What master bubble boi talked about is pretty consistent with what I heard from a VC friend last night: Q4 revenue 18.5B due to headwind from seasonality and distillation block. So this translates to year end ARR as 80B+ roughly? Pretty disappointing and thanks to Opus 5.5 for accelerating the growth. 2027/2028 135/270b revenue run rate, which translates to 200B ARR 2027E I believe, also not high enough to beat sell side expectations? Also heard that my margin expansion thesis wouldn’t last in 2027 unfortunately due to capex expansion. Anyway my original thought of “Anthropic S1 saves AI trades” seems not going to happen. But let’s wait for next week to see how the Mr Market reacts when the real S1 file goes public.

Now is a good time to buy memory stocks. If you want to make a bet on AI, the DRAM companies appear to be some of the most disconnected: high growth, high margins (which will stay well above prior peaks if AI demand holds), critical importance to AI workloads (there’s no AI without DRAM), concentrated industry structure, increasingly strategic customer engagements, lack of substitutes, and record capital return, all at valuations that embed a sharp cyclical decline sometime between 2027 and 2028. This is despite companies like Google reportedly expressing interest in locking up memory supply through 2035. Additionally, Elon’s comments yesterday were an incremental positive for DRAM. He said Terafab would start with an SRAM chip slated for volume production in 2H28, which is 2029/2030 in Elon-speak. This implies Terafab is still years away and, importantly, will target a logic process first, not traditional memory/DRAM. I believe this is because he understands how difficult it is to manufacture memory from scratch (especially at competitive yields/economics). Only 4 players do it at scale globally, and all are vertically integrated with proprietary IP. I believe this is also why Elon has been pounding the table so hard on volumes coming out of CXMT. He’s trying to pressure MU/the big 3 to expand capacity too. Ultimately, I think the best outcome for Terafab/SPCX/TSLA and the memory industry is for them to work together. But Terafab is early, as is a lot of what SPCX/TSLA are trying to build, so the DRAM companies are understandably approaching this with care. The worst thing they could do is ramp capacity for the Elon-universe before the actual demand is there and end up oversupplied. Regardless, I hope we eventually see a partnership, especially between Sanjay at $MU and Elon, who were both seated close to President Trump at last month’s superintelligence summit and are both running American companies that are very pro-American AI/SI.

@DratchCap this SpaceX call a vibe changer!?!? Some v interesting quotes. From 6-12mo paybacks. From being able to rent compute many times over. To 10gw maybe more next year.

I talked to a friendly, from a major hyper scaler, on the GPU order to install lead time (per people freaking out some are ordered and not installed) and the lead time is and always has been a range of one to two quarters. Given data center timing (they call this staged integration) the supply chain is notoriously difficult to manage across compute, cooling, power, etc. Any concern of GPUs waiting to be energized is a nothing burger and is totally detached from the reality of supply chain management, especially that of semis. We remain early in the adoption cycle for AI, we need a lot more compute, and the demand profile only increases to rise. Hope that helps :)

xitter.cf/sama/status/2111131677… bit of preview huh? for greater good of humanity.

I’d estimate Ant was generating something like $5b of ARR directly from Chinese distillation efforts going into its September crackdown. I’d guess the entirety of that spend almost immediately shifted to OAI as the next best available model. Normally one would expect Sam to leave that ARR unchecked until they IPO as the recent angst around Ant coming in below the bogey demonstrates that investors shoot first and ask questions later. However the variable that you have to consider is that the state department likely had a lot to do with encouraging a crackdown behind the scenes as evidenced by rhetoric from Bessent and Lutnik prior to the Ant distillation blog post. So normally I’d expect for that distillation ARR to move from Ant to OAI to eventually MSL and then xAI as the Chinese desperately scramble for the most capable model they can freely ride on. But again that assumes there is no unifying force that upsets the game theory of distillation crackdown. One other component to consider is that we are only now moving from frontier model capabilities being as capable as the median knowledge worker to now starting to eclipse that baseline. One could argue that commoditizing the human baseline forces enterprises to adapt to an AI world far more quickly vs if you were to price tokens at human capacity and closer to human derived intelligence. What that does is prime enterprises by forcing them to disrupt existing workflows anyways simply due to cheap cost of human level intelligence which opens up budget for them to spend on frontier intelligence that far surpasses human intellect. As a maxi I believe that fast takeoff implies we are far closer to those classes of model vs what consensus believes and I also believe that in the next week or so we will get the early glimpse of it via Ant’s new model and OAIs likely fast-follow. At that point while bears will point to OAI NNARR slowing again even as Ant accelerates into November due to OAI crackdown on distillation I suspect the first glimmers of “knowledge worker in a box” will completely shift the narrative for EOY 2027 ARR which imho will end up closer to $1T vs current cons $400b As always, long compute 🫡

Google energy chief reveals its SpaceX compute contract is a short-term bridge for "huge" Gemini Enterprise demand, and says Gemini grew from 400 million to 1 billion users in three months “In the case of SpaceX, that’s a very short-term contract because our demand from Gemini Enterprise customers was huge. “We are working in West Memphis, Arkansas, to build our Google-owned-and-operated data center, where we are excited to do the same thing we did in Minnesota: work with our utility in Arkansas, Entergy, to bring customers over $1 billion of energy savings by paying our own way, bringing 600 megawatts of new solar and 350 megawatts of new battery storage onto the system to bring our own generation. “But that takes time. When you have customer demand like we’ve seen for a Gemini product, we had 400 million users in May, and that grew to a billion in August. That’s our fastest-growing Google product ever. “We will sometimes have to use short-term solutions until we get the data center fully connected. But what’s important is that, in the long run, we have a stable energy supply that’s reliable for our customers, and it’s built the right way.”

Delta, Chase and Amazon are blocking AI agents. The war on agents has begun, and here’s how it probably plays out: Stage 1: Companies block agents because agents skip the ads and upsells. Stage 2: Agents learn to act like humans, and it becomes an arms race. Stage 3: The market splits. Companies that make money on sales let agents in, companies that make money on data charge for access, and the giants build their own agents. Stage 4: A trusted middle layer for agent identity and payments becomes the standard way in. Stage 5: Companies that welcome agents start winning customers from the ones that block them. Stage 6: The agent blockers come back and let agents in on their own terms. Stage 7: Everyone starts competing for agents with better prices, faster checkout and SEO built for agents. Stage 8: The ads come back, and companies pay to be your agent’s first recommendation. Stage 9: The agent war ends, and selling to agents becomes as normal as having a website.

If I'm not mistaken, Anthropic's 85% gross margin and >40% compute margin from this Bloomberg story implies that 70-75% of their compute is now going to R&D. bloomberg.com/news/articles/…

The truth is there are far fewer AI-maxis within the investing class than you might otherwise think. Plenty of bottleneck bros. Lots of hyperscale truthers. Abundant semiconductor enjoyers. Infinite momo lovers. But people who have actually glimpsed the future because they can think non-linearly and instantly understood Leo’s message back in mid-2024… an incredibly small handful. The most fascinating part has been that neither Claude Code or Astra changed this. The same people who can think abstractly before the proofpoint continue to be able to after. And everyone else still needs to be spoon fed. I’d estimate within the TMT L/S community sub-5% are true maxis. And no Coatue isn’t maxi. Neither is Gavin who I respect the fuck out of. But he’s rationally optimistic but cautious. Maxi’s start with the assumption that models can be scaled up well past human intellect at a pace that the human mind can’t project and the economic value of that scale up is exponentially larger than existing gdp. Human bottlenecks are irrelevant as models get smarter. And with that north star we work backwards because path function matters and we have to pay rent until the robots are made. It’s a very different headspace vs starting with the premise that hyperscalers are likely to milk a couple hundred billion in cloud revs if this all goes to shit.

The Trillion-Dollar Question Nobody Really Has a Good Answer To - Where Is the ROI? We spend a lot of time debating how many GW of AI compute will come online. But the bigger question is: who ultimately pays for all of it, and can the economics justify trillions of dollars in investment? I find it helpful to think about AI economics as a chain: CapEx → Compute / Inference Revenue → Model / API Revenue → End Demand One important nuance: the market often uses model company revenue (@OpenAI , @AnthropicAI , etc.) as a proxy for AI demand. But outside direct consumer subscriptions, much of that isn't necessarily ultimate end demand. A company paying for Claude Code is still an intermediate customer in the economic chain. The real economic value comes when that company uses AI to generate incremental reve or reduce costs. --- Let's put some numbers around 2027. 1. CapEx: US$2.5-3.0T Assuming ~50GW of new global AI compute capacity, we're looking at roughly US$2.5-3.0T of total infrastructure investment. ~70% goes toward chips and related compute equipment, with the rest covering buildings, cooling, power and supporting infra. Importantly, this is much broader than the often-quoted ~US$1.2T of hyperscaler CapEx. Neoclouds, colocation providers, private developers and others are also funding the buildout. 2. Compute / Inference Layer: US$1T+ Revenue Potential If all 50GW were deployed and commercially utilized, the capacity could theoretically support over US$1T in annualized compute infrastructure revenue. For context, combined CSP cloud rev = sub US$600B on consensus est. 3. Model / API Layer: How Much Revenue Do We Actually Need? @OpenAI and @AnthropicAI are rumored to be approaching US$200B of combined ARR by YE26 (source: public data). My instinct is that we need to see something closer to US$400B+ by FYE27 to begin supporting the scale of investment we're talking about. Rough sanity checks: 1) US$2.5-3.0T of CapEx implies roughly US$400-480B in annual depreciation. 2) Total GW online grows +70% in FY27E, hence ARR needs to grow at least 70% (to $350B)? --- So, was the ROI ever a slam dunk? Never, really. And I think we should be intellectually honest about that. Don't get me wrong. I believe AI is one of the biggest technology supercycles we've ever seen. But when it comes to justifying trillions of $$$$ of CapEx, even the bull case rests on extrapolating the early success of AI rev. Some extrapolate linearly. Others exponentially or logarithmically. The resulting forecasts can be wildly different. And that's precisely why the market tends to overshoot in both directions. It is not really backed by a solid set of numbers, but rather, extrapolations. --- Which brings me to how I think about trading this cycle. The market isn't necessarily trading the absolute level of AI CapEx or revenue. It's trading the rate of change, particularly revisions to expectations. Think of it as two cars racing: 🏎️Car 1: AI Infrastructure Investment How quickly are CapEx expectations being revised upward? For example, since the beginning of 2026, CSP CapEx expectations have been revised upward by roughly 40-50%. 🏎️Car 2: AI Monetization How quickly are AI rev expectations being revised upward? How quickly is actual end demand catching up? The market wants to see 🏎️Car 2 outpacing 🏎️Car 1. If CapEx estimates keep moving higher while AI monetization fails to keep pace, investors will increasingly question incremental returns, even if absolute AI revenue is still growing rapidly. AI can be the biggest technology supercycle of our lifetime, and investors can still overpay for the infrastructure along the way. Both can be true imho.

He really did it. He built a fucking flying car.

Should be pretty evident CoreWeave's cost of capital is declining over any timeframe greater than a few months.. DDTLs (the ABS, project-finance style facilities backed by specific clusters and customer contracts) comprise the majority of CoreWeave's debt. Pricing on the DDTLs has tightened a lot over the years. From S+9.62% for DDTL 1.0 issued in 2023.. to as low as S+2.25% for DDTL 4.0 done in March 2026. Note how the older, higher-priced DDTLs 1.0 and 2.0 are declining as the facilities are fully self-amortizing.. And new, lower-priced DDTLs are drawn to fund the most recent capex. Will be interesting to see where the next DDTL prices. Not going to argue it will be tighter than the S+2.25% (DDTL 4.0) and S+4.50% (5.0) prints from earlier this year. But probably worth noting that DDTL 5.5 was done at S+5.50% (not shown below as it was completed post quarter-end) in July 2026 when AI financing markets were similarly volatile.. And DDTL 5.5 has a 5 year term yet funds capex associated with non-investment grade customers on 3 year contracts, making it a much more aggressive structure than anything prior. In any event, where DDTL 6.0 lands seems much more important than where its (illiquid) unsecured senior notes are trading. $CRWV