Lawrence D. Loeb

@claudeai @threadreaderapp unroll

Available on Pro and Max. Update your desktop app and pair with mobile to try: #dispatch-and-computer-use" target="_blank" rel="nofollow noopener">claude.com/product/cowork…

You can now enable Claude to use your computer to complete tasks. It opens your apps, navigates your browser, fills in spreadsheets—anything you'd do sitting at your desk. Research preview in Claude Cowork and Claude Code, macOS only.

@institLPGP @Midnight_Captl … what you mean by “llm providers won’t pay for distribution” as I don’t think that’s what anyone is saying.

@institLPGP @Midnight_Captl I believe the idea is that software providers would be paying Anthropic/OpenAI/Alphabet/Azure/etc. for tokens as part of the software solution they would provide to customers, sort of like Palantir does, and charge a premium for the resulting product. I don’t understand…

Everyone’s talking about what Jensen said at GTC about AI capex. But the most important thing he said got almost no attention — that software companies are about to become the biggest resellers of AI tokens in the world. Think about what that means. Workday buys raw intelligence from OpenAI. Wraps it in HR domain expertise. Sells it back as an AI product at a massive markup. That’s not a software company anymore. That’s a refinery. And right now nobody is tracking the economics of that refining layer — what the crude costs, what the refined product sells for, and who’s capturing the spread. This is the most important data gap in AI right now. More on this soon. $NVDA

@institLPGP @Midnight_Captl I don’t understand how the LLM takes the margin if the software developer is reselling the llm’s intelligence wrapped with their special sauce. Are you assuming the labs are going to compete with all of their software clients? That the customers are going to try to cut them out?

@Midnight_Captl not when there’s a compute shortage

@PauloMacro @CRUDEOIL231 …paper on huge rolls for institutional bathrooms. The machinery for one cannot be easily transitioned to the other. When people stayed home, they consumed more consumer paper than was being produced and institutional paper went unused! That was the shortage.

@PauloMacro @CRUDEOIL231 The toilet paper thing was actually very interesting. The problem was the types of toilet paper that are produced versus where they are consumed. There are two types: 1- toilet paper on rolls sold in packages in stores, sold with varying softness; and 2 - institutional toilet…

I’m currently in SK Seoul, and here, you have to throw out your trash in government-issued prepaid bags. That’s the rule. Two days ago, news broke that the Korean Ministry of Environment had launched an investigation into nationwide inventory after reports showed that bag manufacturers only had a month’s worth of raw materials left. With the Strait of Hormuz facing a potential blockade, the supply of naphtha has tightened, leading to growing concerns over production hiccups for the polyethylene bags made from it. My wife nagged me to go out and buy some in advance, so I just got back from the convenience store. I was curious, so i slipped the clerk a few bucks and asked some questions. "How much has the volume of sales increased compared to usual?" "At least several times over." "Is there enough stock here?" "I’m not sure. The owner told me not to sell more than two bags per person from now on." "Does it seem like ppl are panic-buying after seeing the news?" "Yeah i think so. Just in case i even told my boss and set a few aside for myself." This is happening in a country with the world’s fourth-largest ethylene production capacity. Right now it’s just a minor anecdote tucked away in a corner. But in two weeks, this will be felt everywhere on the planet. By then there won’t be any "nothing’s going to happen" talk or pointless questions. The clock is ticking at this very moment. #oott #iran

@The_AI_Investor No. Those H100s were all tossed out as useless. That’s why they’re not easy to find according to Burry.

Suhail reports five GPU providers out of stock for 8xH100 nodes, underscoring a surge in AI inference demand that outpaces 2022's training-focused shortage. Suhail didn't know who to ask, give Michael Burry a call, he know a lot of useless H100 sitting idle.

@Angry_Staffer … for Trump. My wife and I were more concerned about HOW they voted for Trump given that they were 3 or 4 years old in November 2024!

@Angry_Staffer …support the strikes on Iran when the Iranian people were protesting. They were particularly upset at the 30,000 lives lost among those most against the regime and how we may have lost credibility with the Iranian people AND NATO! They were questioning why they voted …

This morning, I was at Pearl’s Bagels eating a bagel and lox, chatting with the lady behind me about the war. Her three-year-old chimed in and said that Trump was vastly underestimating the geopolitical consequences that will reverberate from another war in the Middle East.

@Midnight_Captl @taobanker Stealth doesn’t mean invisible, just hard to find. They’re not invincible either. Throw up enough flak and eventually one was bound to get hit.

@taobanker Bespoke is knowing this means we need more $lmt planes

an f-35 was brought down by 3rd world rockets short $lmt, stealth is a lie

@benitoz Ben, I subscribed to your Substack on 3/5 and just got my credit card bill. It was treated as a cash advance and I was charged a transaction fee and interest! WTF? Why did this happen?

@Midnight_Captl Nice. I think the Arm CEO used to be at Nvidia and Nvidia wanted to buy Arm a few years ago, but there were anti-trust issues.

I’ve officially been invited to the Arm conference in SF next week 🤯🤯 I don’t know a ton about $ARM, so I’m super excited to learn more Things are happening!!

@Angry_Staffer So we’re better than Russia, but we don’t have a plan to help the Iranians choose a better government for themselves. The good thing is we haven’t had the massive losses that Russia has had since we haven’t put boots on the ground (even as we had unfortunate losses).

One thing this Iran operation should do is embarrass the hell out of Russia. In the opening minutes of the war, most of Iran’s leadership was dead, and their air defense was largely neutralized. Meanwhile, four years later, Zelenskyy still stands. There’s levels to this.

@BenBajarin @TShirtnJeans2 Sounds right for base. So roughly $400 mm for FY27 and $600 mm for FY28 (excluding Jan ‘28) after $216 mm in FY26, with a lot of wiggle room in that ‘28 number. Pretty impressive growth at this size.

@TShirtnJeans2 @LDLoeb Yeah agree with that which is why I think base case is ~600B.

2027 more likely to be closer to $600B. Networking alone could be north of $60B.

@BenBajarin @TShirtnJeans2 I thought it was pretty clear that he was just doing easy math to clarify his point, saying if everyone got 25% Groq, it would be $250 billion (which wouldn’t make sense because part of the $1 Trillion is Blackwell). Just rough math.

@TShirtnJeans2 @LDLoeb I'm not sure his comment was to be taken as 20% of all customers. He pointed out in his interview with Ben Thompson today not all customers need will need Groq racks. At least short term. so some customers but not all can have decent attach rate.

@KristinaParts Did they get approvals from Commerce? Is this the small amount reported by Bloomberg yesterday? bloomberg.com/news/articles/…

#Breaking Nvidia CEO on China 2/2: "... but that's our condition today, and and our supply chain is getting fired up. And so, you'll like all of you will likely start to hear about that". $NVDA

#Breaking Nvidia CEO on China: "We have received purchase orders, and we're in the process of restarting our manufacturing. And so, so that's new news for all of you, and it's different than it was two weeks ago or three weeks ago" 1/2 $NVDA

@BenBajarin @SethHWeidman Interesting that Jensen left it out yesterday, particularly given how it was highlighted (was it GTCDC or CES?). Was it because the keynote was already 2.5 hours?

@SethHWeidman NVIDIA

Getting some clarification, but CPX is not dead will go forward on its own product path along side Groq. Both have separate, but very specific jobs to accelerate AI.

@BenBajarin @MajorOcelot45 That will likely change with Feynman I’m guessing. Only so much you can do in 3 months!

@MajorOcelot45 Yep

Groq board. FPGA (likely Altera) and CPU head node (likely Intel) and 8 Groq LPUs on top. Groq rack also on the right paired with VR NVL72.

@The_AI_Investor He also clarified that it doesn’t include Groq or Rubin Ultra in the analyst presentation. And that he’s still accepting Blackwell and Rubin orders (and that it’s through 12/24/27 at 4 pm PST - which was for fun).

So Jensen told Jim Cramer that $1T is the demand for just Blackwell and Rubin thru 2027: 1/ everything else stays on top of it 2/ more demand to come