John Huang

@Linahuaa lol, you think the rest of the world looking at this and think to themselves "woah, US is banning every non-US person using their model. I better bend the knee, get on their good grace so they will allow me to use a weaker version of their best stuff"?

I said it a thousand times.... The whole Mythos glasswing thing. The invisible sandbagging. The export controls. It's ALL about countering China. You really think any American elite cares about some brown terrorist nuking a million White Americans with a LLM-built bio weapon? They have bunkers and canned spam, bro! They can eat this shit for years without getting tired of it! China is the only relevant factor on this game-theoretical battlefield.

You sorta brought this upon yourself no? Also this makes complete sense for trump, can’t control inflation, getting bent over by iran, screwed sideway by bibi, how do you win mid-term? Flip and turn on AI bros to destroy them seems like the only option left.

@JaceHoiX 话虽如此不过kweb每年还派2-8%分红呢

我跟Fable说我满手中概股，什么时候回本。 Fable反手甩了一堆详细数据出来，让我别再碰这种脏东西。

@DarioAmodei tldr: we can't compete, ban them, let me develop new models to compete. Actually, screw that, make it noone can develope or use AI without my generous grace and good will

Today I'm publishing a new essay, Policy on the AI Exponential. AI is progressing extremely fast—much faster than the policy process was built to handle. The essay lays out where I think the technology is now, and the action needed to close the gap: darioamodei.com/post/policy-on…

@RYANHINGSHING @HarryXu920395 @ShanghaoJin 中环汇不错，不过星期天不开门

@HarryXu920395 @ShanghaoJin 不是吃膩了, 香港點心真的變差了, 很多造點心的基本功很差 例如我給一個蝦餃的標準 皮: 是否晶瑩剔透, 皮薄陷靚(看到蝦餃粉嫩), 彈牙微黏性但不黏牙和筷子 蝦: 鮮、彈牙(加筍絲 香港現在沒有了) 形: 飽滿和挺 味: 要蝦鮮味(+筍鮮) 我都不要求蝦餃13摺了, 現在在香港哪裡能找到這種點心呢? 謝謝大家

你们不要给我瞎扣帽子😡 我什么时候不支持“台湾回归”了？！不要让我告别深圳的早茶，香港的太难吃了 btw 经济大发展～～emmm～～我不知道，反正马总统任上合作亲密无间，现在分道扬镳。只好叫台湾小伙伴来评论下

@zephyr_z9 Those powerbus aint getting any smaller

Not so fast Maybe in the early 2030s

@mijiu_btc_eth 什么鬼，spcx没有NOII，直接9:30开

SPCX今晚上市，不是美股9:30开盘就能交易。 开盘后会先进入IPO竞价阶段，NOII（订单失衡数据）预计北京时间22:00左右开始出现，市场可以看到指示开盘价和买卖盘失衡情况。 参考历史大型IPO，SPCX首笔真实成交大概率在23:00-01:00之间诞生。 对于Hyperliquid、Binance、OKX上的SPCX合约来说，NOII将成为最重要的定价参考之一。竞价期间波动可能明显放大，插针风险较高。 注意！今晚别把指示价当成交价！

Good interview

@RYANHINGSHING 不要慌，市场这反应，很有可能oracle会硬拉一个大反转出来，NFA

审视了多方人士信息后，我也决定跟 $ORCL 分手 。除了市场会质疑新投的CAPEX ROI一点外 -云有后发者优势，红利期可能到头。先进者投了旧架构，并不高效。后进者/有现金者投入可能更高效 -未来可能有”新云”加入竞争 和你在一起的3个月，是我人生中最快乐的日子，我跟CDS在一起了，谢谢oracle🥲

that's not how it work for institution though, if you bid $1M to buy SpaceX, you just send an email, there's no capital commitment until you receive your allocation. And usually you indicate to your sales how much you really want/expect. The $1M is what shows up on order book, but real demand? I'd guess around 1/4 or not less.

it’s 5x oversubscribed, and many big funds including Middle East sovereign funds are committing to it. The market is overreacted. @BruceCMaster brought up an interesting idea, if you reserved $1M to buy the IPO but only got $200K shares, what’d you do with the released $800K?

Just doing a quick back of the envelope calc here. I think a U.S. frontier lab is running inference at around a 30% margin. But then you look at DeepSeek’s pricing. They’re literally charging less than 1% of what a U.S. frontier lab charges. And the actual bill might come in even lower than that, because their cache hit rate is so effective at scale. At the volume they’re serving tokens, even DeepSeek couldn’t afford to subsidize it forever, so they must at least be breaking even. So, what does that tell me? It sounds like Chinese labs’ inference costs are just a fraction of what U.S. labs are. Let’s say those US labs shift focus from chasing that extra 5% in intelligence to optimizing efficiency. Could we see them cut inference compute demand by a third? If so, suddenly AI hardware doesn’t look nearly as constrained anymore, does it?

$ORCL half of the RPO is 3 year out, and 1/3 of the rest is customers bringing in their own hardware so margin is going to be tiny. CAPEX off the chart and 20b ATM equity to come, diluted eps is just about market consensus assuming stock price hold (stock fall -> sell more to raise -> stock fall harder). Overall though, it's really not that bad if rates are not at this level, the discount would be bearable and people would be willing to pay for the ROIC.

looking at claude-5, I'm convinced anthropic is stuck on dead tech. Their cache is 5min TTL, which is incredibly hostile for full time usage with any human in the loop, which is exactly enterprise usage -> So this must not be a choice, but rather a feasibility issue -> their model is not trained on proper KV pairs for better cache -> Retraining takes 12-18month but model must keep evolving otherwise they fall behind -> stuck on legacy dead tech...until they devise a way to adopt / post train a better cache knowledge (is that even possible)?

@saso_capital innoscience being the market share leader and only one with actual fab ... didn't even get a mention lol 🤣

@Wujizhuzhu 进了指数，什么价就由不得你了，全世界的对冲基金都拿你当垫脚石（funding short）用

心疼1993港元买智谱的朋友， 也心疼1330港元买minimax的朋友， 你们这辈子估计是回不了本了……

智普摸了一下2000也下来一半了，你只要进了恒科，股价就不是你自己说了算了...全世界的对冲基金都会拿你做funding short 🤣

彭博：中国政府准备未来五年花2万亿人民币做AI基建 就这？美股七姐妹今年投入都1万亿美金/7万亿人民币了...

@zephyr_z9 30-35$ sounds about fair if you map it to the rumor rubin memory cut vs. cost drop.

No, it won't hit $53/GB $35/GB is best best best case scenario Should be in the $25-$30 range

@RJCcapital don't forget the 2x leveraged ETF in hong kong...that's 8b of forced selling waiting on a gap down Monday. -30% on hynix the etf is -60%, even if the fund survive, the holder is perma wiped.

we are so cooked on Monday

Looking at $EWY $KORU , safe to say Monday will be...interesting. The impending doom of 7709.hk , HK listed 2x leveraged HYNIX etf. If hynix trades at -30% limit down, what happens to the leveraged single stock etf? 1) Swap counterparties (who provides the leverage) decide to margin call CSOP (7709.hk's issuer), and CSOP decides fvk it we'll just wind down the fund. ETF holder is completely wiped at whatever price the counterparties decide to charge/dump next day. 2) Let's say the issuer decide to inject capital the save the ETF. But then, next day everyone knows you're under the gun, Hynix open down another 20%, your 2x leverage fund is effectively wiped, your injected capital is wiped, and it's game over for every ETF holder + yourself. So, the rational choice for the ETF issuer is mark everyone down 60%, puke up the difference to save your brand reputation, and call it GG.