AI时代最重要也最神秘的规则：波兰尼悖论
“我们知道的，远比我们能说出来的多。”
We can know more than we can tell.
这是哲学家迈克尔·波兰尼的一个理论，被各路AI大牛反复引用。
AI现在迭代速度极快，它们可以懂得所有人类书本上艰深知识，瞬间破解各种数学难题，完成各种艺术创造，但有一种能力它们现在没有，将来也不会有：默会能力。
“默会能力”（Tacit）——它溶解在肌肉记忆、直觉和无数次试错的生命体验中。它抗拒被提取，拒绝被编码，更无法被标准化地写进教材，更无法被AI抓取学习。
AI的本质，是吞噬并计算一切“可言说、可编码”的显性知识。当所有的规则、数据、理论和套路都被大模型瞬间掌握，甚至以超越人类百万倍的速度输出时，人类的优势在哪里？就在这些无法被编码，“只能意会不能言传”审美与品味中。
AI可以写出符合所有修辞法则的诗句，却无法体会雨夜街头独自徘徊的落寞；
AI能穷尽商业模型的所有推演，却无法拥有在混沌局势中力排众议拍板的魄力。
在AI时代，所有能被清晰表达、步骤化和标准化的技能，都将廉价；而真正昂贵的，只有审美、品味与直觉。
你看了一万本文学作品，你比不过AI，但是你对文学的审美与口味是独一份。
波兰尼悖论在AI时代给我们的最大启示是：不要试图在“计算”上战胜机器，而要去拥抱那些无法被计算的“默会”。在这个算法试图将一切量化、编码的时代，锻炼自己那无法言说的品味、直觉与真实的生命体验，这是我们生而为人最高贵特权和最珍贵的能力。
提高审美、提升口味、强化直觉。
That’s a misconception:
$SIVE is the laser supplier for next gen architectures, not just CPO scale up.
Pluggable, scale out CPO, scale up CPO, NPO, etc.
- Sivers and $JBL went god-mode and developed 1.6T optical transceivers with CW lasers.
Effectively designing around the EML bottlenecks, and even made things more power efficient.
From quotes from Jabil management they created a “relatively dramatic moat”.
So for the next gen of 1.6T pluggable transceivers, Sivers seems immediately used.
Markets missed this nuance too: after Jabil’s announcement, other pluggable players reached out, and SIVE is working with them now (prob codevelopment, qualification stage).
There’s not too many players that build optical transceivers that aren’t vertically integrated like Lumentum (think: Innolight/eoptolink, and maybe others).
So these are active developments, just not public material yet and could be a new press release anytime.
- for CPO scale out which happens H2 2026 onward like $POET, $SIVE is a laser supplier for those players.
- scale up CPO is h2 2027, from Ayar and Nvidia’a NVLink CPO ecosystem and that’s the main volume ramp across optical players.
But this is where $SIVE looks like they mog every other player from supply chain mapping.
Since they’ve likely been working + designed in with $MRVL Celestial, Lightmatter, Lightelligence and the others since they were small way back when.
On top of things like this:
For foundries like $GFS, $SIVE is the reference laser.
Not even including massive companies like O-Net building ELS with $SIVE, likely for the Asian hyperscalers supply chains.
This is why $SIVE is by far my favorite laser chokepoint long.
Feels like they’re everywhere in new optical architectures starting beginning of 2027.
I've been getting a lot of questions about OE Solutions (138080) recently.
Here's my research/thoughts so far on it:
They're a small Korean optical transceiver company, similar to $AAOI.
And they've become one the few EML players in the world (eg. $COHR, $LITE, Mitsubishi, Source, Sumitomo), with scarce 100G EML laser capacity for 800G/1.6T.
OE appears to be trying to make the full transceiver, not just the EML. They also have finished ELSFP CPO products, with UHP CW lasers, which is sampling Q3.
Likewise, OE also seems to be building out the full ELSFP, rather than selling CW laser dies, so that's more market share.
So you can think of it as Korean AAOI but EML instead of CW, and less capacity/qualifications. And playing catch-up to the rest of the world.
However in terms of timelines:
1. 23dBm cooled ELSFP samples start in Q3 2026
2. Sales base for 800G, and 2027 1.6T "full force"
3. Late H2 2027, H1 2028 onward probably their ELSFP enters the volume production.
And it seems they're working on 200G EML capability from their investor snipper 2025 OE IR snippet referenced “100GBaud EML / 200G PAM". This seems promising given their IP/demos.
ELS has just been unveiled recently, sampling starts Q3. 800G/1.6T Optical transceivers are also likely 2027.
So this is basically Korea's sovereign photonics player, playing catchup to $LITE, $AAOI, and the bigger players.
(Disclosure: I have positions in OE Solutions (138080). This is fundamental research for informational purposes, not financial advice).
As for OE solution valuations:
1. Doesn't seem like there's confirmed customers yet for these growth verticals.
2. Probably not many people understood what they're building toward yet.
3. Yields data kinda uncertain
1. Customers: We're actually in a major EML/CW laser shortage, so any independent capacity will be sought after. I personally don't think they'll have a hard time finding customers here. I would assume anything they make might get bought out and would get extra support from Korea.
2. Institutional support: Probably not much since it doesn't meet threshold for many US institution given MC size + KR listing. I also don't think many people understood what they're building yet.
3. Yields/Capacity: OE's disclosed wafer/module utilization is low (i remember was around 31% underutilized off the top of my head), so there's enough material revenue they can generate before they need to spend on capex.
I'm also not sure about EML/CW and other yields. Probably need to go ask the company.
Is this some random crap co? No. It's been doing optical transceivers stuff for more than 20Y, has R&D in the U.S. and Netherlands.
But their entire AI growth vertical seems to happen next year, and hinges mainly around capacity/yields.
And I personally think the EML/CW tech is probably worth a lot more than their current MC, if it were an acquisition target.
Especially from a larger player that wanted to vertically integrate EML for pluggables and CW lasers for CPO.
Markets are probably waiting on more certainty around qualifications after Q3 sampling or earnings projections announcements.
TLDR on thoughts:
Some of my friends discussed this last year, was probably way too early. Saw it got many comments 2 months ago, still too early.
I still think now is early, but later in Q3-Q4 might be more interesting. I personally think it's a lot higher risk than a major CPO player like $SIVE, that's embedded in Ayar, $JBL, $GFS, and many other hyperscaler suppliers. Which also has Win Semi and others de-risking volume ramp.
As OE Solutions looks like a new player trying to build out an $AAOI for the optical transceivers but sovereign EML and CW laser production for CPO products.
And there's a lot of answered questions around customers + volume ramp, which presents material risk.
But if you believe Korea can build out an 800G/1.6T transceiver EML supply chain and ELSFP with CW lasers.
With OE Solutions, it might be worth taking a look into. Still researching the company tho, just initial thoughts.
Apparently gym bros are the new hyperscalers.
They’ve created a new bottleneck with whey protein?
Now there’s $SNDK style price hikes.
Maybe creatine is next with shortages. https://t.co/YY5waMXzKH
Well, looks like prediction markets are about to get a ton of extra regulatory scrutiny.
They should also investigate how Polymarket pays influencers for undisclosed paid comments to game the X algorithm.
I get offered deals like this all the time as the most subscribed to person on X.
But as you know, I don’t do paid promotions since I want my research to be without outside influence.
无限上涨派 Tom Lee 反水了，他认为，现在是暴跌前的狂欢，我们现在做空还是太早了！
真正的修正阵痛期会出现在下半年，四大导火索：
1⃣ 沃什目前正在对美联储进行改革，市场对当前的美联储沟通风格抱有极大的不确定性，极可能在2026 年下半年，尤其是中期选举前1-2个月来发动猛攻测试美联储。
2⃣ Space X 目前市场浮动筹码非常少（仅约900 亿美元），但下半年可能迎来分阶段解锁；再加上Anthropic、OpenAI 两大明星 IPO，市场已经在定价万亿级别的二级市场，可能会吸干流动性。
3⃣ 霍尔木兹海峡的冲突干扰，会造成全球供应链短缺，断裂效应在下半年将会显现出来。
4⃣ 最关键的一点，随着融资余额达到高点，保证金债务无法持续，市场投机的子弹会慢慢打光，将成为引爆卖压的最后一根稻草。
Trump administration is prob realizing now that the key to winning Trade Wars:
Is through frontier supply chains like Quantum, AI, Robotics.
Not cheap Nike sock exports.
With $ASML, $TOWA, $LPK, Ajinomoto,
over in Japan, EU, TW, Korea, and others holding all the cards…
Since each country holds different monopolies and chokepoints needed to make each industry work.
Which all happens to be OUTSIDE the US.
Tariffing all our partners turned out to be not a great idea.
But not too late imo to repair relations and weaponize global supply chains through partners if Trump wants better trade deals.
AggrNews: FAMOUS MEV BOT "JAREDFROMSUBWAY" APPEARS TO HAVE BEEN EXPLOITED FOR OVER $15M: ONCHAIN https://t.co/TKeyafq2Ru
AggrNews: 著名MEV机器人“JAREDFROMSUBWAY”似乎已遭受攻击，损失超1500万美元：链上数据
