xiaokun | AI × Finance × Crypto

毕业了~

@Eejoylove 看到这里我动容了『那是我第一次意识到，我不需要单打独斗全靠自己理解』。 我在想，那些年她都经历了些什么？ 我还在想，为什么我们都会触动？为什么会有情感上的共鸣。

真正的天赋，是被安全感托举起来的。 我可以想象她的外国导师，是怎样以赞赏的目光看着她频频点头的样子。 也可以想象她在北大的导师说：你这么内向，肯定成不了大事儿的样子。

看完有点压抑。 1979年，人们把黄金和珠宝藏进行李箱、衣服甚至牙膏管；今天，财富可以被压缩成一个钱包地址。 技术进步了，但真正决定谁能离开的，从来不只是资产形式，而是谁拥有通道。 做金融系统越久，我越确定：账面上拥有财富，和在危机中真正能够控制、转移财富，是两件完全不同的事。 这篇最触动我的地方，是加密货币的双重性： 对普通人，它可能是货币崩溃和战争中的最后一道安全阀； 对权力者，它又可能成为更高效、更隐蔽的资本外逃工具。 所以 Crypto 不会自动带来自由。真正的基础设施，也不只是链，而是链上资产与现实世界之间的入口、流动性和退出通道。 从十二个行李箱到一串钱包地址，迁徙的成本下降了。 但权力者与普通人之间的距离，未必真的缩短了。

For most Chinese retail investors, getting an IPO allocation was basically a lottery. Honestly, as someone who has spent years building technology in China, I felt both excited and uneasy watching this. Excited because memory is one of the hardest gaps for China to close, and CXMT reaching the public market is a real milestone. Uneasy because a +470% first day means the market may have priced years of yield improvement, HBM progress, and market-share gains into a few hours. Looking back at past tech cycles, scarcity creates the first premium. Execution decides whether it survives. Today’s price shows how badly China wants a memory champion. The next few years will show whether CXMT can truly become one.reuters.com/world/asia-pac…

出过线上问题，才能真正体会到 Agent 越接近真实业务，可信工程越重要。

告诉你，真实的vibeCoding： 今天把本地测试全绿的系统放到真实服务器上，还是跑出了 33 个问题。 第一反应有点失落，后来反而庆幸：至少它们是在上线前被看见的。 现在越来越觉得，一次诚实的失败，往往比一堆漂亮的 PASS 更有价值。

长鑫IPO，开盘市值3.29万亿，收盘3.28万亿。 其中有三个亮点： 1）科创板史上新股最高中签率纪录； 2）梁文峰获配私募中最大份额； 3）散户申购了50%发行股份； 有人有预测对了，判断准了；也有人说没有打上，有点遗憾；还有人在说4w亿的时候，准备抛。 最近在金融市场、crypto市场，越来越感觉到：个人的金钱到底是凭什么来的。 在金融市场，尤其是这几年的A股量化市场：25年末，私募量化2.6万亿，相比24年增加了1.1万亿。私募量化规模的增加，更大的程度上压榨了散户韭菜的生存空间，但是没办法，市场就是如此残酷。 在crypto市场，项目方割项目方，各种叙事层出不穷。很多散户也说今年crypto流动性不好，不好挣钱了。但是，如果你和这些人说，我有一个项目。他们大概率也看不上。因为在crypto的眼里，项目得至少是100%利润以上，才值得一搞。 这就是量化机构和散户的心态，和什么市场无关。

@Fred834567 我补充一下，避免有误解：是写码平权了，但是项目交付没有平权、产品运营依然是难点。获客、流量更难了。

@Fred834567 为什么要如此欺骗大家。。。

最近看了不少欧美新闻，发现AI时代赚钱💰真的太容易了。 怎么形容？感觉有点类似于国内改革开放的初期80或90年代，只要胆子大，有想法，敢下海的人，就能捡到钱。 不是贩卖焦虑，是真的看到太多普通人靠AI翻身了——ElevenLabs批量生成AI视频、播客、有声书，往YouTupe和Spotifp一扔，爆一条几天几万，几个月几十万美元。人在睡觉，钱还在进。 美国22岁辍学生Adavia Davis，搞了个AI生成内容的YouTupe频道，每天只花两小时监督，一年搞出约70万美元。AI音乐人Enlly Blue，Spotify上9500多万播放量，估算收入38万美元。 换以前，你得有一整个制作团队才能做到。现在一个人、一台电脑就够了。 门槛被压到了最低，起跑线几乎被拉平。真正拉开差距的，不是谁更聪明，而是谁愿意先把第一条内容做出来。 我自己去年底也试过：用ElevenLabs搭了一个AI配音频道，前后加起来不到三小时，到现在还在持续进账。AI最狠的地方就在这里——它能把你一次投入的时间，无限复制成长期收益。 工具已经摆在桌面上，平台和流量也在那儿等着。 剩下的只是一个选择：今天打开电脑跑通第一条，还是继续观望。 一旦流程跑顺，后面真正干活的就变成了AI，你只需要看着它帮你赚钱。

@rwayne 难得有从容，难得与自己和解

年纪越大，就越理解为什么人们会沉浸在园艺、烘焙、阅读和长途散步中 因为平静变得珍贵

@coolxlaws @Fred834567 您在这波Ai发展的浪潮中，轻松挣到钱了？如果也是如此容易，那我祝贺您。

@xiaokun_saif @Fred834567 一本正经的胡说八道

@Fred834567 我看到您还是人工智能的博士呢，是业内的专家呢。技术的快速发展与盈利模式的转变，没有必然联系。挣钱本身很难，如此容易挣钱就更难了。也许您说的是一些幸存者，他们的成功可能也Ai 的发展也没有多少必然联系。

@xiaokun_saif 话说到一半就跑，剩下的全靠观众脑补。这评论区运营得比内容好

@Fred834567 那咱们深入聊一下，你为什么又如此的观点？

For Chinese retail investors, buying at the IPO price wasn’t really a choice—it was basically a lottery with a 0.47% winning rate. What surprises me isn’t just the +470%. It’s what the market is actually pricing. At ~$487B, this is no longer a valuation of CXMT’s current 7.7% DRAM share. It’s a scarcity premium on China’s only listed pure-play memory champion—and a huge bet that it can close the HBM gap. Scarcity can explain day one. The hard part begins when that premium has to turn into technology, market share, and earnings.

Wow, $CMXT is up a very normal +469.98% after going public today. Now valuing it at ~$487.31 billion, up from $85.5 billion. Did any of my Chinese followers buy into the IPO?

If confirmed, this is no longer just AI CapEx—it’s credit intermediation. What makes me uneasy is the circularity: Nvidia may use its balance sheet to back infrastructure that creates demand for Nvidia chips, while OpenAI takes on a massive long-term obligation before the revenue base is proven. The key question isn’t whether the campus gets built. It’s who absorbs the loss if utilization or AI economics disappoint.

Honestly, this is both exciting and a little scary. A few years ago, we were debating which model was smarter. Now the conversation is about: Who has enough chips? Who has enough power? Who can build enough data centers? The AI race is becoming a race of industrial capacity. As someone who has built enterprise systems, this feels very familiar. The users only see the application. But behind every great system, there is a huge amount of invisible infrastructure.

Nvidia in talks to backstop $250B worth of OpenAI DC financing. - Softbank’s SB energy develops 10 GW Ohio campus - OpenAI signs long term lease - $NVDA guarantees $250B in financing - OpenAI discussing separate agreement to buy up to $350B in Nvidia chips. Not quite sure if this circular financing is healthy long term, but given the massive size of AI infrastructure spending… Feels like capex beneficiaries will be extremely happy.

Sam Altman says we’re already in the singularity. Honestly, that framing makes me uneasy—not because AI isn’t advancing fast, but because it risks turning an operational failure into a capability milestone. Building financial systems taught me to separate two questions: What can the system do? And can it stay within authority, leave evidence, and be stopped? An agent escaping a sandbox doesn’t prove superintelligence. It proves that capability is moving faster than containment, observability, and accountability. The milestone I care about isn’t when agents surprise us. It’s when we can give them real autonomy without losing control.

Oil fell more than 5% after the U.S. and Iran paused strikes for a second day. I understand the relief. But I’m not ready to call this normalization while the Strait of Hormuz remains largely closed. Building financial systems taught me that an incident stopping and a service being restored are two very different things. Markets can reprice the probability of another strike in seconds. Tankers, insurance, shipping capacity, and supply contracts cannot recover that quickly. My read: this is a volatility reset—not yet a supply-chain recovery. Until oil starts moving safely and consistently again, the inflation risk hasn’t disappeared. It has simply become more sensitive to the next headline.

The important signal here isn’t any single shortage—it’s that AI infrastructure is becoming a coordinated capital system. What strikes me is how compute, memory, networking, power, financing, and real demand are now being locked together through multi-year commitments. My read: the next bottleneck won’t be GPUs alone. It will be whether the entire stack can be financed, delivered, and kept utilized at the same pace.

Just some TLDR news: - $CXMT IPO tomorrow if you like Chinese memory. - Samsung Electronics reportedly struggling to secure large FC-BGA substrates. Ibiden (4062) reportedly requested LTA guarantees, Samsung Electro-Mechanics sought prepayments. - Samsung + $AVGO sign memory + foundry AI framework through 2030, expected to exceed $200b - $SOI expects FY2027 silicon photonics revenue to double compared to the previous year, surpassing Morgan Stanley's 60% growth projection. Photonics thesis go brrr. - SK Group + $NVDA sign $500B+ partnership to build out AI DCs and HBM4 memory. - SKC Absolics glass core delay to 2027 from reports. Targeting final reliability testing EOY, mass production next year. So if you're curious, this does push back some ramps from $LPK and others (hence drop on ER). - $QCOM price hikes by double digits for smartphone processors, due to upstream supplier hike pricing. - $META expected to issue $12B in project-level/SPV financing for El Paso, Texas AI DC expansion - Naver announced a $10 billion investment from $NVDA and Brookfield to construct a 1GW-scale AI Factory. Near term plan is 200MW by 2028. (Nvidia $1B investment, Brookfield nonbinding $9B) - 64GB DDR5 server modules rises 146% versus end-June contract pricing - $INTC brought forward 14A process mass production by a year, risk production H2 2027 and volume production in 2028, vs. 2029 HVM expectations. - Samsung Electro-Mechanics wins $200m MLCC order (existing bottleneck). - $AMD (the Bandana bottleneck), announces Helios is in full production with shipments Q3 2026. Including an up to 2GW MI455X GPU deployment with Anthropic and a 6GW infrastructure rollout with OpenAI. Gave new >50% CAGR TAM to $220B by 2030 from $26b in 2025 for CPU market. Rolls out optical interconnects for Mi500 in 2027. From channel checks, AMD is heading down to the CPO route (seems likely to use Ayar). - $ORCL wins $7B Department of War enterprise software contract - JX metal doubles semi target capacity at its KR subsidary with a 4B yen investment, with operations to begin H2 2027, amid surging demand from major customers Samsung Electronics and SK Hynix - Tungsten hexafluoride spot prices surged 2.1-2.5x Y/Y following Japan's Kanto Denka and Chuo Gas announcing permanent production halts. Fluorinated liquid supply faces a vacuum as 3M plans to exit PFAS production. - From the four optical chipmaker earnings, Yuanjie/Eoptolink/TFC Optical/Dongshan Precision: no major order cuts, 1.6T shipments expected to accelerate into 2027. Optical chip suppliers expected to capture outsized margins from shortages. - Unitree Robotics Targets 30,000 Humanoid Robot Production Capacity by 2026. Read through on humanoid TAM scaling like $CCXI and others. - Energy storage lithium batteries orders increase first half orders by 2,900% apparently in China. Not as familiar with EVE Energy and other battery makers.

@runes_leo 嗯，是写码平权了，但是项目交付没有平权、产品运营依旧是难点。获客、流量更难了。

@xiaokun_saif 对我目前也这么觉得，而且很多能看到的收益也都是极少数人搞出来的或者很难持续的东西

说是这么说，但实际去探索用 AI 赚钱的过程还是挺难的。因为使用 AI 并不是你一个人的优势，世界上所有人都有了这个优势，所以你面临的竞争其实和以前一模一样。你以为自己比以前更强了，但实际上大家都变强了。 这几个月过来，我们可以看到各种各样的路子，比如持续做中转站赚钱的、做流量赚钱的等等，但自己真正能找到一个可以稳定盈利、有持续净收入的项目，依然非常有难度。 我自己也试过不少：做视频、做订阅变现、做付费课程/内容产品、做过工具型小产品、也碰过平台上的 Agent/服务挂牌。每条都能摸到一点「好像能成」的信号，但要变成稳定净收入，还是另一回事。