Macro Equity PM | Dartmouth Fball ’08 “Be bold, and mighty forces will come to your aid.” Views expressed are solely my own and not on behalf of my employer.
NYC
Joined December 2010
- Tweets2.9K
- Following314
- Followers8.7K
- Likes4K
Pinned Tweet
DratchCap Mkt Update. Some thoughts and self criticism.
Been a brutal couple of weeks. An equal-weight AI basket I track fell ~24% in seven sessions vs ~6% for QQQ, which was the worst 7-day window in five years, worse than DeepSeek or July. Thinking about my own mistakes, I did not respect price action and rates enough, especially given the financing reflexivity broadly. Index shorts did start to work but I did not press the way I should have. Another lesson re-learned.
In these types of environments I take things day by day and look for ways to improve the shape of the portfolio. Sometimes options structures line up well for that and can be good stock replacements. Stay in the game, reduce the tail.
Prior to this AM, things looked aggressively oversold. De-grossing has been material (see MS HF data today, last week 3-sigma event) w/ AI long/short near 5yrs lows.
From here, I think the bigger near-term risk is macro. People have been hiding in quality, and the systematic community is still long because the index has held up much better. If the quality rolls, there's plenty of room for the index to catch down.
Meanwhile, the tech demand indicators I'm tracking remain extremely strong. GPU availability is near its lows, rental rates are rising across chips and tenors, and TSMC is stepping up multiyear investment. Recent buildout data points (@SemiAnalysis_ today) are also encouraging. And my recent lab checks have been very strong too, with newer models driving demand and more capacity finally coming online.
At the same time, the market is treating funding as broken across the board. But the big platforms can still raise at prices that make sense given their ROI / ROE and strategic imperatives. A lot of the neoclouds are also financing against contracts and prepayments from those same customers.
Some valuations are also getting extremely attractive. Memory is a good example. Demand is compounding much faster than supply, pricing is going up, and yet the stocks trading near historical trough on fwd book w/ 30-50% type FCF yields.
As for the rest of the year, there is a Mountain of Worry to climb...
Strong megacap prints (soon) and better lab data could get people front-running the November seasonal (predictable AI era pattern). And that loop can feed FOMO into the Anthropic IPO. Expectations there have reset and HF positioning is now light in AI.
Add in the potential for the Fed to hold, power deals to start clearing post midterms, or some relief on the war and rates. We don't need everything to go right but a few will do.
I like to think of the AI buildout era as a super boom w/ mini busts along the way, often when price drives narrative. The mkt is more vulnerable to that than ever bc of new fast info channels (X), and the intangible and exponential nature of the tech. Human's aren't wired to think that way, and skepticism is always more cosmopolitan. On top of that a lot of the people making the big decisions are removed from the details and don't use the tech day to day, so when the tape turns they have limited first principles to fall back on. Staying the course during these moments has been right each time and I think that will prove true again.
Good luck out there!
Tonight is very positive for hyperscalers, I was very nervous the memory companies would break the LTAs. But I guess they are safe... for now. 😜
All this new info today has me wondering what runaway RSI is worth? (Great take attached below as well)
ANT’s S-1 suggests it’s kind of a lot…
First, capability drives everything, from that beautiful rev per dollar of compute chart below to diffusion. As models improve, the opportunity cost of waiting to adopt goes up AND the cost of integration goes down. The part where the AI's own improvement helps make it easier to integrate ("hire itself") is lost on many. It’s partly why I think adoption is a “reverse bankruptcy” in waiting (slowly and then all at once).
Now imagine RSI shatters the open source “6mo behind” credo. Suddenly, open source and other frontier models are expensive in the context of the work they can’t do. Better share and pricing for the lucky winner…
But that capability gap also means negotiating power. ~1/2 of ANT’s revenue runs through marketplaces. My read of the S-1 puts the clouds’ current cut at ~40% of gross profit on those sales.
ANT’s already building out its own compute. Add a bigger capability gap and a war chest to fund more of that buildout, and that smells a lot like a lower vig for the clouds in the making. While we don't know the terms of each agreement, I think it's a good bet there's either sunsets, outs, or other renegotiation triggers.
Put all this together and RSI is potentially a legendary parlay.
Some napkin math. Rev / $ of compute was $2.08 in 2Q. Assuming other operating costs at 15% of revenue, a simple model gets ~$40 of operating profit per $100 of compute.
IF however RSI gets ANT to $4 rev / compute dollar and helps ANT negotiate the cloud’s gross profit share down to 10%, operating profit per $100 of compute grows 5x to ~$220 👀😵💫
The near-term margin warning and training focus makes sense in this context. They’re paying up for short-term compute while investing in the next models specifically to make this bet.
There may be other ways for them to win and demand obviously still has to show up to absorb all the eventual compute. But the RSI success case is extremely powerful if they can achieve it.
And if this all sounds insane… well, add it to the list ;)
(thanks to the AI Maxi homies for helping and inspiring).
Build a sustainable business economically serving inference with a minority of your compute
Direct the vast majority to research and training because you see accelerating progress, which makes each unit of existing compute even more productive
Achieve RSI, at which point compute loops, progress, and productive capacity of each compute unit accelerate even further
Structurally widens frontier gap to also-rans
Allows higher capacity to pay for a unit of compute than anyone else, because you earn differentially better returns on it
Gives you more compute than anyone else, which accelerates the loop even more
You have the best intelligence and it monetizes it in a power law fashion, sustainably
If you don’t believe this none of it works. But if you do there is only one real path.
Well that's bc he's still going through the S-1 the old fashioned way
This quoted post is unavailable.
Look Mom, I can guarantee my own debt!
Anthropic already has $132bn of cash.
Add a ~$140bn IPO, $77bn of committed funding from Google, Amazon and Broadcom and cash from a business growing 14x a year and it could be sitting on roughly $400bn by the end of 2027.
That's more firepower than almost any company on earth, aimed at one thing and one thing only: compute.
In a race where the winner is whoever can build the most, Anthropic is about to have the deepest pockets in AI outside of Big Tech itself.
Startups are the diffusers… ?
Chamath explains how VC funding circularity is giving AI the diffusion problem Ken Griffin is picking up on
“So the cycle here, just to be very precise, is VCs raised from pension funds and LPs and foundations and family offices. They put money into a startup. That startup takes that money and invests in growth. Part of growing is by consuming other people’s AIs. And so that company then starts to show revenue traction. Those folks then say, hey, look how well we’re doing, raises a bunch of money from the same VCs. And it’s just this circularity.”
“Why that’s important is what Ken is pointing out is that for AI to really scale as a predictable driver of long term earnings, it has to diffuse out of the startup ecosystem. And so you can’t be in a situation where on the one hand, you’re talking about breathtaking valuations. But we’re 75 cents of every dollar is actually just a startup sending it back and forth so that they can mark themselves up. That’s, I think, a structural problem.”
“But the difference there is that AWS was abstracted from Amazon retail, which was already then a roughly breakeven profitable business, at least on a free cash flow basis. So what’s different here is you have a bunch of money losing entities just spinning a bunch of money around and it has to diffuse more broadly.”
“They said, we have decreased our spend on AWS and GCP. We now have six completely independent ground up data centers that we’ve built from scratch for ourselves. And I was like, what is going on here?”
“AI has a diffusion problem. It is too concentrated and ripping around in a small little ecosystem. We need to diffuse out. But as it diffuses out, people are picking open source is what I would tell you.”
There is a term for this that I suppose not everyone knows: sandbagging
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.
Banger from @evergreencap3 🙏🙏🙏
I’d add:
- Asian funds hate ✅
- Pods are short ✅
- peak this, peak that ✅ ✅
- “despite being screwed recently on LTAs, all the mgmt teams (and boards) whistled Dixie and signed new ones to get screwed again!” ✅
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.
Overheard: “I don’t think I’ve ever seen this big a difference between prevailing investor wisdom and business reality”
Most should know the clouds will be Rip City this eps season. BUT it’s still nice to see the checks clear…
@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.
To anyone worried GPUs are "sitting in warehouses" or fwd-ing tweets about it, thank you! It is good to know who is who in the zoo.
And thank you to Ben for getting it right below.
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 :)
The kids call this ‘glazing’ I’m told
xitter.cf/sama/status/2111131677…
bit of preview huh? for greater good of humanity.
Great post, I’d only add that competitive open source is already holding back lab ARR at different intelligence/price points. If the distillation crackdown slows near term growth but also weakens that competition, you have to update both sides, i.e. you can’t ding it twice.
And half tongue in cheek... what is open source’s moat exactly? Cost, privacy, control, tuning... yada yada, and I get it. But this mostly rests on an acceptable capability gap which may be a prior that needs updating.
If no more distillation + more scaling + RSI = a 24-48 month frontier lead, things will change dramatically. That seems hard to believe today but… this is all kinda hard to believe?
And to the “we don’t need that kind of capability”... maybe for what you’re doing today. Let’s see what folks with bigger imaginations want to do once they have it :).
BTW, the labs sell cheap intelligence too and a lot of open model benchmaxing hides that it’s more expensive than the sticker price in real world use. I think both will thrive but I also think most east coast investors have too low a probability on a frontier pull away scenario
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 🫡
So even Gemini demand is ripping... what kind of circularity wizardy is this?!?
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.”
Interesting to hear the 'token prices are falling!' hysteria from the macro community when most of it looks like a mix shift to cache reads. Cache read share went from basically 0 to 85% w/ explosion of multi-step agentic tasks. And ironically they are like 90% gross margin. As always, a little context helps (pun intended 😜)
What the consumer wants, the consumer gets…
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.