AVGO: Buy the Dip Before the Rocket Takes Off

Risk disclaimer: 投资需谨慎

大厂纷纷上调AI相关capex，增加ASIC芯片使用，将是对AVGO极大的利好

Broadcom Represents One of the Cheaper Ways to Play Artificial Intelligence

Why Taiwan Semiconductor and Meta Could Be the Hidden Bull Case for Broadcom

Goldman Sachs notes that the TPU v7 offers a 70% reduction in cost per token compared to the TPU v6. This drastically improves the economics of running AI workloads and is “on par” with NVIDIA’s Grace Blackwell 200, according to Goldman.

TSMC also reports growth projections for AI-accelerator-specific revenues. This is the exact category of technology that Broadcom’s XPUs fall into, along with NVIDIA’s graphics processing units (GPUs). Within AI-accelerators, TSMC expects sales to grow at a compound annual rate approaching the mid-to-high 50% range from 2024 to 2029. This is a stark increase over the company’s previous long-term growth projections in the mid-40% range. This is another positive sign for Broadcom, showing that AI-accelerator demand may have more staying power than some expected.

META earnings

In Q1, we will extend our MTIA program to support our core ranking and recommendation training workloads, in addition to the inference workloads it currently runs.

Screenshot2176×1270 99.1 KB

对AI infra建设投入加大

Morning Star Dec 12, 2025

Bulls Say

Broadcom is an exemplar of operating efficiency. It earns excellent operating margins and generates enormous cash flow. It is particularly strong at acquiring companies and trimming excess expenses.

Broadcom’s networking and wireless chip businesses boast best-of-breed technologies, in our view, along with marquee customer relationships with Apple, Google, Cisco, Arista, and others.

We believe Broadcom will be a significant beneficiary of rising AI spending, which we expect to spur significant growth for its networking semiconductor sales.

Bears Say

Broadcom’s chip business bears significant customer concentration, with a small handful of large AI customers driving the bulk of revenue and future growth.

Broadcom’s software portfolio holds legacy and mature businesses, like virtualization and mainframes, which we think will exhibit moderating growth.

Broadcom relies heavily on acquisitions to expand its portfolio and may struggle to find deals large enough to move the needle that can pass antitrust scrutiny.

Analyst Note

Broadcom posted strong fiscal fourth-quarter results above guidance. Guidance for January-quarter artificial intelligence revenue was impressive, as was a new $11 billion AI chip order from Anthropic for the second half of 2026, and the announcement of a new custom AI chip customer in 2026. Why it matters: Broadcom’s AI chip business is accelerating, and we project even greater astronomic growth over the next two years. We see phenomenal demand for Google’s TPU chip, layering in of new customers, and big incremental orders from existing customers driving immense demand. Google is the firm’s lead AI customer, and the latest generation of its TPU chip exhibits superb performance that is driving significant orders. The TPU is no longer exclusively internal to Google—we now expect significant orders from Anthropic and Meta for these chips and systems. Anthropic will accrue revenue in the second half of fiscal 2026, with $21 billion in wholly incremental orders. Broadcom’s brand new customer adds $1 billion in incremental revenue in fiscal 2026. In 2027, we expect OpenAI to supplement the firm’s existing five customers. The bottom line: We raise our fair value estimate for wide-moat Broadcom to $480 per share, from $365, behind a significantly stronger AI chip growth forecast. We see a 5% after-hours selloff as missing the forest for the trees. Investors now have a terrific opportunity to buy into an AI winner. We attribute the selloff to commentary on gross margin dilution from AI chips. We aren’t concerned with this, given that these chips are operating-margin-accretive. The firm’s AI backlog might’ve missed investor targets too, but we see upside to this and view it highly positively. We now expect an even larger surge in second-half fiscal 2026 revenue behind the second Anthropic order. We forecast total AI chip revenue to far more than double in fiscal 2026, and see significant momentum continuing into 2027.

Broadcom’s custom AI chip, or XPU, business, is firing on every cylinder imaginable. With six customers announced and rising unit demand from each, we see a highly constructive demand environment through the next two years. Broadcom’s AI chip backlog of $73 billion over the next 18 months exceeded our model, and we believe the firm will report actual revenue above this figure during that period as more new orders come in.

Google’s latest seventh-generation TPU, dubbed Ironwood and co-designed with Broadcom, is having a moment among AI model builders. Google’s leading frontier Gemini 3 model was trained exclusively with TPUs, and now Google’s competitors are moving partial workloads off of Nvidia GPUs and onto TPUs, led by Anthropic and Meta. Broadcom announced Apple has also been a TPU customer, which was news to us. All of the above is constructive for Broadcom as the designer of these chips. More TPUs sold, either directly to Google or directly to other customers, drives revenue growth. We view positively that Anthropic’s orders are for entire systems, rather than just chips, as we like Broadcom moving toward full-stack systems and following in Nvidia’s footsteps. This also leverages Broadcom’s best-in-class networking prowess.

We expect Google unit growth, Anthropic shipments, and shipments to the new fifth XPU customer (which is not OpenAI, so could be Apple or any number of other firms) to drive accelerating growth in fiscal 2026. In fiscal 2027, we see further TPU demand growth and the ramp of OpenAI’s custom chip as the biggest drivers. We’re skeptical of the three-year timeline for adding 10 gigawatts of capacity with OpenAI, given power constraints, but we see this deal adding significant revenue nonetheless. We estimate roughly $15 billion in revenue for Broadcom per gigawatt added with OpenAI as part of the deal. In 2027 and beyond, we expect to see additional customers onboarded to further add to total AI chip demand, including Broadcom potentially qualifying in Microsoft and Amazon’s custom chips alongside Marvell. Management believes AI chip revenue can double again in fiscal 2027—this is not priced into the stock today and would represent upside to our valuation.

Management was highly constructive on supply, which is positive amid a broadly constrained AI infrastructure environment. They appear to see no constraints to their ability to more than double AI revenue in fiscal 2026 and double it again in fiscal 2027. We liked to hear that the firm is building out new advanced packaging capacity at a fab in Singapore, given that this is one of the main constraints in the AI chip supply chain.

We view the quarter, the guidance, and the deal momentum in AI chips all as extremely constructive for Broadcom and warranting a higher fundamental valuation. We believe a selloff based on gross margin concerns is ignoring the grander picture. Demand is accelerating from already-high levels, not supply-constrained, and accretive to operating margins, earnings, and cash flow. If the selloff is instead due to the $73 billion backlog meeting, rather than exceeding, investor expectations, we also see this as shortsighted. We expect this backlog to grow meaningfully over the next few quarters and for actual revenue to well exceed that figure over the next 18 months. In summation, investors have a rare opportunity to get Broadcom at a discount following a blowout quarter. This is a blue-chip AI share-taker that has a tremendous growth opportunity ahead over the next two years. Buy the dip.

Broadcom posted strong fiscal fourth-quarter results above guidance. Guidance for January-quarter artificial intelligence revenue was impressive, as was a new $11 billion AI chip order from Anthropic for the second half of 2026, and the announcement of a new custom AI chip customer in 2026. Why it matters: Broadcom’s AI chip business is accelerating, and we project even greater astronomic growth over the next two years. We see phenomenal demand for Google’s TPU chip, layering in of new customers, and big incremental orders from existing customers driving immense demand.

Google is the firm’s lead AI customer, and the latest generation of its TPU chip exhibits superb performance that is driving significant orders. The TPU is no longer exclusively internal to Google—we now expect significant orders from Anthropic and Meta for these chips and systems.Anthropic will accrue revenue in the second half of fiscal 2026, with $21 billion in wholly incremental orders. Broadcom’s brand new customer adds $1 billion in incremental revenue in fiscal 2026. In 2027, we expect OpenAI to supplement the firm’s existing five customers. The bottom line: We raise our fair value estimate for wide-moat Broadcom to $480 per share, from $365, behind a significantly stronger AI chip growth forecast. We see a 5% after-hours selloff as missing the forest for the trees. Investors now have a terrific opportunity to buy into an AI winner. We attribute the selloff to commentary on gross margin dilution from AI chips. We aren’t concerned with this, given that these chips are operating-margin-accretive. The firm’s AI backlog might’ve missed investor targets too, but we see upside to this and view it highly positively.We now expect an even larger surge in second-half fiscal 2026 revenue behind the second Anthropic order. We forecast total AI chip revenue to far more than double in fiscal 2026, and see significant momentum continuing into 2027.

Uncertainty

We assign a High Morningstar Uncertainty Rating to Broadcom. As a chipmaker, Broadcom is vulnerable to market supply and demand cycles. Though it has been able to offset cyclicality in recent years with its software exposure and networking strength as a buoy, future cycles may not look similar. It also operates with a high reliance on TSMC for its chip supply, and supply constraints could hamper its ability to ship to customers. Nonetheless, we believe Broadcom is a preferred customer of TSMC and would earn high priority in such a scenario—both for its scale and lengthy relationship. We see risk arising from customer concentration and AI: Broadcom is increasingly reliant on its AI accelerator business, which is made up of a handful of high-spending customers like Google. Changes in spending patterns from these customers and overall AI demand can create fluctuations in Broadcom’s results and affect market sentiment on its stock. We also see medium-term displacement risk in Broadcom’s wireless chip sales into Apple and expect Broadcom to lose its Wi-Fi and Bluetooth chip sales into iPhones and other products as early as 2025. Nevertheless, sales to Apple are a minority of sales compared with the larger networking chip segment.

We believe there is key-man risk with CEO Hock Tan, who is in his 70s. Tan has been integral to Broadcom’s acquisition and operating strategy, and new management may fail to realize the impressive efficiencies of the current team. Even with Tan at the helm, Broadcom’s reliance on acquisitions is risky. The firm is encountering greater regulatory pushback and seeing the size of targets inflate as the business grows. If Broadcom can’t find sufficient targets, or overpays for a deal, it could destroy value. On the environmental, social, and governance front, we see little risk for Broadcom, primarily resulting from hard-to-find engineering talent for wireless and networking chips.

avgo已经套了一个月了

中间不是给过解套机会吗

1/5的仓位，本来收益和googl一样的，最近不知道怎么阴跌少赚一半了都

黄金坑，冲吗

问题不大的，比NVDA增长快。今年HBM 60% 被nvda吃了,20%被Google 吃了，而Google asic 最大的合作伙伴就是avgo.

AVGO又倒车接人了

https://www.reuters.com/world/asia-pacific/broadcom-expects-sell-1-million-3d-stacked-chips-by-2027-2026-02-26/

总结

This breaking headline from Reuters is a major development and provides exactly the kind of concrete revenue visibility that Wall Street institutions are using to justify their $400+ price targets for Broadcom.

The announcement that Broadcom expects to ship 1 million 3D-stacked chips by 2027 ties directly into our previous discussions about AI capital expenditure, valuation, and the shifting profit margins. Here is a breakdown of why this specific news is moving the needle.

What Are 3D-Stacked Chips? (The Technology)

In the context of AI and advanced computing, “3D stacking” (often associated with advanced packaging technologies like TSMC’s SoIC or CoWoS) involves stacking chiplets or memory vertically rather than placing them side-by-side on a motherboard.

• Why it matters for AI: This drastically reduces the physical distance data has to travel, massively increasing bandwidth and reducing power consumption. For hyperscalers (like Google and Meta) running massive AI data centers, power efficiency and data transfer speeds are the biggest bottlenecks. Broadcom supplying 3D-stacked custom silicon means they are providing the exact cutting-edge hardware these tech giants desperately need.

The Valuation Impact: Validating the $400+ Targets

A projection of 1 million highly advanced chips by 2027 is a massive volume for custom, top-tier silicon.

• Billions in Locked Revenue: These aren’t cheap consumer chips; 3D-stacked data center accelerators and high-end networking switches command premium prices (often tens of thousands of dollars per unit). Selling a million of these provides a highly credible, multi-year runway for AI revenue.

• Beating the “One-Hit Wonder” Fear: Skeptics of Broadcom’s valuation often worry that the current AI boom is a temporary spike. By publicly projecting sales out to 2027, Broadcom is signaling that hyperscaler demand is structural and sticky, not just a one-time spending spree.

Tying it Back to the “Margin Squeeze”

This announcement perfectly illustrates the “margin squeeze” dynamic we just discussed.

• High Production Costs: 3D packaging is notoriously difficult and expensive. Broadcom relies heavily on foundries like TSMC to manufacture and package these chips. Broadcom will have to pay TSMC a premium for this advanced packaging, which acts as a “pass-through” cost.

• The “Dollars Over Percentages” Reality: Because these chips cost so much to make, the percentage of profit Broadcom makes on each chip might be lower than their traditional software sales. However, because the final sale price is so high, the actual dollars of profit generated from shipping 1 million of these units will be astronomical.

The Setup for March 4 Earnings

Releasing this kind of forward-looking guidance just days before their March 4 Q1 earnings report is a strong signal of management’s confidence. It gives institutional investors a tangible metric to point to when defending the stock against short-term macroeconomic volatility or options-market anxiety.

周一开空AVGO

ASIC就是个口嗨，这股合理估值就是网络部分加上VMWARE。虽说现在这个价格再做空没有很好的收益率了，但也不至于做多啊。

坚定看多AVGO，但是我财报前可能会考虑减仓/清仓，因为达子那么好的财报都要跌

咋就劝不动呢 这玩意真没内在价值的

AVGO NVDA就算涨不起来，我觉得也是和美股资金出逃有关，像韩股 日股都涨疯了

NVDA是这个逻辑，因为人家有核心技术，有市场，有业绩，宏观改善了分分钟翻倍。

AVGO涨不起来跟这个一点关系都没有，市场上再多十倍流动性也涨不起来，没有技术，没有业绩支撑，不可能虚空造估值

你又来啦？

感谢avgo，我做多avgo拿到了今年第一第二笔盈利。我相信后面会更好。

IMG_38361179×1855 196 KB

但是nvda贵，AVGO的moat就是他便宜，降低芯片成本和ai模型的运行成本。厂商增大非n卡的意思也是要让nvdia降价。

这些不重要，卖put在什么股上都能赚钱，因为市场过高定价波动率。你随便选一只股，选最垃圾的，卖-50%价外put，都是绝对赚钱的。卖价外put都不能称之为做多，最多叫做空波动率。

你跑个回测，如果卖同价位的call赚更多。你再跑个回测，这股的股价绝对值，是不是一直跌的。

那寒武纪、AMD更便宜，是不是应该all in寒武纪、AMD？现在厂家最在乎的明显不是便宜，谁买的芯片性能差谁就会输掉AI竞争，输掉未来。

Expected move \pm 8.79%

期待一份爆炸好的guidance

IMG_72581500×3902 540 KB

【引用自 platinum_go_go】:
那寒武纪、AMD更便宜，是不是应该all in寒武纪、AMD？现在厂家最在乎的明显不是便宜，谁买的芯片性能差谁就会输掉AI竞争，输掉未来。
如何评价最近Meta跟AMD签的大单？是说明Meta不打算要未来了吗？

怎么还没bo起啊。。。难道还要等两天啊，等的我花都谢了。什么时候反弹啊。。

Meta这公司早就确定要完了，你看看Meta现在还有像样的模型吗？这公司就像是人得了绝症，在将死之际总想着整一些偏方。

其实这公司才是我做空的最大仓位，我以为做空Meta和做多Nvidia Google一样已经是共识了，就没发过做空贴。

Here’s a list of the 10 most shorted mid-to-mega cap technology stocks as of February:

SoundHound AI (SOUN) – 34.59%
CleanSpark (CLSK) – 33.42%
Ondas (ONDS) – 33.23%
MARA Holdings (MARA) – 29.34%
Applied Digital (APLD) – 28.73%
IonQ (IONQ) – 22.59%
TeraWulf (WULF) – 22.54%
Core Scientific (CORZ) – 19.11%
Navitas Semiconductor (NVTS) – 18.71%
Dropbox (DBX) – 18.67%

Least shorted mid-to-mega cap tech stocks as of February:

EverCommerce (EVCM) – 0.65%
Microsoft (MSFT) – 0.76%
Ubiquiti (UI) – 0.80%
Apple (AAPL) – 0.91%
Arista Networks (ANET) – 1.01%
Bel Fuse (BELFA) – 1.03%
NVIDIA (NVDA) – 1.05%
Broadcom (AVGO) – 1.05%
Oracle (ORCL) – 1.17%
Shopify (SHOP) – 1.18%

Gemini_Generated_Image_wychl7wychl7wych1888×2240 537 KB

有没有人赌财报？

数据上来说挺多吧，put/call ratio 1.06，或许比达子一边倒的call更容易涨。

有人有VM projection么？FY25大概42%的收入是从 VMware cloud Foundation，财报里面叫Infrastructure Software revenue。我理解硬件没问题，但saas部分很大的不确定，如果有supply chain check/guide能给看看么？

我记得software部分预期是1% decline，因为转订阅制而且免费开放给个人用户

我看FY25 Q3是6.9B Infrastructure Software revenune up 19% Yoy；Q4 6.8B 17% Yoy，如果是1% decline是指yoy?还是什么？

yoy。

前两个月走了大部分avgo底仓，还留了点等着火箭起飞

拷打了一下ai，因为基数变化，yoy增长调整为2-4%了。大概是和gross margin预期记混了。

avgo 财报肯定是beat的。

但是，财报beat不意味着涨。

要大大大大的beat，出乎意料的beat，跌破眼镜的beat。。才可以。。

两位数beat也不一定涨。只能说抓紧volatility大的时间，多sell put

成本还行，跌也不会跌到哪里去，最差应该也就跌到300吧

不知道。但是卖点280-290左右的put不香吗。

put不赚钱呀

buyside survey ~33% vs sell side 29%；但软件的growth会拖累整个给PGM，然后另一个问题就是他们CoWos到底拿到了多少配额。

image1091×287 15.9 KB

tsmc好像并不会透露产能分配信息，我看这个信息大概是去年的fundaai的一篇付费report，准不准就不知道了

可以上put spread 拉杠杆

avgo option flow看起来是挺clean的，都是卖put，但是价格比较低，看起来比较保守。

跟着大军卖的280put 重在参与～

这波动也太小了吧

前几周都是5%起步的

现在挣鸡腿钱还给配韭菜了？

起飞了？

來來回回又回到330 無事發生

希望坚持到明天！这么多回购协议还撑不起吗？

今天HOCK第一次具体说了客户，之前一直含糊。同时，Dario说的30~40GW in 2027，现在能看到路径。

Google — longstanding TPU partner, estimated ~3GW in 2027
Anthropic — 3GW disclosed, at ~$20B/GW pricing (the former “mystery $10B customer” revealed in Dec 2025)
OpenAI — 1GW disclosed, ~$20B/GW (prior to the call, someone asked “is AVGO expected to confirm OpenAI ASIC on the call?”

— suggesting it was confirmed)
Meta — scaling to “multiple” GW in 2027, estimated ~$15-16B/GW
Customer 5 + 6 — still unnamed, ~1GW combined at ~$15-16B/GW

image1038×582 10.6 KB

“Fully secured” demand from 2026 to 2028 — meaning signed contracts, not

aspirational
10 GW of total deployment for 2027 — when Stacy Rasgon floated 6GW at $20B/GW, Hock

corrected him to 10GW
AI chip revenue >$100B for 2027 (includes networking)
XPU customers likely to stay on direct-attached copper
Margins not dilutive from adding new customers — refuting what was said on the

prior earnings call

家人们，AVGO能重返400吗

现在从MS TMT传（根据Dario 1.9b ARR，30-40GW by 2027 projections，几大hyper加单各50b的等值GW，你算下，看看你觉得AVGO NV多少钱合适，涨的快的肯定是分子，如CLS VRT这些，但万亿公司至少能让你睡好觉，大盘股需要低利率这样的催化剂。

问题是openai和anthropic不一定拿得出这么多钱，Meta进度不咋样，搞不好就放弃自研了

你这个问题应该先想清楚，为啥看好AVGO，不能说因为哪些风险或者路边消息，来决定一个投资计划；应该先了解清楚AVGO是什么，对比其他东西比如NV 存储等，AVGO的风险在哪里，然后r/r是什么，多花点时间。

可能是Amazon, 据说Amazon是Alchip + broadcom + mrvl三家供应商

AWS T3是AICHIP大头+MRVL，确实有消息avgo may have a chance。

US mulls new rules for AI chip exports, including requiring US investments by foreign firms

NVDA/AMD的利空就是AVGO利好

大体上，是，AVGO 很可能是这三家里受这类新规直接冲击最小的一家。但要区分“直接冲击最小”和“股价波动最小”——前者我基本同意，后者不一定。 

更准确的判断：

直接监管暴露度排序，通常是：NVDA > AMD > AVGO。

原因很简单：这次讨论中的新框架，核心还是围绕AI 芯片跨境出口许可，尤其是大额订单、主权参与、安全保证、甚至要求对美国 AI 基础设施投资。Reuters 报道称，草案考虑把 20 万颗及以上芯片出口与外国政府承诺、美国本土投资等条件绑定，这本质上最先冲击的是靠高端通用 AI GPU 大规模跨境出货的公司。NVDA 正是这条链条的中心，AMD 次之，而 AVGO 的 AI 业务更多是定制 ASIC + 网络芯片 + 互连，再加上它还有很大的基础设施软件收入缓冲。 

为什么 AVGO 受影响最小：

第一，收入结构更分散。Broadcom 2025 财年总收入里，半导体约占 58%，基础设施软件约占 42%。也就是说，它不是纯粹靠 AI 芯片单一逻辑驱动的公司。即便 AI 出口审批变慢，软件部分不直接受这类芯片许可制度打击。 

第二，AVGO 的 AI 曝险不是 NVDA 那种“标准化高端 GPU 全球铺货”模式。Broadcom 在 10-K 里把 AI 相关半导体定义为 custom accelerators / XPUs、以太网交换与路由芯片、NIC、PHY、光组件、以及相关机架系统，客户主要是 hyperscalers 和 frontier-model 公司。这个模式更偏少数大客户深度定制，不是靠向大量海外买家卖标准 GPU。受监管时，项目审批摩擦会增加，但被一刀切卡住的概率通常低于 NVDA/AMD 的通用 AI GPU 出货模式。 

第三，Broadcom 对中国“最终需求”的真实依赖可能低于表面数字。它 2025 财年按发货/交付地口径，17% 收入来自中国（含香港），但公司自己明确说，其中相当一部分其实最终流向美国和欧洲终端市场，因此“最终依赖中国终端客户”的比例明显更低。换句话说，Broadcom 的中国口径里有不少是供应链中转，不等于和 NVDA/AMD 那种对中国 AI 计算市场的直接暴露。 

为什么 NVDA 受影响最大：

NVIDIA 自己在最新 10-K 里已经写得很直白：

它对中国的数据中心计算市场在现有规则下“实际上被排除在外”；2025 年 4 月的限制导致 H20 相关库存和采购义务产生 45 亿美元费用；到 2026 年 2 月，H200 对中国也只是拿到“小规模”许可，且尚未形成收入。这个级别的监管敏感度，在三家里最高。 

AMD 为什么比 NVDA 稍轻，但仍明显重于 AVGO：

AMD 的 MI308 在 2025 年也因新的美国出口限制遭遇库存和相关费用冲击。AMD 2026 年 10-K 披露，2025 财年与 MI308 出口限制相关的净库存和相关费用约 4.4 亿美元；Reuters 此前还报道 AMD 预计 2025 年营收受影响约 15 亿美元。这说明 AMD 也已经被证明对 AI GPU 出口政策高度敏感，只是绝对体量和市场中心性仍低于 NVDA。 

但 AVGO 不是“免疫”。它有三条潜在受伤路径：

1. 大客户海外数据中心项目审批变慢。

Broadcom 的 AI 业务高度集中在 hyperscalers。草案如果把 20 万颗级别的大单和政府承诺绑定，海外 sovereign AI / hyperscaler 项目推进节奏可能被拉长。Broadcom 不一定丢单，但可能出现确认收入延后。 

2. 客户集中度高。

Broadcom 2025 财年有一个客户占其总收入 32%。这种结构的好处是订单大、粘性高；坏处是任何一个大客户项目节奏变化，都会放大到季度数字。也就是说，AVGO 直接受政策打击较小，但间接受客户 CAPEX 节奏影响并不小。 

3. 市场交易层面会被半导体板块一起砸。

就算基本面受损较轻，只要新闻标题是“全球 AI 芯片出口许可全面升级”，量化资金和 ETF 资金先卖整个 AI/半导体板块，AVGO 往往也会跟跌。区别在于：它的基本面修复通常比 NVDA/AMD 更快，前提是订单没有实质取消。这个判断是基于前述收入结构与产品形态作出的推论。 

结论：

• 按基本面直接受损程度看：AVGO 大概率是三者里最轻的。

• 按短线股价波动看：未必最轻，因为市场会先按“AI 半导体”统一定价。

• 按中期修复能力看：AVGO 很可能强于 AMD，也常常强于 NVDA，因为它有软件收入缓冲，而且 AI 业务偏定制化和网络化，不完全等于“高端 GPU 出口故事”。 

一句话压缩：

如果这条新规继续推进，NVDA 是首当其冲，AMD 次之，AVGO 受的是二阶冲击，不是一阶冲击。

不止V了 还绿了

这个卖av的公司涨得还挺多哈

下周去400

这得拍多少av呀