292: Druckenmiller & Collison, Constellation Software, Nvidia's A.I. Factories, Air Traffic, Aswath Damodaran, Semiconductors, Grid-Scale Batteries, and Blender
"trying to listen to the horribly distorted jazz"
Once you reach a certain level, it's really all about execution.
When you're competing against the top 1% in your field, they all know what needs to be done and how to do it.
What separates the best from the rest of the pack is how well you do it (execution).
—Callicrates
🦢 At this point, I’ll be more surprised when we have a white swan year…
(I know, it’s not the proper use of the term, please don’t tell Taleb 🤫)
🏥 🩻 On Friday, I had my first ever MRI scan (context in the intros of editions #290 and #291). Some observations:
That tube is smaller than I expected. If I had been built like a football player or obese, this would’ve been quite tight (I’m 6’1”, 190lbs).
I expected them to be more paranoid about ferromagnetic stuff in the room where the machine is, but I could bring the little metal key to my locker and the guy was like “yeah, it’s fine”. I guess my mental picture of anything metal flying across the room was a bit overly dramatic…
The sounds that the machine makes reminded me of old PC hard-drive sounds, and old PC speaker sounds (really old games, before dedicated sound cards like Sound Blasters became popular).
There was a minute or two at the beginning when I was trying to hold still, looking at the roof of the tube a few inches from my face when I suddenly questioned things… Like, am I claustrophobic and I didn’t know it? Will I freak out?
The door of the room had some specs, so I know the machine’s power was 1.5 tesla.
It turned out to be fine. I mostly meditated with my eyes closed, trying to listen to the horribly distorted jazz that they were piping through some kind of weird medical headphones (over earplugs)…
I thought I was about 10 minutes in when they took me out and said that had been 20 minutes. I guess it’s a bit like a sensory deprivation tank. Time loses meaning in a MRI machine…
🇺🇦 Been a while since I wrote about Ukraine. Noah Smith has a very lucid overview of the situation, and I agree with his call to not wimp out and hand Putin a victory that will only lead to more pain, rather than lasting peace:
🤩📝 Substack added a feature that is going to make my life so much easier — and maybe yours too, if you need to link to something from this steamboat:
I’m talking about the little chain icon-thingy that pops up when you hover over any sub-title. If you click it, it copies a link to this exact spot down the page in your clipboard.
🍾 🎉 Friend-of-the-show and supporter (💚 🥃) Jim O’Shaughnessy is officially retiring from the firm with-his-name-on-the-front-door at the end of the year (OSAM is now owned by Franklin Templeton).
I just want to offer kudoses (that’s my fake plural of ‘kudos’ and I stick with it — kudii would be funnier if it was Latin…) and congrats, and I’m looking forward to what the next thing will be (Jim teased some new projects here).
If you haven’t heard it yet, check out my conversation with Jim on Infinite Loops, it was a fun one:
💚 🥃 If you are not a paid supporter yet, I hope this is the edition that makes you go:
“Hey, I think I want to support what he’s doing here.”
Thank you for that! 🤠
A Word From Our Sponsor: 📈 Revealera 📊
Revealera provides data and insights for investors into hiring trends for 3,500+ public/private companies + technology popularity trends for 500+ SaaS/Cloud Products.
We give investors insights into:
Job Openings trends: Insights into a company’s growth prospects.
Technology Popularity Trends: Insights into how widely products like Datadog, AWS, Splunk, etc, are gaining adoption.
Vendor Sign-ups (Currently Alpha) tracks the # of companies, as well as the specific companies, that have signed up for SaaS products such as Zoom in near real-time.
Visit Revealera.com for a ✨free✨ trial/demo.
Investing & Business
🔮 Stanley Druckenmiller in conversation with John Collison 💳 (Stripe co-founder)
I haven’t had time yet to watch this video yet — lots of family stuff this weekend — but I read the detailed summary by JSC (the artist formerly known as John Street Capital) and it seems like an interesting convo.
✨ Constellation Software ✨ acquisitions per quarter since 1Q2021
Nice spreadsheet compiled by ZeroExpectations:
Period: # deals, total size, size/deal, size % as CFO
1Q21: 21, $320m, $15m, 65%
2Q21: 15, $260m, $17m, 150%
3Q21: 18, $120m, $7m, 40%
4Q21: 13, $490m, $38m, 144%
1Q22: 13, $210m, $16m, 40%
2Q22: 16 deals so far
All internally funded
No sign of debt raising yet
Click the link on his name to see the list of acquired companies.
Constellation does more acquisitions in one quarter than most companies will do over a decade. And their acquisitions actually create value, unlike most... 🔥
What’s the most important idea to understand CSI and how they add value to their acquisitions?
Why isn’t this just buying low multiple assets and having the markets give them a higher multiple, doing this arbitrage until the music stops and the market realizes that a pile of low multiple assets should deserve a low multiple too?
The primary thing that CSI adds to these businesses is a reinvestment engine.
Take a bunch of capital-light cash cows dominating small niches. They don’t have much to spend the cash on, so in isolation, they’d just pile it on the balance sheet or dividend it out, not compounding value that quickly over time.
But inside CSI, the cash generated can be redeployed via acquisitions, and because of their discipline and coverage of many tens of thousands of VMSes around the world, they have a very good track record since 1995 of not overpaying for assets and keeping the returns high for the whole system. This doesn’t happen by accident — they keep tracking every single business separately forever, have base rate metrics by vertical, built long-term incentives around profitability and organic growth, etc.
They also add value through operational best practices and better capital allocation within units, but most of the value comes from what is layered on top of all these tiny businesses, not changes that take place inside of them.
For more on Constellation, check out the pod I did with my friend MBI (💎🐕):
Jensen Huang of Nvidia on the coming age of A.I. factories and edge compute 🏭🤖
Jensen went into vision-of-the-future mode and made an interesting speech about the 4 past types of data-centers, and the 2 future ones:
The first data center was a supercomputing center, right, Amdahl, Cray, so on, supercomputing centers.
The second is the enterprise computing data center, IBM.
The third, hyperscalers, the invention of Hadoop, in-storage computing, Yahoo!, okay?
And then the next one is cloud computing, which is Amazon. Does that make sense? So I just described the early days of each 1 of the 4 data centers that exist today.
They're all quite large. Each one of them added to the previous data center because it has a different need. It serves a different purpose.
Interesting to note that following this arrow of time, you can also see data-centers move from bespoke, custom computers (like Crays) to more commodity hardware, like the hyperscalers using a zillion parts from Intel, AMD, Nvidia, Seagate, Micron, etc.
There are 2 new data centers that are coming.
The new one that's coming out is what I call an AI factory.
An AI factory does one thing, just like a factory does one thing, it could be manufacturing cars or it can be refining oil or whatever you want to, making chemicals or whatever it is, making plastic, whatever it is. And so at that factory does one thing. — data, in this case, data comes in, it gets refined and what comes out as a model.
Data is coming in continuously. It's being refined continuously, 24/7, and models are being updated continuously. It does one thing.
In fact, if you look at one of the most popular applications in the world and potentially the most disruptive new Internet application in the world, TikTok. There is a factory that is refining the AI model continuously. It's gigantic, utterly gigantic, potentially one of the largest data centers in the world.
We're building many, many of those all over the world. In my opinion, there are 115,000 large factories of traditional industrial revolution times.
Now you're going to see 150,000 giant factories, and their job is just to refine data, create models, AI factories. We're in the beginning part of that.
If you look across all the companies that are doing things, and you think to yourself, is this a service — is this an application that has a continuous ingestion of data and a continuous output of model? Well, we have one. NVIDIA have run some of the largest industrial supercomputers in the world and their AI factories and it completely revolutionized NVIDIA. We ingest data from a fleet of cars. We're processing it continuously, petabytes and petabytes of data, and what comes out is an AI model for self-driving cars.
We're doing that in a whole lot of different fields. And so that's AI factories.
The idea is that hypercsalers have gigantic data-centers, but these are mostly being shared by lots and lots of customers doing all kinds of different things.
But as AI models become larger and larger, requiring more and more compute, storage, memory, and networking to create economic value, there will be an increasing number of data-centers or clusters of DCs that are specialized in doing just one thing, running one or a few super-large models.
One benefit of this is that they can be optimized for that single-use, rather than have to be a bit of everything to everyone.
And then the last data center is what I described earlier at the edge.
Every single factory, every single warehouse, retail stores, cities, public places, cars, robots, shuttles, they're all going to have little data centers inside. They're all going to be orchestrated by Kubernetes. They're all going to be orchestrated from a panel from afar. They'll all run containers.
You're going to OTA new containers to it. All of them are going to work together as a fleet that's going to generate the fleet's memory. The memory is going to be constructed into some virtual world model. That virtual world model will be updated continuously. And that loop will just sit there and just run continuously.
That's a factory.
Okay. And so you got the AI factory and then you have the edge data center. These 2 data centers are brand-new. […]
the largest of all the opportunities, which I think is going to be industrial automation and putting AI literally everywhere. And so everything that moves will be automated. There's no question about it.
It will be safer. It will be easier to manage.
This edge stuff is going to help run factories and all the equipment inside and outside (supply chain, transportation). As more and more stuff is automated, you’ll want to have local low-latency AIs to run a lot of it.
It’s an interesting vision, and I think it’s hard to argue that it’s not correct in the broad outline. What will be interesting to watch unfold is the details of how all this will be implemented, and who will be the winners in these new paradigms.
Will markets fragment for a while and then consolidate into power laws, or will existing giants be able to own this transition because it’s so hard and expensive to develop all the software and hardware and developer ecosystems needed to get these things off the ground?
‘Air traffic back to around 90% of “normal” in the US’
Via Chartr (not to be confused with the cable company)
🎧 Interview: Aswath Damodaran on inflation, ROIC, FANG, and ESG ♻️
Everybody was talking about this interview by Patrick O’Shaughnessy (☘️) one after it came out. Chances are you’ve already heard it — I’m slow, but I finally got around to listening to it, and it was very interesting:
I think the takedown of ESG at the end was particularly good.
It’s one of those things where if you don’t really look into it, it’s easy to be in favor of it. But the more you learn about it and how it actually works, the more you realize that there’s a wide gap between what you could imagine it is, and what it actually is.
One thing I’d add to what Prof Damodaran said: I won’t be able to take ESG seriously as long as nuclear energy isn’t included in the ‘E’. If the goal is to be evidence-based and effective at creating a better world for our kids, and not just to play politics, and create jobs for armies of consultants and box-checkers, then it makes no sense at all to lump nuclear power with tobacco and coal.
🎧 Interview: Doug from Fabricated Knowledge on the semiconductors industry
Great discussion by friend-of-the-show and supporter (💚 🥃) Andrew Walker and friend-of-the-show Mule (you’ll always be an Asimov character to me!):
Science & Technology
🔌🔋🪫 Grid-Scale battery storage problems…
U.S. renewable energy developers have delayed or scrapped several big battery projects meant to store electrical power on the grid in recent months, scuttling plans to replace fossil fuels with wind and solar energy.
At least a dozen storage projects meant to support growing renewable energy supplies have been postponed, canceled or renegotiated as labor and transport bottlenecks, soaring minerals prices, and competition from the electric vehicle industry crimp supply.
The slowdown in utility-scale battery installations threatens the pace of the U.S. transition away from fossil fuels as the Biden administration seeks to decarbonize the grid by 2035. The delays could pose a threat to power reliability in states that already depend heavily on renewable energy, like California. (Source)
Shouldn’t we be seeing acceleration and rapidly falling prices if our civilization is going to be largely dependent on these stationary storage solutions within a decade or two? I know, probably just a bump in the road…
While in the abstract it makes a lot of sense to pair intermittent renewables like wind and solar with large stationary batteries, if you zoom out and look at the whole system, there are many challenges and downsides.
The most obvious one is probably the opportunity cost:
Battery production capacity is already tight because of the electrification of transportation. If lithium batteries (or the capital goods/machines and raw materials that produce them) are diverted to make storage for the grid, it’ll slow down the transition to EVs and/or make it more expensive. This lowers net benefits from adding batteries to the grid.
In short: batteries should go where they don’t have good alternatives to displace oil (transportation).
The grid can be cleaned up without having to build hundreds and hundreds of GWh of battery storage, a scale that boggles the mind (some would help, but the scale that some people are thinking of isn’t necessary if we don’t tie our hands behind our backs by refusing to use the most energy-dense and reliable clean energy source).
Batteries are simply not the best way to clean up the grid.
I think it’s a great idea to build wind and solar in the best sites in the world, but there’s a big difference between the phase when you have 10-20% of intermittent sources on a grid — and can make up for them going offline with other sources — and when you get closer to 100% everywhere (ie. can’t be bailed out by a neighbor who has lots of stable baseload).
Having supply swings of tens of percent on a day-to-day basis is incredibly hard to design against and makes the system more brittle, expensive, and Rube-Golberg-like.
A point that I rarely see: even if you have storage, it’s more valuable with nuclear power than with wind/solar.
With a mostly renewables grid, you need a *gigantic* amount of batteries to store days worth of that power in case of long periods of cloud cover and windless days, and you have to design around the peak demand during the year (winter/summer), so those batteries may spend whole seasons being mostly redundant (lower utilization = higher cost).
If you had a clean grid that was largely nuclear (+ hydro and medium quantities of wind/solar), adding relatively small amounts of storage to the grid would help optimize the use of your nuclear plants because you want them to produce flat out all the time to maximize ROI, but there’s less demand at night than during the day.
With storage, you can charge batteries at night and discharge that power during peak time during the day and evening, time-shifting that precious energy when you need it most.
💾 Do you know what Blender is? I promise it’s quite cool 👨💻
While I’m not a 3D animator or designer or modeler or whatever, I like to keep an eye on the tools used by the people who create all the cool stuff that we see everywhere (games, films, TV shows).
Blender is particularly cool because 1) it improves fast and 2) it’s free and open source.
The short version of the history is:
The Dutch animation studio NeoGeo (not related to Neo Geo video game hardware) started to develop Blender as an in-house application, and based on the timestamps for the first source files, January 2, 1994 is considered to be Blender's birthday. Version 1.00 was released in January 1995…
On January 1, 1998, Blender was released publicly online as SGI freeware…
In May 2002, Roosendaal started the non-profit Blender Foundation, with the first goal to find a way to continue developing and promoting Blender as a community-based open-source project.
What can you do with Blender? What *can’t* you do with it!
Check out the video above to see some of the features of the latest version, but even just the list of features and uses is endless:
[Blender is a] software toolset used for creating animated films, visual effects, art, 3D-printed models, motion graphics, interactive 3D applications, virtual reality, and, formerly, video games.
Blender's features include 3D modelling, UV mapping, texturing, digital drawing, raster graphics editing, rigging and skinning, fluid and smoke simulation, particle simulation, soft body simulation, sculpting, animation, match moving, rendering, motion graphics, video editing, and compositing.
Really cool stuff.
The Arts & History
🎸🎶 Playing around with loops 🔁
It’s the small things in life.
I love when you can see *the moment* when an artist, after a lot of sometimes-painful trials and errors, finally makes it work and how happy they are.
It doesn’t have to be some superstar writing a classic.
Even a relative beginner learning a new skill is something that we should promote and cherish more in our civilization. Here’s a good example (warning: some swearing in the video):
(if you’re curious, the song is Brazil by Declan McKenna. I was not familiar with him before this, I guess I’m not the target demographic ¯\_(ツ)_/¯ )
Young musician "Playing around with loops" is so fun/refreshing! 🤘🚀