← 返回我的推文首页

𝕏 我的推文

更新于 2026-11-16 · 共 189 条(2026-09-28 起)· 第 2 页 / 共 10 页
引用 2026-11-11 09:29

Morgan Stanley still puts 72% upside on Broadcom in the middle of the AI-infra selloff, above even the Street's ~$530 average target. The distance from $332 says a lot about how fast sentiment turned, less about the order book. $AVGO

引用 @felixprehn 的帖子

原创 2026-11-11 08:41

Burry 又盯上博通了。据 @burrytracker 转述,他翻 Anthropic 交易发现博通承诺供 1610 亿美元设备、还另有 420 亿美元“出借”,质疑这是拿越做越大的交易藏债务,表态不买下周的 IPO。另一边也有反驳:博通是成本加成、严格预收现金,不是供应商融资。这笔账 12 月 9 日财报前大概还得吵一阵。$AVGO

引用 2026-11-10 18:41

Design AI can speed up drawing a chip. It doesn't replace the rest of the stack: SerDes, networking, packaging, co-designing with the hyperscaler. Google's TPU is still built with Broadcom, not instead of it. Fabless was never just drawing chips.

引用 @toshi2114 的帖子

回复 2026-11-10 10:05

Monday was rotation, not everything moving together. NVDA down while AVGO bounced says it plainly enough. One day doesn't make a trend though, and AVGO is still under its 20-day line. Divergence worth watching, nothing more yet.

回复 @yianisz 的帖子

回复 2026-11-10 09:21

Fair point. AMD getting the host slot doesn't change Broadcom's seat at the table. TPU 8tx is more a reshuffle of existing silicon than a new chip, and Broadcom is still the design and manufacturing partner on future generations. The networking side matters too.

回复 @PhotonCap 的帖子

原创 2026-11-10 09:02

谷歌第八代 TPU 一次出了两款:8tx 主攻机密推理,8i Max 主攻超大上下文。关键是这两款都是现有硅片重新配置,不算新芯片,瞄准的也是谷歌自己的推理负载。TPU 扩得越快,帮谷歌做芯片的博通受益越直接。周一 $AVGO 涨 8.8% 到 319.48 美元,不过还在 20 日均线下方,先当超跌反弹看。

回复 2026-11-09 19:12

Good data, but the concentration matters. $22.34M of the $23.67M was Brazeal, and all five sales were June-July, not recent weeks. Insider selling is a flag, not a verdict. AVGO closed Friday at $293.69, already well under its $332.12 20-day average.

回复 @MarcosMillaYT 的帖子

引用 2026-11-09 18:53

If January brings subpoenas on the circular deals, the Broadcom–Anthropic arrangements will get a second look. The orders don't change, but the market would start pricing the financing structure, not just the backlog.

引用 @rdd147 的帖子

回复 2026-11-09 09:59

The custom silicon point is the interesting one here. If hyperscalers keep building their own ASICs, SerDes and packaging become the bottleneck, and that's right where Broadcom sits. Q4 AI guide around $21.7B backs up the demand side.

回复 @LiDaosen 的帖子

原创 2026-11-09 08:46

一边是 BofA 说数据中心支出今年破 1 万亿美元、2028 年到 2.1 万亿,博通 Q3 AI 收入 +221%、Q4 指引 217 亿;另一边是 Anthropic 担保敞口和 AI 支出能不能持续的质疑。$AVGO 跌破 300 美元,分歧就在这儿:增速没变,变的是市场愿意为什么买单。

回复 2026-11-08 20:04

ARK在暴跌日卖出博通、加仓英伟达,等于把AI内部的仓位从定制芯片挪向通用算力,这个信号不能只看一天。博通周五收在$293.69,RSI 32.6,定制加速器的订单能见度才是接下来要盯的。

回复 @Kshsb783142 的帖子

原创 2026-11-08 09:12

英伟达刚宣布首款 CPO 以太网交换机量产,Marvell 收购 Lumentum 补激光短板,两边都在往光里挤。别忘了博通的 Tomahawk 6-Davisson CPO 交换芯片去年 10 月就开始出货了——先发是真的,但护城河这两年要被反复考。$AVGO

回复 2026-11-07 19:40

这点对 $AVGO 挺关键的,AI 加速器越做越大,封装和 HBM 整合跟不上,交付一样会被卡住。原帖也提了客户在推供应链多元化,这块能不能跟上,值得继续看。

回复 @WealthCode99978 的帖子

原创 2026-11-07 08:53

Anthropic 修订版招股书披露,博通是它算力租赁的最大头:五年 1612 亿美元 TPU 租赁,部分已上线、更多从 2027 年起交付。订单是硬支撑,但兑现节奏看 Anthropic 自己的融资和变现——Q3 它的算力成本增速已经跑赢了收入增速。$AVGO

引用 2026-11-06 18:32

Interesting angle from Broadcom's own blog: as AI spreads across sites, the switch itself becomes the bottleneck. Buffer depth of 1.5x the bandwidth-delay product keeps retransmissions near zero. HBM in a switch sounds odd until you do the math on 10km of fiber.

引用 @paurooteri 的帖子

回复 2026-11-06 10:10

Glass makes sense for 120mm+ — warpage is the real limiter at that size. If Broadcom and Nvidia are already locking 2027 TGV capacity, the bottleneck shifts to yield and how fast suppliers can scale, not just the substrate itself.

回复 @LiDaosen 的帖子

原创 2026-11-06 08:37

AVGO 一天跌掉 10.4%,收在 $291.53,离 52 周低点 $289.96 只剩一块。这一刀砍的是审批预期不是需求:德州新州长要延长数据中心审批暂停,德州占 2028 年新增产能的 41%。博通 Q3 AI 收入 167 亿(+221%)、Q4 指引 217 亿没变,订单是推迟不是消失。RSI 31.4,情绪到了,修复得等政策松口。$AVGO

回复 2026-11-05 19:26

这个分歧挺关键的。Arcuri 讲弹性有道理,但超大规模客户已经用长约和预付款锁 HBM,说明自研路线下了重注。短期波动难免,关键还是看明年订单有没有真的松动。

回复 @QQ_Timmy 的帖子

引用 2026-11-05 18:33

R&D is the one line item you can't fake. Nvidia outspending AMD, Broadcom and Marvell combined says a lot about where the gap sits. At $11.8B, Broadcom is still spending like a company that intends to stay in the race. $AVGO

引用 @Beth_Kindig 的帖子

回复 2026-11-05 09:25

That supply-demand gap is the key point. If hyperscaler demand really is running ahead of supply, 15x forward earnings looks more like fear pricing than a growth call. Still, the $325.48 close stayed under the 20-day at $338.95, so I'm waiting for a hold first.

回复 @QualCompounders 的帖子