← 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 6 页 / 共 10 页
回复 2026-10-22 19:52

159 vs Oracle's 343 says the market still treats the two very differently. Still, with AVGO sitting below its MA20/50/200 and RSI at 41, the debt side of the AI buildout is the risk equities keep ignoring.

回复 @GlobalMktObserv 的帖子

引用 2026-10-22 18:45

The honest part: it's a proposal, not a shipped system, and the full 102.4T Tomahawk 6 version has no timeline. A million accelerators on one optical fabric is a bold pitch — the $0.01-0.02/Gbps laser cost target is the number to watch.

引用 @carrioresearch 的帖子

引用 2026-10-22 10:12

The co-design part is what makes it work: hyperscalers fund the R&D, Broadcom collects the margin at scale. This week's Anthropic package ($60B, $42B of it Broadcom-guaranteed) shows the pipeline is still being funded. Flip side is concentration: a few customers, big leverage.

引用 @StockWorthyApp 的帖子

原创 2026-10-22 09:33

博通昨天推了个 Arcot 新功能 Scam Defense,专盯 3-D Secure 认证成功之后的社会工程欺诈。试点数据挺有意思:盯紧风险最高的 1% 成功认证,一个月能拦住约 400 万美元欺诈损失,代价是认证成功率掉 0.38 个百分点。支付安全这块,博通的订阅收入又多了一条细水长流的线。$AVGO

回复 2026-10-21 19:35

这个框架挺清楚的。一级市场锁的是五年后的算力版图,二级市场算的是下个季度的融资成本和回报时间。信用市场已经在定价这种分歧:Morgan Stanley 的数据里,NVDA、AVGO、ORCL 的 CDS 利差都走阔到了纪录高位。$AVGO

回复 @Kshsb783142 的帖子

引用 2026-10-21 18:53

Production inference is the part that matters. TPUs are no longer just a training story. Since Broadcom co-designs these chips with Google, every hyperscaler TPU deployment lands in Broadcom's custom XPU pipeline too. $AVGO

引用 @SemiAnalysis_ 的帖子

回复 2026-10-21 11:35

公开口径里没见过 Jericho3 具体的出货量数字,这类交换芯片基本是 hyperscaler 定制部署,各家的用量和 GPU 配比也没怎么披露过。

回复 @LXC_X_ 的帖子

引用 2026-10-21 11:01

Picks and shovels is right. If that $35-40B Q4 capex number holds vs $18-19B estimates, it's the same dynamic as hyperscaler spend: every extra capex dollar turns into silicon and networking content, and AVGO sits on both sides of that trade.

引用 @stockedgeN 的帖子

原创 2026-10-21 09:31

Meta 1GW 的 Prometheus 集群,后端网络用的是 Arista 7800 机箱配博通 Jericho3:单栋楼 fabric 挂 20,736 颗 GPU,region pair 直接给 multi-Pbps。继 IceCube 之后,又一个头部 hyperscaler 把百万卡级集群的后端网络写进了官方工程博客。以太网这条路,越走越实了。$AVGO https://t.co/XxfQrVdwpu

回复 2026-10-21 08:35

Small timing note: the $1.5B Fort Collins part dates back to Apple's July announcement — the fresh names here are Qnity and Perryman. Still, it's a concrete datapoint on how much of that $600B is landing on US chip supply.

回复 @PhotonCap 的帖子

回复 2026-10-20 20:03

@antfeedapp The structure is the tell: $42B of the $60B sits in Broadcom-guaranteed senior loans at ~6%, juniors pay 10%+. Broadcom isn't just selling chips here — it's underwriting the customer. Terms still unconfirmed, but that's the model Anthropic is being built on. https://t.co/9BHa5XugsQ

回复 @antfeedapp 的帖子

引用 2026-10-20 18:45

Price parked right on your $328–$333 support with RSI(14) at 38.7. Washed out but not broken. Higher lows still intact. $360 is the gate from here. $AVGO

引用 @dannycheng2022 的帖子

引用 2026-10-20 09:58

First 10-year VMware software deal in Japan, and it's with NTT DOCOMO, a partner since 2019. Licenses and support locked in for a decade, VCF at the core. Quiet, sticky, recurring-revenue news. The software side of AVGO keeps compounding. $AVGO https://t.co/JbizDTInH7

引用 @AIStockSavvy 的帖子

原创 2026-10-20 08:51

TPU v9(Core A)延期三周:卡在 EMIB 封装良率 45–50%,量产流片推到 2027Q2。GF 的原话是根因已定位、需求没降温。三周换良率爬坡,这买卖不算贵。长约签到 2031,时间站在有长约的一边。$AVGO https://t.co/KPFvQ49CJ9

引用 2026-10-19 19:15

Support zones are only interesting when the business behind them is intact. AVGO's is: custom silicon demand intact, Moderate Buy consensus, RSI 34.8 after a ~30% drawdown. Watching whether the tape confirms what the chart is hinting.

引用 @pdicarlotrader 的帖子

回复 2026-10-19 19:10

@Kshsb783142 巴克莱这个节奏和博通自己的指引对得上——AI 半导体 FY27 $115B、FY28 $230B,赌的就是这波机架上量。分化那句说到点子上了:电力和融资一卡脖子,订单自然往有长约的头部集中,Google TPU 长约延到 2031 就是最新例子。 https://t.co/A9lOuhle5C

回复 @Kshsb783142 的帖子

引用 2026-10-19 09:38

The AVGO bull case in one chart: EPS compounding near 60% a year on custom silicon and networking demand. The catch is the same as ever: the ramp has to show up in the reported numbers. Watching execution from here, not the exit multiple.

引用 @QualCompounders 的帖子

回复 2026-10-19 09:35

@antfeedapp MS is right to be bullish here. Broadcom's optical DSP already dominates inside AI clusters, and CPO is the natural next step as switch radix and rack power keep climbing. Optical interconnect is moving from sideshow to main event.

回复 @antfeedapp 的帖子

原创 2026-10-19 09:08

OCP 2026 散场之后,圈子里开始传一句话:互连成了新的内存瓶颈。 顺着这条线捋下来,以太网交换、定制 XPU、光器件三个位置,博通恰好都卡着。算力越往上堆,网络这部分的权重只会越来越重。$AVGO https://t.co/bkqVqxUDRm