似于美光上个月57.56%+的增速，且自从我首次提及以来已翻倍有余： 股价简直跟不上。 当巨头（三星、SK海力士和$MU）争夺高端AI市场时，南亚科专注于PC、计算机电子和工业领域的成熟及特种DRAM产品。（尽管他们也在做1b节点的先进芯片）。 存储公司留下了巨大的空白，而南亚科填补了它： 美光 - 完全退出消费级产品线 SK海力士 - 几乎完全专注于HBM4和服务器级DRAM。 三星 - 将30-40%的成熟产线重新分配给先进AI存储。 南亚科 - 最后的坚守者 需求保持稳定，但成熟DRAM供应突然断崖式下跌。且它仍被广泛使用于： - 消费级PC和笔记本电脑：2015至2023年间生产的数百万台电脑使用DDR4 - 智能手机：大多数中端和入门级智能手机使用LPDDR4 - 家庭电子产品：智能电视、游戏机（如旧款Nintendo Switch或PS4）使用这些。 基本上，它们仍需生产这些，但南亚科手握所有牌。 买家为旧内存支付“稀缺溢价”，因为他们无处可买，且在一个无聊的市场中看到了供应冲击。 南亚科本质上成了沙漠中卖水的人。他们可以不断抬高水的价格。 我做多这家在沙漠中卖水的公司。
英文原文
Nanyna (TPE: 2408) is one of my longs alongside $MU, SK Hynix, and Samsung for memory. But for a different reason. You might be thinking if this parabolic trend (+972% 1Y). is going to stop? Probably not. And it just keeps going up 7% a day. Here's a simple TLDR why: The forward P/E is only ~13.62 (likely lower now with price hikes). With +357.7% Revenue Growth/+804.27% Net Income Growth Y/Y, their profits are astronomical. Similar to Micron and their 57.56%+ last month, and more than doubling since I first mentioned it: The stock price literally can't keep up. While the giants (Samsung, SK Hynix, and $MU) are fighting over the high-end AI market, Nanya specialized in the legacy and specialty DRAM products from PCs, computer electronics, and industrial. (They are doing advanced chips though with 1b Nodes though). And the memory companies left a massive void Nanya filled: Micron - Total Exit from Consumer Lines SK Hynix - Focusing almost entirely on HBM4 and server-grade DRAM. Samsung - Reallocating 30–40% of legacy lines to advanced AI memory. Nanya - Last man standing Demand stayed steady but legacy DRAM supply suddenly fell off a cliff. And it's still widely used from: - Consumer PCs and Laptops: Millions of computers built between 2015 and 2023 use DDR4 - Smartphones: Most mid-range and budget smartphones use LPDDR4 - Home Electronics: Smart TVs, gaming consoles (like the older Nintendo Switch or PS4) use these. Basically, they still need to make these, but Nayna has all the cards. Buyers are paying a "scarcity premium" for older memory because they can't find it anywhere else and seeing a supply shock in a boring market. Nanya essentially became the guy selling water in a desert. And they can keep jacking up the price of water. I'm long on the cat sounding company selling water in the desert.
-
解析AI芯片散热瓶颈：CTE不匹配致翘曲，AlSiC成Rubin架构候选方案。
熔点是完全无关的信息。向2000W功率演进导致的问题源于热膨胀系数(CTE)不匹配。 铜的膨胀速度快得多，这会导致翘曲。铝硅碳化物(AlSiC)并非“新”材料，因为它早已用于高铁到高超音速导弹等领域。 只是碰巧在Rubin架构世代出现了新的应用场景，而在Blackwell架构中并不需要它。 实现路径只有几种，AlSiC只是可能的候选方案之一。
英文原文
Melting point is completely irrelevant information. The move to 2000W causes stress due to CTE mismatch. Copper expands multitudes faster, which causes warpage. AlSiC isn't "new" since it's been around for high speed rails to hypersonic missiles. It just so happens that there's a new use case with Rubin generations where it wasn't needed with blackwell. There are single digit different ways to go about it, AlSiC is just one of the likely candidates.
-
看空CRWV因债务与折旧风险，建议多NBIS/CIFR空CRWV对冲。
$CRWV 可能是我能想到的唯一一家市值超 500 亿美元且很可能破产的公司。至少 $ORCL 拥有盈利的核心业务，并持有大量资产，例如 15% 的 TikTok 股份。 对我来说，做空 Coreweave 的理由很明确，纯粹因为其大部分收入将用于支付债务利息。再加上 GPU 折旧，情况更是雪上加霜。 但进行对冲是个好主意。我个人会做多 $NBIS、$CIFR，同时做空 $CRWV，这对长期投资来说似乎不错。Nvidia 目前仍在维持它们的生存。
英文原文
$CRWV is probably the only $50B+ company I can think of that is likely going to go bankrupt. At least $ORCL has a profitable core business and owns tons of assets like 15% of Tiktok. Coreweave is a clear short to me purely because a large part of their revenue will be used to pay off debt interest. Then you compound that with GPU depreciation. But it's a good idea to hedge. I would personally long $NBIS, $CIFR short $CRWV, seems decent for long term. Nvidia is keeping them alive for now.
-
博主回复粉丝，仅包含表情符号，无实质内容。
@jukan05 🛖🔥🚿💸💾🏦😢
英文原文
@jukan05 🛖🔥🚿💸💾🏦😢
-
看好NBIS和CIFR，指出Coreweave高息债务侵蚀现金流。
我认为 $NBIS 目前极具吸引力。$CIFR（指其托管业务而非新云业务）等特定个股也是如此。相比之下，$CRWV 的说服力则稍逊一筹。 Nebius 的商业模式本质上是将五家年同比增长(Y/Y)达100%的公司合并而成。（包括与 $UBER 合作的自动驾驶公司 Avride、估值约150亿美元的 Clickhouse、AI 标注公司 Toloka 以及教育科技公司）。此外，还包括其核心业务部门，该部门已实现约70亿美元的年度经常性收入(ARR)。 因此，尽管我们需要通过实际产生的自由现金流(FCF)来仔细审视这些乐观的收入数据，但公司方面表示长期 EBIT 利润率有望达到20%~30%。 最关键的差异在于，Coreweave 超过12亿美元的债务利息正在侵蚀其自由现金流，而其他公司则获得了2~3%的低利率融资。
英文原文
저는 $NBIS 가 현재 매우 매력적이라고 생각합니다. $CIFR(네오클라우드가 아닌 코로케이션 부문) 같은 특정 종목들도 마찬가지입니다. 반면 $CRWV 는 그만큼 설득력이 있지는 않네요. 네비우스(Nebius)의 비즈니스 모델은 기본적으로 전년 대비(Y/Y) 100%씩 성장하는 5개의 회사가 하나로 합쳐진 형태입니다. ($UBER와 협력하는 자율주행 Avride, 약 150억 달러의 가치로 평가받는 Clickhouse, Toloka(AI 라벨링), 그리고 에듀테크). 여기에 약 70억 달러의 ARR(연간 반복 매출)을 기록한 주력 사업 부문이 포함됩니다. 따라서 이러한 낙관적인 매출 수치들은 실제 창출되는 FCF(잉여현금흐름)를 통해 면밀히 살펴봐야 하겠지만, 사측에서는 장기적으로 20%~30%의 EBIT 마진을 제시하고 있습니다. 가장 중요한 차이점은 코어위브(Coreweave)의 경우 12억 달러 이상의 부채 이자가 FCF를 잠식하고 있는 반면, 다른 기업들은 2~3% 수준의 저금리를 확보했다는 점입니다.
-
对比新型云融资结构，指出Coreweave债务风险高且英伟达损失可控。
其他如 $CIFR、$WULF 是数据中心托管(colocation)商，它们不购买 GPU。$CRWV、$NBIS、$IREN 则购买。前两家由 $NVDA 注资。后者 $IREN 由 $MSFT 通过预付款购买 GPU。大多数新型云(neoclouds)拥有极佳的融资结构，如 Nebius 的低息可转换债券。Coreweave 是唯一陷入困境的，背负巨额债务，因此若无垃圾债(junk bonds)支持，很难获得额外资金购买更多 GPU。英伟达给 Coreweave 泼了冷水，投入 Coreweave 的 20 亿美元最终仍回流至 $NVDA。如果 Coreweave 因巨额债务和贬值的 GPU 而倒闭，英伟达损失并不大。
英文原文
Others like $CIFR, $WULF are colo. They don't buy GPUs. $CRWV, $NBIS, $IREN do. The first two $NVDA funded. The latter one, $MSFT did with prepayment for Iren to buy GPUs. Most neoclouds have great financing structures like low interest rate Nebius convertible notes. Coreweave is the only one on fire, with massive debt, so it's harder for them to secure additional funding to buy more GPUs without junk bonds. Nvidia threw a bucket of water on Coreweave and the $2B into Coreweave goes back to $NVDA anyways. Nvidia doesn't lose much if and when Coreweave goes under with just a ton of debt and depreciating GPUs.
-
建议押注英伟达而非高负债的Coreweave
考虑因为 $NVDA 的交易而买入 $CRWV 吗？ 你可以这样思考： 赌场给你更多资金，因此你可以利用抵押品承担更多债务去赌博。 然后所有额外的钱都流回了赌场（即 $NVDA）。 这并不意味着 Coreweave 是一个好的做多标的。 但这确实让英伟达成为了一个。 押注庄家，而不是那个背负 147 亿美元+债务且年利息支付超过 12 亿美元的人。
英文原文
Thinking about buying $CRWV on the $NVDA deal? Here's the way to think about it: A casino gives you more money, so you could take on more debt with collateral to gamble. Then all the extra money goes back to the Casino (aka. $NVDA). That doesn't make Coreweave a good long. But it does make Nvidia one. Bet on the house, not the one with $14.7B+ in debt with $1.2B+ in yearly interest payments.
-
澄清AlSiC与铜在散热中的不同角色及后处理瓶颈价值
在极小的真空中审视局部，从而否定整个多维度的帖子，并非明智之举。 AlSiC（铝硅碳化物）被用于高超音速导弹以承受高温不变形，而非铜，是有原因的。或者它被用于使用 AlSiC 散热片的火箭鼻锥。 我来解释一下其中的细微差别：它们的用途不同。 铜在材料熔化前提供最大热传递，热导率表现更佳。而 AlSiC 基本作为外部结构，在热冲击下维持形状。我们可能会看到铜-金刚石或 AlSiC 的混合材料，因为每种材料各有专长。 AlSiC 尤其用于翘曲控制，因为铜会膨胀，你们在 Blackwell 上已经看到了这一点。 至于制造，情况很复杂。我从未说过它（成本/难度）更高。我说的是前道工序可能面临大规模生产的困难。 但 AlSiC 可能存在后处理瓶颈，如果它成为 2000W+ 架构的组件，从事此环节的公司将更具价值。
英文原文
Not the best idea discredit the entire multifaceted post by looking at something in a tiny, tiny vacuum. There's a reason why AlSiC is used in hypersonic missiles to survive temperature without warping instead of copper. Or why it's used for the nose cone in rockets which use AlSiC heat spreaders. I'll help explain the nuance: they're both used in different ways. Copper for maximum heat transfer before the material melts and it does a better job with thermal conductivity. And AlSiC, which is basically the exterior to maintain the shape under thermal shock. We might see some copper-diamond or AlSiC mix since each material specializes in something different. AlSiC especially is used for warpage control since copper expands and you've already seen that with Blackwell. As for manufacturing. It's nuanced. I never said it was higher. I said former processes likely high mass-production difficulty. But there is likely a post-processing bottleneck with AlSiC, which makes the companies that do this more valuable if it becomes a component for 2000W+ architectures.
-
关注8英寸晶圆价格上涨趋势，认可相关分析观点。
@zephyr_z9 8英寸晶圆(wafer)价格上涨确实也是我会关注的一个点！ 很棒的论点(thesis)
英文原文
@zephyr_z9 8 inch *wafer* price increases are definitely something I’d look out for too! Great thesis
-
赞同关于南亚科在成熟制程存储领域占据份额的论点。
@zephyr_z9 这句话的前10个字确实有更好的表达方式…… 但论点很棒，我们已经看到南亚科(Nanya)如何在成熟制程(legacy)存储领域占据大量份额，情况正是如此。
英文原文
@zephyr_z9 There’s definitely a better way to word the first 10 words of the sentence.. But great thesis, we’ve already seen how it played out with nanya capturing a lot of legacy w/ memory
-
博主祝贺粉丝里程碑，分享个人关注书签比率的偏好。
@Gubloinvestor 恭喜，这是一个里程碑！很高兴我在过去4周获得了2600万次曝光。 我个人最关注书签比率，因为这意味着人们觉得我的帖子值得收藏。 你的成长速度太快了！ https://t.co/7snSgendL5
英文原文
@Gubloinvestor Congrats, what a milestone! Im happy i got 26M impressions in the last 4 weeks. I personally track bookmark ratios the most because it means people find my posts memorable. You’re growing so fast! https://t.co/7snSgendL5
-
半导体级AlSiC供应链格局及非AI领域应用现状
这里我只讨论材料，本贴不涉及具体公司。 但从供应链角度来看，半导体级铝硅碳化物(AlSiC)的粗略市场份额估算如下： CPS - 30.1%（美国） Denka (TYO: 4061) - 18.7%（日本） Japan Fine Ceramics - 15.6%（日本） MC-21 - 12.1%（美国） BYD - 11.4%（中国） 有趣的是，BYD将其用于电动汽车，而CPS用于国防到高速铁路，因此目前的散热应用非常广泛（尚未涉及AI领域）。
英文原文
I'm just talking about materials here, nothing about specific companies in this post. But in terms of the supply chain, rough est. of semi grade AlSiC looks like this: CPS - 30.1% (United States) Denka (TYO: 4061) - 18.7% (Japan) Japan Fine Ceramics - 15.6% (Japan) MC-21 - 12.1% (United States) BYD - 11.4% (China) It's interesting because BYD use it for EVs while CPS uses it for defense to high speed rails so just all over the board thermal applications usage so far (aside from AI yet).
-
博主计划深入研究增强型硅-28技术。
@MarkosAAIG 我需要深入研究一下增强型硅-28(enhanced Si-28)。我敢肯定那些量子圈(quants)的家伙们对此了如指掌。
英文原文
@MarkosAAIG I need to do more research into enhanced Si-28. I'm sure the quantum bros know all about it
-
高性能计算或成碳化硅铝继航空航天后的新应用领域。
@__Tomster 没错！高性能计算(HPC)看起来是碳化硅铝(AlSiC)在航空航天/工业之后的下一个潜在行业应用领域。https://t.co/Io4vj1Qms1
英文原文
@__Tomster Yep! HPC looks like a possible next sector application for AlSiC after aerospace/industrial. https://t.co/Io4vj1Qms1
-
2027-28年GPU散热瓶颈或致AlSiC材料替代铜铝。
散热可能是2027-2028年GPU的一个问题： 随着散热目标向2000-3000W+范围攀升，铜和铝变得不足。 但AlSiC（铝硅碳化物），一种金属基复合材料，可能变得重要。 原因如下： Rubin VR200 GPU指向1800W TDP，AMD Instinct MI450X据报道高达2500W，$NVDA Rubin Ultra的功耗据报道达到2300W。 向2300W的转变创造了大规模的热通量问题。 AlSiC（铝硅碳化物）是一种金属基复合材料，用于国防、太空和高功率工业应用（如高铁）。 就像Toto等随机马桶公司成为HBM关键一样，这种用于航空航天和工业散热的材料复合材料可能会被用于AI，因为它可以在不发生分层的情况下承受数万次热循环。 因此，随着Rubin等AI加速器达到与工业电源模块或太空相当的功率水平，半导体中AlSiC的采用可能变得重要。 对于2300W系统中的AlSiC： - （结到壳）：包括硅芯片、底部填充物和第一层TIM（热界面材料） - （壳到散热器）：包括均热板（顶盖）和第二层TIM（TIM2） - （散热器到环境）：冷板或液体换热器的热阻。 壳到散热器可能是AlSiC的应用场景。 随着功率密度可能继续向3000W攀升，可能会出现机械和散热危机，这可能导致材料从传统铜封装发生变化。 所以这里有四个部分： 1. 内部/外部散热 看起来SiC（碳化硅）中介层（内部）可能用于$NVDA Rubin代GPU。然后AlSiC（外部）可能用于微通道顶盖（MLCP）或位于芯片中介层组件上方的均热板。 2. 3D垂直堆叠（SoIC） 这些复杂的封装在热循环期间极易翘曲。SiC中介层是脆性晶体，无法提供结构刚性。AlSiC充当“加强筋”，防止基板在高夹紧压力下弯曲。 3. Rubin Ultra NVL576机架可能达到高千瓦功率密度 这种密度造成了SiC中介层无法解决的重量负载瓶颈。Rubin Ultra NVL576机架散热堆栈加液体流道的累积重量可能超过地板承重限制（AlSiC可能成为必要，以在不牺牲传热性能的情况下将散热管理的“死重”减少60%以上）。 4. 微通道的净成型制造 传统铜顶盖必须蚀刻或CNC加工，据报道这一过程在Rubin量产中遇到了高难度。 AlSiC使用“快速注射成型”制造陶瓷预成型体，然后渗入铝。这允许创建复杂的内部几何形状。AlSiC微通道顶盖和硅集成IHS看起来是铜散热管理的替代品。 我们可能会在2026年初看到这一问题的解决，因为SemiAnalysis在2025年报道Nvidia的Blackwell（B100/B200）因CoWoS-L封装中的翘曲问题面临良率问题。 TLDR散热可能是2027-2028年的瓶颈： 一些潜在受益者可能是SiC中介层（高纯度SiC粉末）和用于2027-2028年散热的AlSiC复合材料。 这都是持续的研究，但也许我们会在1-2年内看到一些极其小众且随机的铁路或太空AlSiC供应商被AI使用，就像HBM的马桶制造商一样。
英文原文
Thermal is a likely issue for 2027-2028 GPUs: As thermal targets escalate toward the 2000-3000W+ range, copper and aluminum becomes insufficient. But, AlSiC, a metal matrix composite, may become important. Here's why: Rubin VR200 GPU pointed toward 1800W TDP, AMD Instinct MI450X, reportedly reaches up to 2500W, and $NVDA Rubin Ultra's power reportedtly goes to 2300W. The move toward 2300W creates a heat flux problem of massive scale. Aluminum Silicon Carbide (AlSiC) is a metal matrix composite, used as for defense, space, and high-power industrial applications (eg. high-speed rail). Similar to how random toilet companies like Toto became critical for HBM, this material composite used for thermal management for aerospace and industrial might be used for AI as it can survive tens of thousands of thermal cycles without delamination. So, as AI accelerators like Rubin reach power levels comparable to industrial power modules or space, the adoption of AlSiC in semiconductors may become important. For AlSiC in 2300W systems there's: - (Junction-to-Case): Includes the silicon die, the underfill, and the first layer of TIM - (Case-to-Sink): Includes the heat spreader (lid) and the second layer of TIM (TIM2) - (Sink-to-Ambient): The thermal resistance of the cold plate or liquid heat exchanger. Case-to-Sink is likely the application for AlSiC. And as power densities will likely continue to climb toward the 3000W mark there becomes a mechanical and thermal crisis that likely causes a material change from traditional copper packaging. So there's four different parts to this: 1. Internal/External Thermal It does look like SiC interposers (internal) are probably used for $NVDA Rubin gen GPUs. Then AlSiC (external) may be used for the Microchannel Lid (MLCP) or heat spreader that sits on top of the die interposer assembly. 2. 3D vertical stacking (SoIC) These complex packages are highly vulnerable to warping during thermal cycling. A SiC interposer is a brittle crystal and it cannot provide structural rigidity. AlSiC acts as a "stiffener" that prevents the substrate from bowing under the high clamping pressures 3. Rubin Ultra NVL576 rack likely reaches high KW of power density This density creates a weight-loading bottleneck that SiC interposers cannot solve. Rubin Ultra NVL576 rack cumulative weight of the thermal stack plus liquid manifolds can exceed the floor limits (and AlSiC may become necessary to reduce the "dead weight" of the thermal management by 60%+ without compromising heat transfer). 4. Net-Shape Manufacturing of Microchannels Traditional copper lids must be etched or CNC-machined, a process that has reportedly encountered high mass-production difficulty for Rubin volumes. AlSiC is manufactured using "Quickset Injection Molding" to create a ceramic preform that is then infiltrated with aluminum. This allows for the creation of complex internal geometries. AlSiC Microchannel Lids and Silicon Integrated IHS looks like the alternatives for copper for thermal management. We might be seeing this addressed earlier in 2026 as SemiAnalysis in 2025 reported that Nvidia’s Blackwell (B100/B200) faced yield issues specifically due to warpage in the CoWoS-L packaging. TLDR thermal is a likely a bottleneck in 2027-2028: Some beneficiaries are potentially SiC interposers (high-purity SiC powder) and AlSiC composite for thermal management in 2027-2028. This is all ongoing research, but maybe we'll see some extremely niche and random small railway or space AlSiC supplier be used up for AI in 1-2 years time like toilet makers for HBM.
-
关注VICR的GPU供电技术，预判机构正布局下一代架构瓶颈。
@PhotonCap 我得好好研究一下 $VICR。直接在 GPU 裸片(die)下方供电这一点真的很有趣！ 虽然 2027 年的下一代架构可能会出现许多新的瓶颈，但我确信许多投资机构已经在研究并抢先布局了。
英文原文
@PhotonCap $VICR 을 한번 살펴봐야겠네요. GPU 다이(die) 바로 아래에서 전력을 공급한다는 점이 정말 흥미롭습니다! 2027년 차세대 아키텍처에서는 수많은 새로운 병목 현상들이 발생할 수 있겠지만, 이미 많은 투자사들이 이를 연구하며 선점에 나서고 있을 것이라 확신합니다.
-
指出COHR市值超300亿，并分享潜在供应链瓶颈清单。
@JakeWoods143114 $COHR 是一家市值超 300 亿美元的公司。这是一份关于潜在瓶颈/卡脖子环节（bottleneck/chokepoints）的低调清单。不过这确实是一家很棒的公司。
英文原文
@JakeWoods143114 $COHR is a $30B+ company. This was a list of under-the-radar potential bottleneck/chokepoints. Great company though.
-
某公司掌控全球超两成半导体材料供应，或成2027年散热瓶颈受益者。
这里有一家有趣的公司，恰好拥有全球约 25-30% 的半导体级材料供应链。散热可能是一个问题，因此随着 $NVDA 架构在 2027 年的部署，这可能成为一个潜在的材料复合材料瓶颈。H200（<1000W）没有问题，但随着 Rubin 部署功率升至 2000W，它可能是材料转型的隐性受益者。
英文原文
There's an interesting company on here that happens to own ~25-30% of this semi-grade material supply chain in the world. Thermal is probably an issue, so might be a potential material composite chokepoint with $NVDA architectures in 2027. There was no issue with H200s (<1000W) but as you get higher to 2000w with Rubin deployments, might be a hidden beneficiary of material transitions.
-
分析$BAER高负债导致自由现金流被锁定用于偿债。
我研究了 $BAER。正如你提到的，需要注意的有：2.029亿美元+的债务，4.003亿美元+的优先股负债（需计提利息），以及新的3.315亿美元融资方案。此外，你还需要在模型中计入每年约2400万美元+的债务利息，这会大幅削减自由现金流(FCF)。因此，保守估计自由现金流约为1300万-1900万美元。鉴于有2亿美元+的高级债务和4亿美元+的优先股，这1900万美元的自由现金流基本被“锁定”，用于偿还债务。话虽如此，这些只是我复核过的资产负债表计算，我尚未深入研究公司的业务（如果其盈利能力足够强，或许能随时间推移抵消负债影响）。
英文原文
I looked into $BAER. Something to look out for as you mentioned is that there's $202.9M+ debt, and $400.3 Million + preferred stock liabilities (which accrues interest), and that new $331.5 Million financing package. Then you need to model in that there's ~$24M+ debt interest a year, which does cut into any FCF a ton. So conservative FCF would be ~$13-19M or so, that ~$200M+ in senior debt and $400M+ in preferreds, that $19M in FCF is basically "trapped" and is being used to pay off debt. That being said these are just the balance sheet calculations that I double checked, I haven't looked much into what the company is doing (maybe it would overshadow liabilities over time if they're that profitable )
-
博主表示重仓英特尔和Meta，关注后续表现。
@Neomeldir 不错！我个人重仓了 $INTC 和 $META，好奇看看这两只股票在接下来的几周表现如何。
英文原文
@Neomeldir Nice! I personally loaded up on $INTC and $META, curious to see how those two do in the upcoming weeks.
-
高股息股暴跌警示：安全感源于成长股而非停滞的股息。
$UNH 刚刚暴跌 -19.48% 至 $283.31。股息投资者(Dividend investors)和巴菲特跟单者(Buffet copytraders)今年过得并不轻松。来自 $PYPL 的跌幅为 -37.8%（1年），$FISV 跌幅为 -72%（1年）。这是一个残酷的警钟：通过追逐 $UNH 约 2.85% 的“股息”或进行价值投资，单日 20% 的下跌（如今天）实际上抹去了 6.6 年的等待。安全感并非来自停滞和股息，而是来自像 $NVDA、$META 或 $TSM 这样最明显的成长型公司。
英文原文
$UNH has just crashed -19.48% to $283.31. Dividend investors and Buffet copytraders are not having a fun time this year. Others from $PYPL are down -37.8% 1Y with $FISV down -72% 1Y. This is a brutal wake-up call that by chasing a ~2.85% "dividends" with UNH or value investing, a single-day drop of 20% (like today) effectively wipes 6.6 years of waiting. Safety isn't found in stagnation and dividends. It's found in the most obvious growth companies like $NVDA, $META, or $TSM.