ChrisTech @ChrizTechInvest

@ShanghaoJin Dont think it will collapse. AI is going to dominate the world

One of the most interesting parts of $LITE’s latest earnings call was not the growth rate itself, but the commentary around Scale Across. $CIEN Ciena -- coherent DSP, NPO/CPO, FST, Scale Across, DCI $CSCO Cisco / Acacia -- coherent pluggables, 800G DSP, coherent transmission $NOK Nokia -- 1.6T coherent, Coherent-Lite, InP/SiPh PIC, Scale Across transport $MRVL Marvell -- 1.6T coherent DSP, 2nm DSP, optical connectivity silicon $AVGO Broadcom -- 400G/lane DD-EML, InP laser, DSP, CPO, switch ASIC $COHR Coherent--InP EML/PD, 400G/lane optics, SiPh, coherent components $TSEM Tower Semiconductor -- silicon photonics foundry, PH18 SiPho, 400G SiPh MZM 6503.T Mitsubishi Electric -- high-speed InP EML, 400G/lane device technology Lumentum said that, for one hyperscaler, the network capacity required to connect just two AI data centers is roughly twice the capacity of the entire global backbone it built over the previous decade. And Lumentum explicitly described this as Scale Across. That immediately reminded me of the six OFC 2026 papers I analyzed back in April. The direction I saw then was fairly simple. 400G/lane IM-DD is becoming technically feasible at the component level, but the real challenges are moving toward link budget and power. On the coherent side, 1.6T pluggables are running into a similar wall as lasers, drivers, DACs and DSPs all have to fit within a roughly 35–40W power envelope. At the same time, NPO and CPO pull client optics closer to the switch ASIC. As that happens, coherent line optics can increasingly move back outside the router and into the transport layer. Coherent-Lite and full-band transponders are particularly interesting in Scale Across, where the network expands from building-to-building to campus-to-campus and ultimately data-center-to-data-center. Back in April, much of this was still visible mainly through conference papers, architecture proposals and roadmaps. Now earnings calls are beginning to show the actual hyperscaler demand behind that architecture. AI networking is no longer just about connecting GPUs or racks. The data centers themselves are becoming the nodes. I covered this transition in detail in my OFC 2026 Part 4 review, connecting 400G/lane IM-DD, 1.6T coherent, CPO/NPO, Coherent-Lite and full-spectrum transport. photoncap.net/p/ofc-2026-par…

keyboard_arrow_left Previous keyboard_arrow_right Next

Photon Capital @PhotonCap

4 months ago

[OFC 2026] Part 4 of 5: 400G/Lane and 1.6T Coherent — Two Fronts, One Power Wall Six papers from $AVGO, $CSCO, $COHR, $CIEN, $NOK. Zero measured link demos. Broadcom's 99 GHz DD EML? Open eye only — no BER, no TDECQ. Coherent Corp's 400G link budget? The 5 dB number everyone

3 6 28 35K 11

@ShanghaoJin But 800V DC is being introduced in Q3 2026 by Delta, the price is not going to catch up for power semi to make profit?

@ShanghaoJin They eventually realised……

[Exclusive] Singapore's Temasek to Buy Into Samsung, SK Hynix… Its First Ever Investment in Korean Stocks Temasek, one of Asia's largest sovereign wealth funds, is making a new investment in Samsung Electronics and SK Hynix. This marks the first time Temasek has invested in the Korean stock market. (..) Temasek is said to hold the view that within the artificial intelligence (AI) value chain, memory semiconductors are the most undervalued area. An investment banking (IB) industry source said, "Temasek has an investment style of deploying large amounts of capital into industries that lead the market, from a long term perspective," adding, "Samsungnix has risen as much as more than 880% from last year's low, but this shows conviction that it will go higher still."

@cherryPayment 这个梗是因为Dr Lin在几块钱的时候一直跟我们说： 会超过中际旭创（注意不是成为第二） 当然他现在想超过的应该是Lite了

为什么要买增长公司：PE高不代表贵；PE低不代表便宜 当你毛利率到顶，且无空间继续之后，能讲的故事就只有三部曲： 毛利率会不会降、怎么降，什么时候降 PE低 vs 毛利率面前就是渣渣，要不怎么说Hock Tan厉害呢。人家一早就看穿了

Torres9 @jiangchenglu_

2 weeks ago

@ShanghaoJin @BigbirdflyChan 毛利即使到顶，还有给pe的故事吧

1 0 1 55K 0

@ShanghaoJin @bboczeng 應該拿SOX 去比吧。今年的槓桿不在Nasdaq

@cherryPayment 業績發佈日，必定會放量啊。雖然漲很多感覺有勢

告诉大家一个不幸的消息 我不知道高盛说momentum factor说“deleveraging nearing its conclusion”结论怎么来的 我单纯看GS的数据似乎并得不出这个结论，暂时谨慎采纳他们看法吧 因为没爆仓去杠杆不够彻底，需要等平空之后观察市场卖盘情况才知道

一群野人在头领的带领下去猎杀野牛。拼尽全力做空猎杀，准备分尸的时候，高盛把整个牛都让头领Citadel吃独食了 所有人都精疲力竭，还握着空单，没有肉吃 = 空头平不出来了 Boss快打完的时候，爆变成自己满血的队友。高盛和Citadel从队友瞬间切换成对手盘了，这时夺路还逃不出来 最近市场跟行业无关

Herman Jin @ShanghaoJin

2 weeks ago

Situational Awareness还是太年轻 当成为几百亿带杠杆的那个蓝鲸，全世界鲨鱼都会围过来的 很多杠杆赌徒，都没想过市场是会被扭曲来专门绞杀你的

23 13 301 193K 88

@ShanghaoJin 問題是兩個月後，要還有錢進場…

@ShanghaoJin @TiezhuCrypto 2027的CAPEX 一定高吧，畢竟內存太貴了。往後scenario 應該有二，1. Hyperscaler 的CAPEX 慢下來，其他公司/政府補上，2. Hyperscaler 的CAPEX 維持不變/growth 減少，其他公司/政府瘋狂建設。 第一個情況的話會有一陣大跌，但半導體的生意應該還是不受影響直到2030左右。

Let see how far can we go from here.

@ShanghaoJin @jimmyhuli 所以其實是去槓桿，讓業績說話，把價格打到認為閉上眼買都可以的程度，才買對吧？我本來認為，股價應該要推升，才更容易融更多的資金來支付這筆帳，讓韓國的冤大頭推高一點，但他們卻把他們幹掉了。會不會是Wall St 本來持股不夠，要推倒股價自己進場？

@PandaCakeSaber @ShanghaoJin @jimmyhuli 資本支出本來就是不可能一兩年收回的吧？只是因為以前軟件行業以前不是資本密集的公司，captial intensive 的公司像半導體，一向都是以起碼數年作為單位來投資CAPEX 的啊，還要產能爬坡到理想的yield 才能收回。

@PandaCakeSaber @ShanghaoJin @jimmyhuli 不對啊， 據我理解，Anthropic 的盈利應該是從Fortune 500 那些大企業，政府，等等而來的。你說的左腳踩右腳的是Anthropic 的CAPEX 和 Nvidia 的銷售吧？如果只有這個閉環還說得通。事實是金融業自己，像GS，Citadel， 還有SaaS 也因為AI 而增長加速了。

@ShanghaoJin @jimmyhuli 當然現在市場的確如你所推演的去運行，我也從難以置信到現在強迫相信了，因為不相信，回報也回撤了很多，所以我也想知道更多你的看法。

@ShanghaoJin @jimmyhuli 其實我不明白帳為何算不過來。之前說大模型公司在蝕本做，現在Anthropic 都能盈利了，連最多現金，而且不碰科技股的巴郡都投GOOG 了。GOOG，AMZN 等公司發債都能籌到天量資金，Covid QE 發的錢從美債轉到AI 標的，應該已經足夠了吧？TSMC 都沒有擔心過客戶付不起錢，去增加CAPEX，Wallst 操心什麼？

SOX dipped to 11867 today, which is a critical low from 06/09 and 07/07. I will give it 1-2 days to reclaim the 50d EMA. Otherwise, I have to stop out any non-Mag7 positions. My thoughts are that semis should remain in a consolidation range instead of any directional breakout before the Mag7 ER, as bulls and bears are debating AI capex and ROI heavily with some balance. Any breakout of the range before the Mag7 ER means the market has reached some weird consensus for reasons we don’t know, and we should be very cautious.

3X Long Labubu @labubu_trader

a month ago

@vix8b No cut, I still hold lots of cash. If there is a clear break down of EMA50 of SOX, I’d have to stop loss and wait for lower level to enter.

2 0 10 62K 3