Puppet

[Fox News @FoxNews]

BREAKING: President Trump says he is going to continue bombing Iran "very hard" after it shot down a U.S. helicopter over the Strait of Hormuz.

"We're going to be attacking them and attacking them very hard."

"I've been working with Iran for a number of months, and they should sign their deal. It's a good deal. It doesn't give them the right to have a nuclear weapon. In fact, it totally prohibits them from ever having a nuclear weapon."

[Javier Blas @JavierBlas]

BREAKING: US President Donald Trump says US will resume attacking Iran "very hard" as soon as "today."

Trump cites downing of US helicopter as reason for resuming attacks; he adds Iran should sign deal. "I don't know what they're doing," he tells reporters at the White House.

[Aaron @Aaronwei3n]

"EMIB to 90-95%"

The figure is irrelevant, substrate yield matters more.

[Meer | AI Tools & News @Meer_AIIT]

🚨News: Meta is building its first AI data center in India, and it's teaming up with Reliance to do it.

the project is a 168-megawatt AI-enabled facility in Jamnagar, Gujarat.

Meta is leasing the capacity while Reliance handles everything from design and construction to power, connectivity and day-to-day operations.

the center will run on renewable energy and use desalinated seawater for cooling.

Meta is covering the full cost of the energy and water it uses.

Reliance says it'll be ready within two years and can expand over time.

It will also plug into Meta's global AI computing network.

On top of that, Meta has contracted nearly 1 gigawatt of new renewable energy in India through CleanMax and Fourth Partner Energy.

the deal builds on Meta's $5.7B investment in Jio back in 2020. Value of the new agreement wasn't disclosed.

[Nick Timiraos @NickTimiraos]

The core CPI rose 0.21% in May, very close to where forecasts had it, and this pushed the 12-month measure up to 2.9%.

This is the first time since Dec. 2022 where the 12-month core CPI reading was higher than the year-earlier reading.

[*Walter Bloomberg @DeItaone]

ISRAEL SAID TO PREPARING STRIKING IRAN AGAIN

[*Walter Bloomberg @DeItaone]

TRUMP: IRAN TAKEN TOO LONG, NOW THEY WILL HAVE TO PAY THE PRICE

[Noam Brown @polynoamial]

https://t.co/oWqzT12RtZ

[bubble boi @bubbleboi]

The commodity thesis is half right, and it’s the half that’s doesn’t pay.

Yes, tokens at the “good enough” tier are already a commodity open weight models from DeepSeek, Qwen, Llama have crushed pricing for mid-tier intelligence, and that tier is now a brutal cost-per-token knife fight. But that’s also where margins go to zero, winning it is like winning a refinery, not an oil field.

The frontier however behaves very differently because the demand isn’t for tokens, it’s for completed labor. When a model is doing agentic coding or multi-hour knowledge work, a model that’s 10% better doesn’t only earn a 10% premium because it completes tasks the cheaper model fails at entirely, and the buyer on the other side is comparing the token bill to a salary, not to another token bill. That’s why the quality-price curve at the top is so inelastic.

And being “Good enough” is a moving target every time capability jumps, the set of economically valuable tasks expands, and the frontier model captures the new tasks before the commodity tier catches up 12-18 months later. The moat isn’t the weights, those depreciate fast, it’s the compounding loop of capital access, RL environments and post-training data, product surfaces that generate proprietary feedback (coding agents especially), and increasingly co-designed silicon and inference infra.

Even worse the argument for commodity models as gotten weaker in recent months. We’ve seen pre-training plateauing but the scaling axis only shifted, it didn’t end. RL/test-time compute is arguably worse for commoditization, because it’s harder to distill and replicate than next-token pre-training was. The recipes, environments, and reward engineering are where the secret sauce lives now, and they don’t leak through a weights release.

[拖拉机 @tuolaji2024]

看完了报告，

1，800VDC 电源 → 推迟到 2028+
2，CPO 上量 → 推迟到 2029+
3，铜和可插拔光模块 → 继续赢
4，CPO系统良率仅19%且涨多了，交易非常拥挤 → 短期承压
5，存储器（HBM/DRAM），真正的物理瓶颈，供需紧张不受影响。
6，宽禁带半导体（GaN/SiC）放缓。

整体来讲，全面否定了光通信的下一次升级，利好老干部光模块和 npo，就这样。

[Boris Cherny @bcherny]

Fable 5 is the biggest step up I’ve felt in our models since Opus 4.5 back in November. After 4.5 came out I uninstalled my IDE when I realized that I’d been doing 100% of my coding in a terminal for a few weeks. With Fable, it’s felt like Claude has stepped up from being a coding agent to a thought and design partner in building the product. Fable has judgement, taste, and dimensionality in a way that previous models didn’t, leading me to trust it more with the most complex work.

I think the first time I had this realization was when I asked Fable to debug something. It is the first model I have used that was so methodical and precise, taking measurements and adding logs then verifying that it truly fixed the issue before declaring victory.

There’s nothing in claude code’s prompting telling the model to do that, it’s just part of its personality. It really has this “big model smell” that I haven’t felt before.

[ClaudeDevs @ClaudeDevs]

Claude Fable 5 changed how we work on the Claude Code team day to day.

We used to verify that Claude did the work right. Now we verify that it's doing the right work.

Here’s the 3 biggest changes:

[Chubby♨️ @kimmonismus]

Claude 5 Fable Benchmarks!

Holy moly, significant jump even to Mythos

[*Walter Bloomberg @DeItaone]

TRUMP ON IRAN: UNITED STATES MUST, OF NECESSITY, RESPOND TO THIS ATTACK ON HELICOPTER.

[*Walter Bloomberg @DeItaone]

TRUMP ON IRAN: IRANIANS SHOT DOWN ONE OF OUR HIGHLY SOPHISTICATED APACHE HELICOPTERS WHILE PATROLLING OVER STRAIT OF HORMUZ

[駿HaYaO @QQ_Timmy]

Semianalysis:
電源關閉，燈熄滅。800VDC推入和CPO延遲
作者： Konrad Wang、Nigel Chiang、Nick Doyle、Doug O'Laughlin 和 Daniel Nishball
日期： 2026年6月9日

Nvidia原生單端800VDC設計的大規模採用，800VDC出貨量已推遲至2028年。400VDC，一個獨立的高壓直流架構，仍按計劃用於超大規模晶片組部署。

行業傳言表明，超大規模企業正對採用由英偉達推動的單端800VDC架構的做法進行抵制，因為800VDC對魯賓來說並非必需，其仍可使用50VDC。他們認為，以350-450VDC的電網電壓供電，然後提升至800VDC再降回至50VDC以供計算機托盤使用，這種操作方式效率低下。相反，我們看到超大規模企業正日益推動在降低電壓前以更高電壓輸送電力至計算機托盤。

+400VDC產品按計劃將於2026年第二季度如期推進，這主要適用於超大規模企業的內部專用集成電路（ASIC）部署。我們預計400VDC側掛式產品訂單將於今年晚些時候達成，而生產工作則將於2027年第一季度逐步展開。

• 總之，這意味著原本由Rubin Ultra/Kyber交付驅動的側車出貨量將被移入2028年窗口期。

2027年CPO預期顯得過於激進，與當前的普遍預期相比，共封裝光學器件（CPO）的交付將有所延遲。我們預計會下調2026年和2027年CPO出貨量規模的預期。與此同時，儘管市場一直呼籲從2029年開始啟動CPO的規模化生產，但我們始終認為從2029年起CPO的規模化生產進程將切實推進，儘管市場曾預期CPO的規模化生產時間點或為2028年甚至2027年。相反，許多NPO項目將開始加速推進，但這可能更有利於收發器供應商的發展。

對於Scale-out CPO交換機，CPO系統級集成是關鍵問題，而產量經濟性仍然是主要障礙。在樂觀的情況下，光引擎的連接產量為95%，每個ASIC有32個COUPE，系統的產量僅為約19%。市場預計到2027年每年將生產70萬至100萬台以上的Scale-out CPO交換機，但這些問題導致當前的產量水平大大降低，使我們無法達到這些數字。在即將發布的網絡模型更新中，我們可能會下調Scale-out CPO交換機的出貨量。

總體而言，我們從Computex展會回來時，對安費諾、Vertiv、Forgent Power Solutions、Legrand 和 FormFactor 的看法逐漸變得更加積極；相反，我們對Lumentum、Himax、Navitas 和 Wolfspeed 的看法則逐漸變得更加負面。

[Herman Jin @ShanghaoJin]

好～Semianalysis终于给机构发文说CPO和800V延后了

美国投研简直太滞后了，一般得到信息往后数2周都没一个人提。这还是先信的，等晚点大家都讲还得一个月

我不方便把批发来的信息公开说，小范围朋友们都看到了吧

其实，CPO和800V是时间问题，眼睛盯着模型就行，模型不弱智啥都无所谓

[Herman Jin @ShanghaoJin]

@himself65 CPO概念的不是个好的入场点

因为全链没都经过量产考验，在还没看到“扑街”式新闻之前，还是小心点

我不可能想象如此大规模需求，产能、良率会“顺利”完成

[AZKicks @RealAZKicks]

I got super lucky as a VC in Solana’s early rounds and then went degenerately into every shit scam that SBF told me he was going to pump

We were up gorillions and every billionaire in the world was calling me and SBF every day to try and get into the ecosystem cabal

Then SBF went to jail and I eventually crashed out as most of our LPs tried to exit via secondaries in the fund at a huge discount just to avoid getting zeroed

After that I nailed the perps thesis intellectually but proved I suck at asset selection by backing Drift, which got continually exploited and rugged, after also backing Alon at TGE and getting rugged

Then I started a DAT and we absolutely bazookad our whole treasury into SOL at the picotop, it hasn’t gone up once since we deployed

Then I got kicked out of my fund and they started buying Hype in my absence and are already up more than I am cumulatively on all my deals from this current cycle

Oh wait, that is wasn’t me, his name was Kyle Salami

[Papa Johns @SVTrivo]

In the past, the memory cycle was led by set makers and chip designers (Qualcomm, MediaTek, Apple, etc.). They would say, “We need this much LPDDR5X at this price,” and suppliers would follow.
Now the situation has completely reversed.

[The new memory-driven market]
* Samsung, SK Hynix, and Micron are reallocating wafer capacity heavily toward HBM and high-density DDR5 because AI hyperscalers (Microsoft, Google, Meta, etc.) are locking in long-term contracts at premium prices.
* Production of mainstream mobile memory (LPDDR4X/LPDDR5X) is being cut or phased out. Suppliers are focusing only on 12GB+ high-capacity products.
* Chip designers and phone makers were still designing around 8GB being sufficient, but memory suppliers notified them:
“8GB production will stop from X date. We will only supply 12GB and above. If you need it, place orders now. If not, supply will become difficult or unavailable later.”

[What happened as a result]
* Most models were forced to jump from 8GB → 12GB.
*Once the 12GB was in place, engineers said, “Since we have 12GB now, let’s fully utilize it for on-device AI, local LLMs, enhanced multitasking, etc.” — creating a positive feedback loop.
* The jump to 16GB is still limited to ultra-premium flagships (Galaxy S25 Ultra, iPhone 17 Pro Max, etc.).
* Meanwhile, mid-range models face the opposite pressure: RAM downgrades and higher prices.

This is clear proof that memory has become the new silicon gold. Non-memory chip designers now have to adjust their AP/NPU roadmaps to match the memory suppliers’ plans.

#MEMORY #SKHYNIX #SAMSUNG #MICRON