39: My Thoughts on Nvidia CEO Interview, The Trade Desk, Visa & Mastercard, Liberty Global, and Cool Birds
“My mother still has no idea what I do,” he said.
Communication usually fails, except by accident. —Osmo Wiio
Here’s something I recently wrote to a reader of this newsletter who sent feedback (Hi M!), and I thought I’d share it here:
“I often try to remind myself that behind all the twitter avatars and email addresses are real people all around the world. Isn’t modern communication technology amazing?
Our brains aren’t very good at understanding what is really going on behind the pixels on screen, but it still blows my mind that I can be sitting at home in pyjamas, typing stuff in the computer, and moments later all kinds of people from around the world, with all kinds of backgrounds and life experiences and expertise, are reading these words — even reading itself is cooler than we realize, because we’re so used to it. It’s pretty close to telepathy: I think of stuff, and now these words are INSIDE YOUR BRAIN. haha”
→ If you’re a US citizen, please vote. ←
Investing & Business
A Word From Our Sponsor
One issue most are familiar with is the usability (or rather inaccessibility) of data sets. Please join Sentieo, a fundamental research platform and a challenger to the legacy terminals, and alternative data provider Earnest Research as they go over several companies to demonstrate the combination of alternative data, NLP and ML applications.
The Trade Desk vs The Rest of Advertising
It’s almost like there’s a law of conservation of market cap in the industry, and as the others melt, TTD is picking up what others are dropping.
Good line on how complex the space is:
Jeff Green, chief executive of the Trade Desk, admitted it is still hard to explain the business model to people unfamiliar with the industry. “My mother still has no idea what I do,” he said (Source)
Graph showing how digital advertising has been growing:
Everything else getting squeezed between it and TV (btw FT, what’s the deal with the bottom part of the chart? Why is there this woobly negative space above the X axis? Is this the unlabelled “others” category?).
“The Trade Desk is a true platform in that it enables the agencies to build their businesses on top of it, not just technologically but financially,” he said. Some holding groups have looked at ways to reduce reliance, but have in the past struggled to develop in-house tech capability.
Laura Martin, an analyst at Needham, said The Trade Desk has “re-intermediated” the agencies as middlemen by offering them volume discounts. Brands have a reason to buy advertising through agencies because it is cheaper, while agencies “do not feel threatened” by The Trade Desk because it is protecting their place in the supply chain. (Source)
Interview: Jensen Huang, Nvidia CEO (50 minutes)
Before I go to the interview, a really good line from Jensen at a recent investor presentation:
Nvidia is an IP company because I'm pretty sure that TSMC makes the chips, and I'm pretty sure that what we delivered to them was effectively an e-mail after completion of a multibillion-dollar project.
Let’s go through some highlights of this good interview with Nvidia CEO Jensen Huang at The Next Platform (video above). They discuss the ARM deal about 11 minutes in (I wrote some of my thoughts about it in edition #26).
Q: Why buy the cow? When you can get the arm license for almost free?
A: Nvidia is not a chip company Nvidia. Nvidia from day one has been an intellectual property company. We're not a chip company no more than Microsoft is a CD-rom company.
You can tell because you you know what all our people are doing. They're all typing […] our population of engineers, they're not manufacturing people, you know, semiconductor manufacturing is very complicated, it's deep science, but that's not our field. We type and we create IP.
In a lot of ways, we’re exactly like ARM, we’re two IP companies. One of us has IP that is ‘hardened’, so it’s easier to deploy… The other has ‘soft’ IP, so it has the benefit of greater reach.
Then there’s a great joke about zero nanometer chips, but I won’t ruin it by trying to type it.
Both Jensen and the interviewer agree about the end of Moore’s Law, and they point out the very important fact that even if we’ve been shrinking transistors lately (5nm node chips are coming out, like the new Apple A14 in the iPhone 12 and new iPad Air, fabbed by TSMC), the cost per transistor has been going up and the performance per transistor has been going down (it’s so expensive to develop the techs and build these new fabs, almost nobody is left standing at near cutting edge: TSMC, Samsung, and Intel, basically, and Intel has been struggling).
To bring it back to ARM, Jensen makes the great point that wherever the wall of Moore’s Law is, ARM is going to have a longer runway there because it’s so energy-efficient to begin with.
In other words, within a certain thermal envelope, you can pack a lot more ARM transistors and thus probably achieve either better absolute performance per watt, or lower energy-use for equivalent performance to the competition. That’s very important since power and heat management are huge costs and design constraints in any datacenter.
Their DPU chips (and the Mellanox acquisition) ride on a similar logic. When you accelerate a lot of the data moving around in a datacenter, you not only gain performance, but you also offload it from chips that would do it less energy-efficiently (ie. a general purpose CPU can do that work, but it’ll use more power and cost more per operation than a dedicated SmartNIC chip).
Here’s Jensen on Nvidia’s efforts to optimize software in various industry verticals to be accelerated on their chips:
It's a full stack problem. It's a domain specific problem.
We go into computer graphics, from top to bottom, we refactor the algorithms and we refactor the code so that it could be acceleratable.
We go into a life sciences, computational chemistry, computational biology and we refactor all the code. We go into robotics, we've refactored the code.
We go into data analytics. We have to refactor the code… Notice that it took us until Spark 3.0, with hundreds of people working on Spark, to be able to accelerate Spark and the HPC solvers that are at the heart of some of these giant simulations are also now being refactored to use the same math units that AI software uses. [...]
Jensen makes the point that in the end, it’s the lowest cost that wins… Not the “cost for the thing”, but the “cost to do what you want to do”.
So it’s more than the price of a product, it’s about the price per result of what you’re doing. If you can replace a whole data-center worth of compute with one Ampere node, the fact that the node itself is pretty expensive doesn’t matter if its utility is equivalent to many many cheaper nodes that, combined, cost more (upfront and to operate, because a big part of total cost of ownership is electricity, maintenance, etc).
Jensen seems to see ARM well positioned on that front to invade the data center (the interviewer makes the point that historically it’s technologies going up from the bottom that win, while Intel is attempting to go down from the top to compete with ARM).
On the ARM deal, this answer is telling on how big an opportunity he sees:
I could realize all my hopes and dreams, and our company can realize all of our hopes and dreams, without ARM. This opportunity that came up because Masa and Softbank were interested in selling ARM. It was like, my mind exploded. It was that good. It was that good, it's a once-in-a-lifetime opportunity [...]
This is an ecosystem… This is a company, this is a team that won't get built again. This is a 30-year journey. It took 30 years to get here. It will not happen again. It will not happen again. Something else will happen, but it will not be this right.
Hard to phrase it more strongly than 🤯
Jensen doesn’t seem to be a big fan of FPGAs (AMD’s is apparently in “advanced talks” to buy FPGA maker Xilinx):
Q: Do you need an FPGA?
A: No, no, not at all. I believe that chip designers ought to design chips because, and the reason not not because it's a work that should be protected, but because it's unpleasant work. You know, nobody woke up in the morning and say, you know what I wanted to do more than anything is to to take three years to make my algorithm work. So as much as possible, we want to be able to do it with with software.
Of course, there are many algorithms that don't fit well with CPUs and that's the reason why FPGAs exist…
But whenever you can, you want to do everything software-defined, you want it to be programmable, and the combination of a programmable CPU, a programmable GPU, a programmable DPU, and on top of it this tensor core AI processor, that's a universal function approximator that we established between those four processors. I have a hard time, thinking of of a reason to do an FPGA, or another way of saying it is, the number of reasons is reducing over time.
Here’s another interesting way to think about where to allocate scarce company resources (R&D):
Jensen: I could give you the the the algorithm in our company for for what to do […] this is a a universal universal algorithm that we apply.
We will do work that is incredibly hard to do that nobody else can or should do, or want to do. If other people want to do it, we would more enjoy to let them have it.
Notice we don't fight for share in almost anything, we always go and build original work, and the reason for that is because we don't want to squander the resources of our company and we want to make a unique and fundamental contribution to to the industry, and so we would highly prefer to do work that the world's never done before. And it is the case that there's plenty of work that that's never been done before, then we're richly employed.
How to measure success at Nvidia?
The metrics we're trying to optimize for is the number of developers, the richness of our ecosystem and, ultimately, the number of applications we accelerate. Those are the KPIs for the company. Those are indicators of future success.
And to finish on a totally pointless and trivial note (but hey, I notice things), I like that Jensen’s kitchen is actually like it was in the Nvidia official keynotes this year, with the now famous spatulas. There’s even a really cool bonsai tree:
Thinking Back on Liberty Global
Liberty Global (LBTYA) is an interesting case study of a stock/company that just never worked after 2015 despite all the various theories (some very detailed) on why it should turn around...
Looking at notes, bought in 2014 & sold 2017. Owned it longer than I should've, but like classic value traps, it seems too cheap to sell, so you hold out for a better price that never comes.. What did it for me: lost confidence in management. Things they said would happen just didn't happen.
The actual loss wasn't that big in dollars (about 10%), but the time wasted/opportunity cost was really high.
I think I would've soured on it a lot faster if the Malone halo wasn't rubbing off on it.. The whole TCI 2.0 (and LiLAC was supposed to be TCI 3.0) narrative, Malone saying Fries was the best CEO he knew, etc..
Visa & Mastercard Volume Growth Since 2010
Drowning in SaaS Apps
Science & Technology
Pranked Engineer Does What Was Supposed to be Impossible
Marvin Pipkin’s 1925 fool’s errand at General Electric:
When the war ended, Marvin was hired by [General Electric’s National Electric Lamp Company (NELA Park)] Park as a chemical engineer. As the story goes, in 1925 he was given the task of creating a strong, inside frosted light bulb. This task had almost always been given to every new scientist or engineer who first came into NELA Park, by all accounts as a practical joke because no-one ever believed that an inside frosted bulb could ever be created. And of course, none of them had ever succeeded. But Marvin, despite the odds and to the astonishment of his fellow employees, hit upon the solution after several weeks of trial and error. [...]
He first started with the method used by many of the scientists before him - spraying acid inside the bulb to etch its interior. [...] any object… that has a sharp fracture or cavity in it such as a crack or a notch, by nature then has a weak spot. And this spot is the weakest link in the object’s chain of strength. As such, the object will easily break if enough stress is applied to it [...]
The previous scientists and inventors who had earlier attempted to make an inside frosted bulb had always and erroneously concluded that a second application of acid to the inside would merely weaken the bulb even further. But not Marvin. Here is where the Marv’s next process of inside frosting comes in. To re-strengthen the bulb, Marv created another acid solution, this time with a different mixture of chemicals. The application of this new acid mixture to the interior merged the tiny sharp crevices into smooth dimples that re-strengthened the glass. [...]
Marvin was a showman, apparently:
The story continues that Marv decided used a little bit of his dry wit to demonstrate the bulb’s strength to his manager [...] holding six inside frosted light bulbs in his hands, some having been treated with just the first application of acid and the remaining bulbs with the subsequent strengthening treatment.
The first few bulbs, again treated with just the first acid application, shattered very easily as expected, scattering broken glass on Enfield’s desk. However, the remaining bulbs with the second, strengthening treatment, easily retained their structural integrity when tipped over and remained unbroken. But that was not the end of Marv’s demonstration.
Marvin then took the unbroken bulbs and, in a confident but seemingly cockeyed manner, dropped them straight to the office floor. The bulbs did not break but remained intact, bouncing several times. Marvin had successfully created a strong, inside frosted light bulb with nearly perfect light diffusion, another milestone in the history of lighting. And in spite of the broken glass on his manager’s desk, Marvin was not fired but instead was subsequently awarded the time-honored and prestigious Charles Coffin award for his discovery
The Arts & History
I Just Like This Woodpecker…
…and this Red-tailed Black Cockatoo
I may as well make it a special bird edition!
A female red-tailed black cockatoo. The species is sexually dimorphic, unlike most parrots, and only the female has that "starry night" thing going on. Source
58th Cuban Missile Crisis Anniversary
October 16 to October 28 is the 58th anniversary of the Cuban Missile Crisis, a very dangerous few days in modern history, an inflection point where things could’ve taken a turn for the worse even thought none of the parties involved wanted a nuclear war. (Time for some game theory, right?)
“The Destroyer of Worlds” podcast by Dan Carlin at Hardcore History covers this, as well as “the dangerous early years of the Nuclear Age and humankind’s efforts to avoid self-destruction at the hands of its own creation”. I haven’t heard it yet, but based on the quality of what I’ve heard from Carlin in the past, I can pretty much recommend it blind.
h/t Mike Dariano