Welcome to Stuff I Thought About Last Week, a collection of topics on tech, innovation, science, the digital economic transition, the finance industry, holograms inside of holograms, and whatever else made me think last week. Please grab me on Twitter with any thoughts or feedback.

Click HERE to SIGN UP for SITALWeek’s Sunday EMAIL (please note some ad blocking software may disrupt the signup form; if you have any issues or questions please email sitalweek@nzscapital.com)

In today’s post: our new whitepaper redefines “Margin of Safety” in terms of adaptability; will the $40B radio business shift to streaming? Solar power comes to high-heat manufacturing; the low cost of EV fleets; inside a TikTok hit; blending John Malkovich; mammoth poop; and lots more below...

Stuff about Innovation and Technology
Solar Powers Industrial Manufacturing
Industrial manufacturing of materials like cement and steel requires very high temperatures and accounts for 20% of global emissions. Heliogen is a new company backed by Bill Gates that uses concentrated solar power (CSP) – an array of mirrors that focuses reflected sunlight to power a steam turbine. The system uses four cameras and software (and computing power that is greater than what was commercially available just five years ago) to constantly adjust the mirrors. Heliogen gets to 1000 ℃ now and could reach above 1500 ℃, which opens up lots of new markets. I recently read that manufacture of a single aluminum can requires the same energy as running a TV for two hours or the equivalent of burning 1/4 of the volume of the can in gasoline. While the Vox article doesn’t mention aluminum, I suspect this tech could apply to it as well, possibly creating a near-zero carbon container? I also wonder if there could be tiny Heliogens on buildings and houses to more efficiently heat water and supply heat?

EVs offer Cost-Effective Alternative for Fleets
Electric vehicles may dramatically lower the costs for fleet vehicles, a 12M unit annual market in North America and Europe. The operator of a California-based shuttle service, Tesloop, is seeing Teslas operate without degradation well into hundreds of thousands of miles driven. Fleet vehicles are typically sold off before they hit 100,000 miles. Tesloop’s Teslas operate around 6c/mile, which is comparable to gas cars; however, the all-in costs are 18-25c/mile over the lifetime of the car compared to standard gas sedans at 32-35c/mile. The EV numbers could improve even more if battery swaps become easier in the future. I think this is an underappreciated angle on Tesla’s “truck” launch this week. Some version of the Tesla truck could be the ideal fleet vehicle for maintenance and construction fleets. Tesla has logged over 140,000 pre-orders for the truck so far.

Streaming Ears and Radio Ad Dollars
At Liberty Media’s analyst day this week, Greg Maffei discussed the radio advertising market (estimated at $18B a year in the US) and how the “ear is under-monetized versus the eye” at $25 cpm (cost per 1000 listens) on podcasts. Maffei said 121M Americans listen to spoken word content daily – most of that on traditional radio. Many companies are working on building exclusive podcast content (this long Hollywood Reporter interview with Spotify founder Daniel Ek dives into that very topic). But for now, the value to podcasters lies with their audience size, and it doesn’t yet make sense to get locked in to one distribution platform. What will change that? Data. Data created the Netflix flywheel that funded their expansion into original content and the rest is history (of course I am over-simplifying – great user experience, distribution, and many other factors were at play too). Data could create an audio flywheel as well. This is true, to a smaller degree, for music, but to a larger degree for talk programming – podcasts, news, sports, local, etc. The Hollywood Reporter article mentions a $40B radio market (I think that’s global, but it’s not cited). I suspect, at an average of over four hours per day of listening, that $40B vastly underestimates the market for audio advertising content, if data can improve matching programming to consumer preferences and target fewer, but more impactful ads. Here is a fun thought exercise: for someone like Spotify to get massive data scale in a hurry on streaming talk programming, why not buy an asset like iHeartRadio and turn all of their programming into streaming? The heart of the problem is how best to achieve escape velocity on the flywheel of listening hours+talk content+ad dollars: would acquiring local advertising knowhow and talk radio content achieve it faster than the current path the streaming audio world is on?

A Voice Assistant that doesn't Gossip
Sonos has acquired a voice assistant technology called Snips that can run music-focused AI queries locally without the need to process the request in the cloud. There’s probably a good market for a smart speaker that offers privacy and hands-free operation, or could one day provide other conversational answers without going to the cloud. Sonos smart speakers are now filling 9M homes with streaming audio.

TikTok’s Quirky Authenticity Seems at Odds with Monetization
Pitchfork explains the Anatomy of a TikTok Hit. The viral Chinese music meme-ing app, which is taking US teenagers by storm, is now moving songs, often obscure, up to the top of the Billboard hits list and driving huge streams across Spotify and other apps. According to the magazine, songs that become viral on TikTok tend to have an abrasive, homespun vibe (like “something was created on the cheap in someone’s bedroom”), and are heavy on imitation, outrageous lyrics, ‘drop’, and distortion “...to work with viral dance moves, the beat has to be loud, energetic, and visceral, as if a supernatural current is jolting the dancer awake.” In case that’s not enough information: "TikTok hit songs “‘Pumpkins scream’ and ‘Asshole’ gathered momentum, in part, because of e-boys—chain-sporting, black nail polish-wearing, Timothée Chalamet-types who serve as a digital update of previous generations’ goth mopes. They film their TikToks in dark bedrooms lit by colored LED lights, where they pout, mimic choking themselves, and roll their eyes while tapping their temples.” On the monetization continuum for social networking sites, Facebook and Instagram are the easiest followed by Twitter, and then there’s a big gap to Snap (which struggles to generate revenue above its cost to operate). And then there’s TikTok, which is so organic and viral it seems impossible to effectively monetize without destroying its authenticity. For now, TikTok is just a machine that turns money from its parent company Bytedance in China (Bytedance is now the 2nd largest advertising platform in that country) into ads on Snap and Facebook. This NYT fluff PR piece on TikTok says it remains independent of Chinese control, but it’s highly likely CFIUS (committee on foreign investment in the US) will demand it be separated from Chinese control for security reasons.

FedEx CEO Challenges NYT Publisher to Debate
I often get pushback for saying FedEx has no reason to exist (SITALWeek #211), but this NYT article tried to one-up me with its criticism of FedEx for passing tax cuts onto shareholders last year instead of investing in their business. FedEx faces significant competition as the package business evolves into vertically-integrated, locally-sourced (shipments originating in local cities rather than being flown/driven cross-country) platforms. The upstream sorting and other advantages the FedEx network had in the past are rapidly evaporating. FedEx responded to the NYT article, which I don’t believe they actually read, by challenging the NYT’s controlling family to a tax debate - maybe they could have a tax duel at high noon? 🤣

RBC Takeaways: VCs Cooling their Heels, Data Advantage, Transportation as a Service, and iBuyer Liquidity
As I read through transcripts from the RBC Technology Conference this week, a couple of things stood out. The freezing (or chilling) of private capital is likely to help companies that are already public and have a path to growth and profits (see below for real estate and ride sharing examples). Another theme was data advantages: I think that we will see more and more vertical integration as the Internet infiltrates more and more legacy, 20th-century businesses; a key driver of that vertical integration is privileged access to data. For example, Zillow gets 100M signals a day from its consumer portal that can be used to understand the housing market on a very local, real-time level. (See also data comments on Spotify above.)

The average car in the US costs $9,000/year to own and operate, which provides a significant pricing umbrella for transportation as a service. Speaking at the RBC conference, Lyft is also seeing availability of drivers as less of a gating factor today (which is interesting given the tight labor market and concerns over wage inflation) – which allows them to raise the bar on cars to require built-in advanced safety features, which improves rider safety and lowers insurance costs. (Also lowering insurance rates is Uber’s use of video recording, as the NYT reports.) Lyft remains focused on harvesting share in verticals, like healthcare and business, while Uber is taking a broader, horizontal “super app” approach.

RBC Internet analyst Mark Mahaney interviewed Zillow CEO Rich Barton, who discussed the “buy box” for the iBuyer market getting coverage to over 50M homes in the US in the future. Next year, Zoffers will be live in 26 markets, allowing Zillow to shift their large marketing budget to the product. Over the last 20 years, the average time homeowners stay in one home has ballooned from six years to 13 years (discussed in SITALWeek #218). Could new liquidity from the iBuyer market reverse this trend? If it drops back to <10 years, the market for iBuying and other real estate services will boom. With the benefit of the Zestimate and its large audience, Zillow is moving into pole position for iBuying just as the startup funding environment is likely to keep new entrants at bay. A lot of this applies to Redfin as well, who might even have built a better mousetrap today than Zillow – to be determined!

BERT Trending, but too Energy Intensive?
BERT (Bidirectional Encoder Representations from Transformers) was a hot topic at the Super Computing conference in Denver this week. NVIDIA explained BERT and its massive demand for GPU-accelerated inference of conversational AI on their earnings call last week: “Google's breakthrough, introduction of the BERT model, with its superhuman levels of natural language understanding, is driving a way of neural networks for the language understanding...training these models requires V100-based compute infrastructure that, in orders of magnitude, beyond what is needed in the past. Model complexity is expected to grow significantly from here.”

But, can the planet afford the energy burden of you talking to your BERT smart speaker? The Guardian states that the latest NVIDIA natural language model was 24x larger than its previous version yet only 34% better – and training the model took nine days on 512 V100 chips at an energy cost equal to the yearly usage of three Americans.

Homegrown Semiconductors Remain a Dream for China
Junko Yoshida at the EE Times went digging to see how successful China’s “Big Fund” – of nearly $30B to spur the domestic semiconductor industry – has been. “Thus far, results from the first round are mixed. Pressed to name the Chinese semiconductor champions the Big Fund helped build, most executives have no answers.” As I’ve mentioned a few times recently – the resolution of the US-China trade war is heavily dependent on the reality that China will not be able to create a viable, leading-edge chip capacity for decades.

Machine Learning Comes to Semiconductor Design
Google’s head of AI Jeffrey Dean posted a paper discussing how machine learning could dramatically change the process of chip design: “By having a reinforcement learning algorithm learn to ‘play’ the game of placement and routing...it may be possible to have a system that can do placement and routing more rapidly and more effectively than a team of human experts working with existing electronic design tools for placement and routing.” As Moore’s law slows and the IoT explodes demand for all types of chips, this could greatly accelerate innovation and dramatically increase the scale of the semiconductor market over the next couple of decades. Chips designed with the aid of machine learning are likely to advance artificial intelligence faster than many other efforts underway today.

Ex-Amazon Employees Carry Culture to Seattle Startups
This article in the Seattle Times, which interviews Amazon veterans who now have local startups, discusses the difference between having leadership values and actually living them. “The leadership principles at Sears, as articulated, were almost identical to the leadership principles at Amazon.” This ability for values to actually be in the blood of an organization is very difficult to accomplish. One of the key things that Amazon’s values maintain is speed: most companies grind to a halt as they grow; fighting that inertial weight is a challenge for every company.

No Shoes, No Shirt, No Cash Register
As restaurants move to turn their locations into delivery/pick-up only, their tech stack changes a lot. In particular, with no need for cashiers, there is no need for cash registers, point of sale systems, etc.

Miscellaneous Stuff
Holographic Information Escapes from Black Holes
We generally think of a black hole as a place where gravity is too strong for anything, including light, to escape. But, there is another way physicists view them: as a holographic projection of 2D quantum waves with no gravitational force. We know information escapes black holes (known as Hawking Radiation), and this new work suggests it might do so via a hologram within a hologram, which projects out of the black hole back into our slightly more understandable part of the universe.

Blending John Malkovech
John Malkovech has been blending cabernet and pinot noir grapes at his Southern Rhone vineyard in France. It’s a combination of two grapes out of their original homes that, traditionally, are never mixed. “It’s like any other form of self-expression,” he says. “I have very specific taste in wine. These are reflections of what I like.”

Winter Weather may have Spurred Feather Evolution
Dinosaurs near the South Pole had downy feathers to keep warm. The discovery is the first to suggest that dinosaurs and early birds lived in regions of extreme cold with “months of polar darkness”. As Jamie Woodward said on Twitter this week: “When you hunker down tonight in your warm cosy bed - spare a thought for those feathered dinos in the polar south 100 million years ago working hard on the prototype for your duvet filling.”

Ancient, Frozen Bacteria Get New Lease on Life
Scientists have revived 40,000 year old bacteria from mammoth poop in an effort to glean potential new antibiotics.

Avocado Cartels are Deforesting Mexico and Terrorizing its Population
The LA Times reports on the bloody cartel war for Mexico’s “green gold” avocado industry.

Marie Kondo’s Crafty Retail Scheme
Marie Kondo rose to fame largely through a Netflix documentary that said to throw away all your crap that doesn’t bring you joy (I haven’t seen it). Now she’s gone into business to sell you crap to fill the void left by all they crap you threw out. One of the items is a tree branch, er, sorry, a Shiatsu Stick, being sold for $12.

Stuff about Geopolitics, Economics, and the Finance Industry
A new whitepaper from me and Brinton: Redefining Margin of Safety. The classic Ben Graham term gets a fresh analysis as we think about adaptable companies and their ability to slow down time. And, we couldn’t help but explain how special and general relativity can help us understand this concept. Below is a summary/excerpt, and the entire paper can be read here.

Ben Graham posited the idea of “margin of safety” in The Intelligent Investor back in 1949. At its core, the concept is about investors’ inability to predict the future. We heartily agree with Graham – we don’t believe it’s possible to accurately predict the future with any meaningful degree of precision. Instead, we’re always looking for situations where we can make predictions with the broadest range possible, thus giving us higher odds of being correct.

Over the years, however, investors seem to have placed too high an emphasis on valuation for determining margin of safety, and tend to use the concept to justify ownership of dying businesses. As explained in Complexity Investing, we’ve found that a company’s ability to adapt is a much stronger indicator of a wide margin of safety. We’ve always observed a strong correlation between adaptability and quality management; but, as we explain in the new paper, adaptability doesn’t seem to be determined exclusively by the management team per se, but also by the time constraints under which the management team has to adapt, which is governed by the speed of their business. Specifically, slow, long-duration growth allows for timely innovation, decelerating the game clock so managers can make smart decisions and maintain their lead through adaptation. Hallmarks of gentle sloping ‘S-curve’ businesses that we look for are negative feedback loops, tight-knit customer relations, and positive NZS.
✌