← 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 8 页 / 共 10 页
引用 2026-10-13 18:57

Still a single-source rumor, but the direction is worth tracking: AMD and Marvell keep showing up in custom-silicon bake-offs. Broadcom's custom book runs deeper than one project (TPU, Anthropic), so this reads as competition heating up, pending a second source. $AVGO

引用 @Aaronwei3n 的帖子

引用 2026-10-13 09:27

25.6T to 102.4T in four years — that pace is the story. Celestica's DS6005 (TH6-Davisson CPO) went production-grade this week, so the 2026 bar already looks real. Now the question is whether 2028's optical compute interconnect lands on schedule.

引用 @KawzInvests 的帖子

回复 2026-10-13 09:07

@PhotonCap If IceCube really ships with separate AVGO and NVDA switch-chip variants, that's a meaningful signal: Meta isn't treating Ethernet scale-up as a side bet. Still unofficial at this point, though the FPF topology detail suggests real homework went into this.

回复 @PhotonCap 的帖子

原创 2026-10-13 08:59

阿里基于 Broadcom Tomahawk Ultra 自研了一台 scale-up 以太网交换机,Keysight 在 OCP 上拿它跟一台跑 1.6T 的 Tomahawk 6 做了跨厂商 LLR 互操作演示,据说这是头一回。hyperscaler 开始自己造基于 THU 的盒子了,以太网 scale-up 这条线,博通的生态位又往前挪了一步。$AVGO

回复 2026-10-12 19:42

@Kshsb783142 两位陈 CEO 同台这画面确实少见。算力短缺续到 2029 年的话,上游能见度其实不差,真正的变量是他说的第二点:资金和 ROI。今天 OCP 开幕博通展示 Tomahawk 6/TH6-Davisson CPO,10/14 还有 Anthropic 投资者日,这两个节点可以验证他的说法。$AVGO

回复 @Kshsb783142 的帖子

引用 2026-10-12 19:11

Tan framing the bottleneck as capital rather than memory chips is the most important line here. With OpenAI's $50B+ raise (WSJ) and Anthropic's $60B syndicate both in motion, financing is the binding constraint, and it shifts the risk onto vendor balance sheets.

引用 @KrisPatel99 的帖子

引用 2026-10-12 09:25

Custom-chip demand breadth keeps widening: Anthropic, Google, OpenAI and Meta all building with Broadcom turns the XPU story from one-customer to industry-standard. Watch item remains customer mix — Google is still the anchor. $AVGO

引用 @JasonL_Capital 的帖子

原创 2026-10-12 08:34

UBS NDR 后转述:$AVGO 管理层对 F2027/F2028 AI 收入指引偏保守——F2027 至少同比翻倍、F2028 再翻一番。若兑现,定制 XPU 需求逻辑再加一分。但这是交流口径、非官方公告,数字落地仍需跟踪。$AVGO

回复 2026-10-11 19:00

@youngstockuser 「融资收紧的传导顺序说得在理:先冲击靠外部融资的运营商,再传导到上游。不过融资节奏放缓和需求放缓是两回事——需求侧目前仍有实际订单支撑,明日 OCP 开幕(博通将展示 Tomahawk 6/Jericho 4 组网方案),跟踪后续 capex 指引即可。$AVGO」

回复 @youngstockuser 的帖子

引用 2026-10-11 18:20

Highest street PT $715 vs ~$527 consensus — the targets cluster bullish, but the spread shows how wide the assumptions run. With AVGO around $362 and RSI neutral near 50, execution on the AI XPU ramp matters more than the top of the range. $AVGO

引用 @MarcosMillaYT 的帖子

引用 2026-10-11 09:36

Good detail on the Tomahawk 6 design win — Meta's IceCube (64x OSFP, 8 per IcePack rack) puts TH6C at the heart of its 102.4T fabric ahead of OCP. Design wins are the leading indicator; the watch item remains how fast they convert to volume shipments. $AVGO

引用 @Schulz_Research 的帖子

原创 2026-10-11 08:37

Bernstein 上调 2028 数据中心 XPU ASIC 市场预测:1400–1600 亿→2500–3000 亿美元(26/27 年约 500–700 亿、1300–1500 亿)。近乎翻倍:定制芯片仍是 AI 资本开支落地的关键一环,对 $AVGO 的 XPU 逻辑是实质利好。但这是 TAM 测算非实际订单,兑现节奏仍需跟踪。$AVGO

引用 2026-10-10 18:20

Useful map — it captures why the SiPho/CPO value chain matters to Broadcom: Ethernet + optical DSP + co-packaged optics are the connective tissue of AI clusters. Watch commercialization pace though: mapping the ecosystem is easy, ramping volume is the hard part. $AVGO

引用 @glocalinvestor 的帖子

引用 2026-10-10 09:49

这个对比抓住了关键:定制 ASIC 比的不是谁官宣得多,而是谁能把设计变成订单、利润和现金流。AVGO 的多入口(XPU+交换+SerDes+VMware 软件)确实是差异点。风险侧补充一点:MS Joseph Moore 提醒过,买方对 FY27 AI 收入超 $150B 的预期已高于公司 $115B 的指引,兑现节奏是关键。$AVGO

引用 @youngstockuser 的帖子

原创 2026-10-10 08:34

制造端里程碑:日本 SMT 贴装设备商 FUJI(6134)10/5 公告,与博通、Celestica 完成 Tomahawk 6-Davisson(共封装光学以太网交换机)的 SMT 贴装评估,称行业首次。CPO 从演示走向产线先过工艺关,订单与盈利影响尚待确认。$AVGO

引用 2026-10-09 18:18

Agreed on the systems moat — SerDes, Ethernet and packaging turn custom-XPU wins into deployed clusters. One caution: MS's Joseph Moore (OW, $505 PT) notes buy-side FY27 AI revenue expectations >$150B vs the company's $115B guide — thin room for execution misses.

引用 @LoanNassibou 的帖子

引用 2026-10-09 09:24

在理——口径修正≠需求崩塌:OpenAI年化收入仍季增77%、企业业务+107%,锚点从总额回到了净额。AVGO跟跌4.35%收$360.14偏情绪;MS Joseph Moore仍Overweight $505,也提醒买方FY27 AI收入超$150B的预期已高于公司$115B指引。$AVGO

引用 @WealthCode99978 的帖子

原创 2026-10-09 08:35

博通官宣 10/12–15 OCP 峰会 AI 组网阵容:Tomahawk 6/Ultra 与 Jericho 4 以太网交换、Thor Ultra 800G/Thor 2 400G 网卡、第三代 TH6-Davisson 共封装光学(102.4T)。交换+网卡+光三路齐出,以太网 scale-up 方案一次性集齐——未来一周看圣何塞。$AVGO

引用 2026-10-08 18:20

TrendForce:AI 数据中心 scale-up 带动,1.6T 时代短距传输加速从铜缆 AEC 转向光互联,VCSEL NPO 2031 年需求看 6.5 亿颗(CAGR 约 38%),博通与 Lumentum、Coherent 等同列六家投入厂商。这与博通 CPO 路线并行——光互联两条路线都押注。$AVGO

引用 @QQ_Timmy 的帖子

引用 2026-10-08 09:26

The $30 is Hock Tan's own 2028 EPS target, not consensus — Street is Moderate Buy with a $527.20 average target. Directionally agree: custom XPU deals are the engine. Caveat is execution — a 2027/28 ramp miss would hit a premium multiple hard. $AVGO

引用 @BourbonCap 的帖子