emperinter retweeted
现在的市场有多差,参加了一个 100U 挑战赛
赛季队伍 冠/亚/季军可获得 300U / 200U / 100U的奖励
现在
冠军队伍盈利 - 72U
亚军队伍盈利 - 231U
季军队伍盈利 - 334U
$AAOI is a shitco imo, and demand has nothing to do with it. Demand is fine. The problem is that management's forecasts last about one quarter before they get walked back.
Guidance vs reality:
> Q3'25 call: "meaningful" 800G shipments in Q4'25 → came in below the $4M to $8M they'd expected (firmware)
> Feb: 800G "more than $25M" in Q2 → $12.8M
> Aug: 800G to roughly 5x in Q3 → 3.7x ($47.5M)
> Aug: >$70M of 1.6T revenue in Q4 → now $50M to $70M
> Aug: CEO floats Q4 above $500M → guide is $385M to $420M
> FY26 ">$1.1B" (raised in May, softened to "around $1.1B" in Aug) → implied ~$1.015B, quietly not reiterated
Margins tell the same story:
> Exit 2026 GM target: 35% (May) → 32 to 33% (Aug) → 29.5 to 31% Q4 guide (Nov)
> Q3 non-GAAP GM 28.7% vs a 29 to 30.5% guide, and below 31.0% a year ago
Timelines slip too:
> End 2026 capacity: 650k units/month → >500k (Texas Phase 2 delayed "several months" on power)
> $471M/month revenue run rate: mid 2027 → end 2027
> 2025 capex plan $120M to $150M → spent $209M
This isn't new either. In 2017 the Q3 guide of $107M to $115M got cut to $88M to $89M when Amazon pulled back. In 2018 a laser defect led to suspended shipments, a guidance cut and a securities suit (later dismissed). Same founder CEO throughout.
What bugs me more is in the 10Q:
> Receivables of $413.8M on a $270M quarter. That's ~140 days of sales unpaid
> $234.5M of that is owed by one CATV distributor (Digicomm), which was billed ~$257M in the first nine months. Mgmt says it gave them longer than typical payment terms
> Inventory ~$378M, roughly doubled since Dec
> Q3 GAAP profit came from $8.2M of government subsidies. The operating line was still a $1.4M loss
> ~68M shares a year ago → 90.8M now. Three ATMs this year, plus $125M of 2030 converts now convertible
To be fair, the bull case is real:
> Revenue +127% YoY, 6 straight record quarters
> Demand exceeds supply. A hyperscaler pulled 1.6T forward, and the CEO says there's ~$450M of 1.6T orders in hand
> Signed a multi year 1.6T deal with a Tier 1 hyperscaler. Mgmt sees $1.0B to $1.5B, with a min 10% share through May 2029
> In house InP lasers while the whole industry is laser constrained
> The CATV 1.8GHz cycle is printing: record $110.9M
> Credit where due: the last ATM raised ~$588M at ~$105/share. The stock is ~$48 now
> Q4 non-GAAP EPS guide is $0.48 to $0.63. Annualise the midpoint and it's a low 20s P/E
The caveat is that the deal has built in price cuts and no customer capital commitment. The "5x larger" second deal is in "final stages", which is what this team says about everything.
Demand isn't the question. Credibility is. A company that's capacity bound should be the easiest to forecast, and they still miss on nearly every timeline. Until they hit a number they set two quarters earlier, I'm haircutting everything they guide. Q4 is the test: $385M to $420M revenue, 29.5 to 31% GM, $50M to $70M of 1.6T.
Poke holes if you see them.
1000 美金开 10 倍,和 2000 美金开 5 倍,最大的区别是什么?
先看名义仓位:
1000 美金 × 10 倍 = 10000 美金
2000 美金 × 5 倍 = 10000 美金
两种方式的实际持仓规模都是 1 万美金。因此,在不考虑费用和资金费率的情况下,币价每上涨或下跌 1%,账面盈亏基本相同。
真正的区别,不在于赚得多少,而在于你能承受多大的反向波动。
因为两种方式占用的保证金不同,爆仓距离也不同。
以逐仓、无额外保证金为例,粗略来看:
10 倍做多,价格下跌约 10%,本金就可能接近亏完。
5 倍做多,价格下跌约 20%,本金才可能接近亏完。
做空则相反,价格上涨时面临同样的风险。
当然,实际爆仓价格还会受到维持保证金、手续费等因素影响,通常会比理论值更早触及爆仓线。
简单来说:同样是 1 万美金的仓位,1000 美金开 10 倍,资金缓冲更小;2000 美金开 5 倍,抗风险空间更大。
再换一个角度看资金分配。
假设你的账户总资金有 1 万美金:
1000 美金开 10 倍,理论上可以分成 10 份,每份对应 1 万美金仓位。
2000 美金开 5 倍,理论上可以分成 5 份,每份同样对应 1 万美金仓位。
前者单笔投入的保证金少,资金可以分得更细;后者单笔保证金更多,能够承受更大的反向波动。
但这里有一个关键问题:能分成多少笔,不等于你真正能承受多少次亏损。
交易不能只看方向对不对,更要看进场之后如何管理仓位。
价格朝有利方向走,想加仓可以,但要控制整体风险,避免仓位越滚越大;价格朝不利方向走,就要按照预设的止损和减仓计划执行,而不是不断加仓摊平,把小亏拖成大亏。
归根结底,仓位管理其实就两件事:
控制投入: 每一笔交易到底拿多少资金作为保证金?
控制亏损: 如果判断错误,这笔交易最多愿意亏多少钱?
所以,不要只问自己应该开 10 倍还是 5 倍。
真正应该先问的是:
如果这笔交易判断错了,我最多会亏多少?亏完之后,账户还剩多少资金?我还有没有下一次交易的机会?
记住,交易最重要的不是某一笔赚得有多快,而是当你连续判断失误时,依然有资金、有机会留在市场里。
杠杆决定仓位大小,风控决定你能在市场里走多远 #btc
emperinter retweeted
现在的美股大盘太难做了 像今天这种大利空该跌的时候不怎么跌 上周四那种毫无征兆没有利空的时候反而大跌 上涨也是无厘头 要么像昨天一样很突然的大涨 要么像前一段时间一样一直不涨 都不知道该如何操作了 🤭
emperinter retweeted
通胀又炸了,老黄的麻烦也越来越大。
美股科技股好不容易反弹了两天,今晚又被狠狠砸下去了。
这次主要是两件事撞在了一起:美国通胀超预期,还有就是英伟达老黄那档子事。
先说通胀。
美国刚公布的10月CPI,环比上涨0.4%,核心CPI上涨0.3%,双双超过市场预期。
美国2年期国债收益率一度飙升21个基点,市场对12月加息的预期直接升到了接近99%。
更麻烦的是,美联储理事Waller甚至开始讨论,12月是否应该一次加息50个基点。
利率突然往上冲,对于本来估值就不低的科技股,显然不是好消息。
偏偏在这个节骨眼上,英伟达又出事了。
真正麻烦的是,这件事现在已经不只是私人生活问题,还开始影响到老黄在英伟达的领导地位。
大家可能不知道,英伟达的管理结构非常特殊。
黄仁勋从1993年创办公司开始,一直担任CEO。英伟达有超过60名高管直接向他汇报,大多数还是技术负责人。
从芯片研发、产品路线,到大客户关系、政府沟通,他都深度参与。
也就是说,英伟达虽然已经是全球最大的公司之一了,但仍然保持着非常鲜明的创始人管理模式。
更棘手的是,公司至今没有公开明确的CEO接班人。
而且,黄仁勋虽然是创始人,但持股只有约3.6%,并没有绝对控制权。
甚至,英伟达11人的董事会甚至没有董事长,几位独立董事早在上世纪90年代就已经加入。
现在调查黄仁勋的,恰恰就是这批与他共事多年的董事。
董事会虽然公开表示仍然信任黄仁勋,但市场已经开始质疑,这场调查能否真正保持独立。
而且事情还在继续发酵。
美国国会的共和党议员已经要求黄仁勋11月18日参加闭门说明,解释英伟达如何防范外国情报活动。还有股东起诉英伟达,指控公司此前的信息披露误导投资者。
如果最终只是一场私人生活丑闻,老黄大概不会失去CEO职位。但如果调查进一步牵扯到出口管制、商业机密或信息披露问题,性质就完全不同了。
对于英伟达这样的公司,失去黄仁勋,意味着整个公司的决策体系、客户关系,以及战略方向,都可能面临一次非常困难的调整。
这恐怕才是英伟达这两天被市场集中抛售时,投资者真正担心的另一层风险。
$NVDA
emperinter retweeted
0.3% Core CPI opens the door for Fed's Waller to back a 50 bps rate hike at the December FOMC, which would be the largest single-meeting hike since December 2022.
emperinter retweeted
$AAOI
AAOI beat Q3 estimates, but its Q4 revenue guide tops out below the Street's $427M forecast. The year-end capacity outlook also moved from roughly 650,000 units a month to more than 500,000 as expansion runs into power and construction delays.
AAOI Q3 2026: A beat, with a slower capacity ramp
1. Q3 delivered growth, with a margin shortfall
Revenue was $269.9M, up 40.6% sequentially and 127.5% year over year. That beat the $268.1M consensus by 0.7% and landed inside management's $255–290M guide.
Adjusted EPS of $0.17 beat the $0.14 consensus by 3 cents, within the $0.11–0.26 guide. It used 87.8M diluted shares.
Adjusted gross margin was 28.7%, below the 29–30.5% guide and down from 29.8% in Q2. GAAP gross margin improved to 28.2% from 27.7%. Those two measures tell different stories and should stay separate.
GAAP net income turned positive at $6.3M, or $0.07 per diluted share. Operating income remained negative at $1.4M; $10.1M of net other income helped bridge the gap. A profitable bottom line has arrived before GAAP operating profitability.
2. Q4 falls short on revenue; EPS is near consensus
Q4 revenue guidance is $385–420M. The $402.5M midpoint is 5.8% below the $427.3M consensus, though it still implies roughly 49% sequential growth.
Adjusted EPS guidance is $0.48–0.63, using about 96M shares. Its $0.555 midpoint is essentially in line with the $0.56 consensus. Adjusted gross margin is guided to 29.5–31%, a recovery from Q3.
The outlook therefore points to less revenue than analysts expected, with earnings holding closer to expectations. Margin recovery now carries more weight in making the profit forecast work.
3. Answers to our preview
Q1 — Does Q4 support consensus? Answered: revenue falls short across the range; adjusted EPS broadly brackets consensus.
Q2 — Where is the growth? Partly answered. Datacenter revenue reached $155.6M, up 44.5% sequentially; CATV reached $110.9M, up 37.6%. The release does not split 800G and 1.6T revenue. It discloses a multi-year strategic agreement with a Tier 1 hyperscale customer, without naming the customer or quantifying its economics.
Q3 — Is the capacity target intact? Partly answered, with a lower year-end outlook. Installed equipment capacity is approaching 330,000 units a month. Management now expects equipment sufficient for more than 500,000 monthly 800G/1.6T units by year-end, versus roughly 650,000 previously. It expects more than 1M by end-2027, while warning that US expansion will take longer because of electrical-power access and construction issues. Installed equipment is not the same as qualified, shipped output.
Q4 — Are profit, cash and dilution improving? Partly answered. Adjusted net income rose to $14.8M from $5.5M in Q2, despite lower adjusted gross margin. September-end cash, cash equivalents and restricted cash totaled $621.5M. The release does not provide a cash-flow statement, so it cannot establish operating cash generation or quarterly capex. The Q4 share assumption is about 9% above Q3's diluted weighted average.
4. What the call must resolve
C1. The revised buildout timetable, power constraints and qualified output behind installed capacity.
C2. 800G/1.6T shipment mix, customer acceptance and the new hyperscale agreement's commitments.
C3. The margin recovery supporting Q4 earnings, operating cash flow, capex and the share-count bridge.
Our ramp thesis is weaker on timing. Demand has additional support from the hyperscale agreement and datacenter growth, but the revenue outlook and expansion delay reduce confidence in the delivery schedule. Q4 margin recovery and qualified output are the next tests.
emperinter retweeted
美股这次暴跌,可能是AI从造梦到还债的开始。
今晚美股这波暴跌,纳指盘中一度跌了快6%,很多个股全是两位数起崩,真的有种天塌了的感觉。
事情的导火索,毫无疑问就是美国中期选举之后,数据中心建设遭遇的政治阻力突然被市场放大。
以前大家觉得,只要微软、谷歌、亚马逊这些巨头愿意掏钱,美国就能不断盖机房、买GPU、建电站。
结果现在,美国越来越多地方的居民不干了。
凭什么科技巨头赚AI的钱,普通人却要承担电费上涨、缺水、噪声这些代价?
美国至少38个地方辖区的选民支持了限制或禁止数据中心的措施。瑞银甚至估算,美国原计划2028年新增的数据中心容量中,约30%可能受到潜在建设限制的影响。
这件事可怕在哪里?
可怕在它可能引发一连串连锁反应。
机房审批慢了,项目就得延期;项目延期,GPU、服务器、电力设备的采购就可能往后推;订单兑现晚了,原本靠这些订单支撑的增长预期和估值,自然就得重新计算。
但更麻烦的还在后面。
很多AI公司根本不是拿自己的钱建机房,而是借钱买GPU,再把算力租出去赚钱。
像CoreWeave这样的公司,前期投资极其庞大,得不断融资才能维持扩张。
可问题是,项目可以延期,银行的利息不会延期。
最近CoreWeave的信用风险指标已经冲到很高的位置,说明债券市场越来越担心它的偿债能力。甲骨文这周宣布计划再融资约200亿美元,股价也遭到重挫。
这就形成了一个危险的循环:
项目建设受阻,未来收入可能推迟;收入推迟,市场开始担心偿债能力;融资变贵,继续扩张就更困难;扩张放慢,又反过来削弱未来的增长预期。
说白了,AI公司最怕的可能不是没有需求,而是钱先烧完了,需求却还没来得及兑现。
偏偏这个时候,市场又接连收到几个不太妙的信号。
微软决定不再扩大一项潜在价值数百亿美元的数据中心合作。虽然不是今晚临时取消的,也不影响原有合同,但说明巨头已经开始筛选扩张项目。
Datadog最新财报又透露,它最大的AI客户正在减少云监控服务使用量。
两件事都市场更加担心,科技公司是不是开始控制支出了?
再看AMD,刚公布的季度收入同比增长45%,数据中心收入更是增长97%,业绩相当漂亮。
可今晚股价照样一度暴跌近16%。
这就很说明问题了。
而且,这场抛售早有征兆。
10月份,纳斯达克已经跌了5.3%,但标普500等权指数竟然上涨了5%。
说明资金早就在撤离科技巨头,只是此前没有集中爆发。
现在,政策限制、融资压力、支出放缓的消息一起出现,原本就拥挤的AI交易开始加速平仓。VIX恐慌指数盘中暴涨超过50%,进一步反映出市场恐慌情绪的急剧升温。
有意思的是,今晚美国十年期国债收益率反而从4.90%左右降到了4.84%。
所以,这次还真不能简单怪到利率上涨头上。
更像是投资者对科技股增长前景失去信心,开始集中降低风险。
如果接下来真的出现融资困难、项目停工、订单取消,那恐怕就不只是股价调整,而是整个AI投资周期开始发生变化了。
我们真的都低估了美国中期选举的威力。
emperinter retweeted
应用光电AAOI三季度财报大概率会超。
这家伙在8 月下旬开始在 105 美元左右卖新股,到 10 月初一共卖了 569 万股,净拿到 5.88 亿美元,11 月 4 日收盘只剩 59.27,最近一周就跌了 28%。
美东今天盘后它出三季报,这季的利润我看问题不大,难的是第二天。
公司自己给的指引是调整后每股 0.11 到 0.26,华尔街估 0.14,离指引下沿只差 3 分,上季它还超过了自己给的上限,要担心的是缺货。
一个客户买不够 100G 交换机,三季度 100G 收入要少 2000 万到 2500 万美元,800G 也卡在 DSP 芯片上。
第二天才难猜。
最近 9 次财报 6 次高开,其中 4 次收盘比开盘低了 4.8% 到 17%,今年 5 月开盘涨了 10.3%,收盘反倒跌了 5.5%。也有 2 次一天涨了 55% 以上。
这次期权算出来是涨跌 16.5% 左右。
所以周五就算高开,我也先看它守不守得住开盘价。
公司 8 月说今年营收 11 亿美元左右,倒推四季度要做到 4.7 亿到 5 亿,四季度指引给得比这低,高开也难守,电话会在北京时间 11 月 6 日早上 5 点半,我先听 800G 卖了多少和四季度指引,周五收盘再回来对这条。
我自己的判断,仅作参考,高开低走那张图,我放我欧易星球的长文里了。
oyidl.co/ul/m8uXR1g
emperinter retweeted
Expected numbers for $AAOI (Applied Optoelectronics) earnings today after close:
emperinter retweeted
LITE的业绩确实出来了,但有一点得先跟各位说清楚:这不是一份按原定日程发布的正式财报,而是一份提前放出来的业绩新闻稿。
LITE原定11月5日盘后发FY2027 Q1财报,因为今天同日官宣被Marvell收购,特意提前到了11月2日盘前公布。完整的10-Q文件通常会在新闻稿之后一两天才正式提交SEC。所以更准确的说法是:业绩数字是官方的,但披露节奏是被收购事件催着走的。
数字本身挑不出毛病。Q1营收12.63亿,同比增136.5%,略超预期;非GAAP EPS 4.57美元,预期4.2左右;毛利率53.2%,营业利润率42.3%。Q2指引营收15.75亿到16.25亿,比市场共识的14.8亿高出一大截,EPS指引5.90到6.20,也远超预期。CEO还确认拿下了第二个NPO/CPO大单,采购承诺超过50亿美元,OCS的需求能见度在快速扩大。
但盘中LITE和MRVL都在跌。LITE盘中跌幅差不多10%。
原因不在经营层面,在换股结构上。Marvell这笔1150亿美元的全股票收购,LITE股东每1股换4.5股MRVL,MRVL因为大规模增发承压,LITE作为被换股的一方,股价和MRVL绑死了,MRVL一软,LITE就跟著跌。盘前的抛压,定价的是交易结构带来的筹码压力和股权稀释担忧,不是LITE自己出了什么问题。
一句话:业绩是硬的,提前发布是为了配合收购官宣;目前的跌,是换股交易自带的副作用。
$LITE
免责声明:本文纯属个人观点,不构成任何投资建议。
emperinter retweeted
$AAOI has now given up all of this week's gains. Black Friday in the markets, my boys.
emperinter retweeted
Replying to @aleabitoreddit
Max pain right now with $AAOI.
Just holding until we get catalysts next year.
emperinter retweeted
Replying to @aleabitoreddit
$AAOI to me seems mismanaged, they announced that offering late on a Friday night…

































