24: My Thoughts on Peer-Based Relative Valuation, Reed Hastings on WFH, Amazon + India's Reliance ($20bn Deal?), Snowflake, and DUNE!!!

"Mr. Hastings: Twelve hours after a vaccine is approved."

Curiosity killed the cat. —Various

Ignorance killed the cat; curiosity was framed! ― C.J. Cherryh

🐈


Investing & Business

My Thoughts on Valuation Based on Peers

Earlier this week, I was reading an analyst report on a company I know fairly well, and after basically slapping my forehead a bunch because of one ridiculous claim after another, I came to the conclusion that I should consider myself lucky that I don't have to "precisely" value businesses by pulling together grab bags of "peers" that are mostly nothing like the business I'm looking at, and then comparing multiples...

I think it’s being done in good part because it’s precise (falsely) and can be done mechanically over and over again for every stock, and because others do it too, so there's safety in numbers (nobody got fired for buying IBM… oh wait), and because it sounds smart and objective until you think about it too much and notice that it's like this:

"Why is company A worth this?"

"Because that's about where company B, C and D trade"

"Why does company B trade there?"

"Because it's where company A, C and D trade"

"Why does company C..?"

“Because A, B, and D..."

The selection of the “peers” is itself an editorial choice, and you can make something look cheap or expensive just by deciding what you include and leave out. (Yes, I know relative valuation isn’t the only method used, but it’s the one I want to focus on here.)

It reminds me of compensation committees that use “peers” to benchmark against, and they all aim to be a bit above average.. And they all use each others as peers, and avoid low-paying peers, so there’s constant inflation based on this self-referential, recursive, BS metric.

So you end up with this:

Image

I’ve also noticed there’s also a lot of this going on:

Image

PCMag’s Real-World Test of Wireless Networks in the U.S.

Download speeds by major carriers over 10 years

It’s PCMAg’s 11th year doing testing of the wireless networks in the US. The drove around to 26 cities and tested both 4G/LTE and 5G. The good news is that average download speeds keep going up across the board.

If you want to go straight to the bottom line, this graph shows how Verizon won this year, and how each network ranked in various categories (average download speeds, upload, latency, 5G coverage, etc).

But most interesting was their section on 5G, which has been hyped like crazy in recent years:

So Wait...5G Isn't as Fast as 4G?

We admit it, we bought into the 5G hype. Carriers, phone makers, and chip makers alike have all been selling 5G as faster and more powerful than 4G, with lower latency. So I was shocked to see that our AT&T 5G results, especially, were slower than 4G results on the same network.

This is a crisis for marketing, not for performance. All three US carriers showed significantly higher download speeds and better broadband reliability than they did in our 2019 tests. It's just that these gains, particularly on AT&T, are largely because of improvements in 4G, not 5G networks.

The good news is, with few exceptions, you don't need to rush out and buy a 5G phone.

❄️ Understanding Snowflake & Data Warehousing Better

Those of you trying to understand better what Snowflake does (I wrote about their S-1 here), this podcast interview with Christian Kleinerman from 2018 gives a decent overview. Of course, some things have changed since, but the fundamentals havent. Via Rishi G.

Speaking of Snowflake, Tae Kim has found a filing that appears to show that the purchase by Berkshire was done by Todd Combs. Doesn’t mean he’s the only one involved, but it’s his signature…

TikTok Drama & Algorithm Briefings

So the sale process that seemed to be almost over when Bytedance announced that it was picking a winner seems to have been derailed when China added A.I. algorithms to its export control list. Now all parties to a potential deal seem to be scrambling to figure out what the politicians will allow, and how to structure a deal that would be approved:

TikTok’s Chinese parent, ByteDance Ltd., is discussing with the U.S. government possible arrangements that would allow the popular video-sharing app to avoid a full sale of its U.S. operations, according to people familiar with the matter. [...]

A number of options remain on the table, the situation is fluid and a sale is still a possibility, the people said. Even if there isn’t a full sale, the outcome would likely involve some sort of restructuring of TikTok, one of the people said.

Now you’ve even got VCs meeting with the CIA:

A group including a representative from at least one of TikTok’s major investors—Sequoia Capital, General Atlantic and Coatue Management LLC—met last week in Virginia with representatives from the Central Intelligence Agency to discuss data security (Source)

Separately, TikTok has been briefing people on its algorithm, trying to demystify a bit this controversial thing that very few non-technical people have much understanding of (so it’s easy to claim all kinds of wild things about it).

Here’s some of what the briefing revealed (nothing too shocking to anyone who knows how these type of recommendation engines work, and no real details about the secret sauce):

  • When users open TikTok for the first time, they are shown 8 popular videos featuring different trends, music, and topics. After that, the algorithm will continue to serve the user new iterations of 8 videos based on which videos the user engages with and what the user does.

  • The algorithm identifies similar videos to those that have engaged a user based on video information, which could include details like captions, hashtags or sounds. Recommendations also take into account user device and account settings, which include data like language preference, country setting, and device type.

  • Once TikTok collects enough data about the user, the app is able to map a user's preferences in relation to similar users and group them into "clusters." Simultaneously, it also groups videos into "clusters" based on similar themes, like "basketball" or "bunnies."

  • Using machine learning, the algorithm serves videos to users based on their proximity to other clusters of users and content that they like.

  • TikTok's logic aims to avoid redundancies that could bore the user, like seeing multiple videos with the same music or from the same creator.

Netflix’s Reed Hastings is NOT a Fan of Work-From-Home

WSJ: Have you seen benefits from people working at home?

Mr. Hastings: No. I don’t see any positives. Not being able to get together in person, particularly internationally, is a pure negative. I’ve been super impressed at people’s sacrifices.

WSJ: It’s been anticipated that many companies will shift to a work-from-home approach for many employees even after the Covid-19 crisis. What do you think?

Mr. Hastings: If I had to guess, the five-day workweek will become four days in the office while one day is virtual from home. I’d bet that’s where a lot of companies end up.

WSJ: Do you have a date in mind for when your workforce returns to the office?

Mr. Hastings: Twelve hours after a vaccine is approved.

For those who will point out to Hastings that vaccines don’t act that quickly, I don’t think he meant it literally, but I do think he was serious about wishing it would go that fast.

Source.

Interview: Peter Offringa and ‘Muji’

Good podcast interview of the authors of Software Stack Investing and HHHYPERGROWTH (what a name) about investing in fast-growing software and tech infrastructure companies by Austin Lieberman at 7Investing.

A lot of the discussion is around Fastly and Cloudflare and the future potential of edge computing (as an expansion of their CDN businesses). If you’re not familiar with this area of tech, Muji has a good post about it, and Peter does too.

Amazon + India’s Reliance (Potential $20 Billion Deal)?

Indian billionaire Mukesh Ambani’s Reliance Industries Ltd. is offering to sell a roughly $20 billion stake in its retail business to Amazon [...]

Mumbai-based Reliance Industries is willing to sell as much as a 40% stake in the subsidiary to Amazon [...]

“The deal would be a highly complementary one that builds on the strengths of either side,” said Utkarsh Sinha, managing director at Bexley Advisors in Mumbai. “Amazon comes with the might of its warehousing capabilities and the ability to streamline supply chains and sweat assets for maximum returns.”

Source.

Amazon: JOBS JOBS JOBS

Amazon has clearly understood that one of the things that makes it harder to attack by politicians is all the jobs it creates. And also, well, it needs lots of people, that’s not fake… From its recent “career day 2020” press release:

As a part of the nationwide virtual event, Amazon – the largest job creator in the U.S. over the last decade – will give attendees the opportunity to learn about the 33,000 corporate and tech jobs currently available across the country, and the thousands of additional hourly positions in the company’s operations network to be announced soon. Amazon will also mobilize 1,000 of its recruiters to provide 20,000 career coaching sessions to attendees in a single day


Science & Technology

Image

Physicists retort: “On the other hand, physicists like to say physics is to math as sex is to masturbation.” Source.

Microsoft Reveals More Details on Xbox Series X & S

It looks like the Xbox will be segmented primarily by resolution. Pricing for the top end Series X will be $499, but there will also be a smaller, less powerful Series S selling for $299 starting November 10.

Looking at the tech specs of both, it looks like there’s a modest drop in CPU performance for the S, using the same AMD Zen 2 but clocking it at 3.6ghz vs 3.8ghz. Where things differ more is with the GPU: the AMD RDNA 2 in the bigger X will have 52 ‘compute units’ clocked at 1.825ghz while the smaller S will have the same RDNA 2, but with 20 CUs clocked at 1.565ghz. In practice this means 12.15 teraflops of GPU power for the X and 4 teraflops for the S. There’s also 6 gigs of RAM missing from the S, the SSD is half the size (but this is because lower resolution games take less RAM and have smaller assets to store)…

All this to say: The X is optimized for 4k resolution at 60 frames/second minimum (up to 120), and the S is optimized for 1440p at 60 frames/second (up to 120 too).

Microsoft says the Series S "delivers approximately 3x the GPU performance of Xbox One," compared to an 8x relative boost for the Series X.

Here’s the exploded view of the S, just because it’s cool:

ARM’s First 64-bit Real-Time CPU (Cortex-R82)

Most of the operating systems (OS) and CPUs out in the world aren’t designed for 'real-time’ operations. What this means is that because of the complexity of modern software, they do a bunch of stuff at the same time (or often alternating extremely rapidly between things, which looks to us like parallel work) and will queue and prioritize various things, making how long it takes to execute a task somewhat unpredictable.

For some mission-critical applications, this isn’t good enough. If you’re flying a plane using fly-by-wire controls, or applying ABS brakes in a semi-truck, you need immediate and predictable response, every time. In other words: Real-time.

A lot of these applications in cars and planes and healthcare and military hardware are run on special OSes and CPUs that are designed to have that level of predictability (tightly coupled memory, dual-cores in lockstep for fault-tolerance, etc), but that comes with other trade-offs (life is trade-offs, remember?). This can mean lower performance, the inability to run full-fledged modern OSes or run the software made for these OSes, etc.

I’m explaining all this because the new ARM real-time CPU (Cortex-R82) makes some interesting progress on these fronts.

First of all, its the first 64-bit real-time CPU by ARM, which is important because you can’t allocate more than 4 gigabytes of memory with 32-bit CPUs (some CPUs have used workarounds to expand addressable memory, but still, everything else is on 64-bit now and there’s no point in going halfway).

The Cortex-R82's cores can be dynamically assigned to either real-time or applications processing—the same device might have a different profile during the day than it does overnight.

It’s also designed so that the cores can be dynamically used either for real-time processing, or switched to a mode where they can run non-real-time OSes and apps. Very clever. Via Ars Technica

Aqua Dams

Maybe it’s just me that has been out of the loop, but it’s the first time I hear of this very simple, yet seemingly very effective way to create temporary water barriers.

The tweet I’m quoting below includes a video showing someone saving their house from flood-waters with an aqua dam that they bought and installed themselves in a few hours.

Here’s the Aqua Dam website if you’re curious to learn more and see more photos.


The Arts

DUNE!!! (…Part 1?!)

The trailer to ‘Dune’ by one of my fave working directors, Denis Villeneuve, has just dropped. I have such a man-crush on him.

While looking up info on it, I realized that I had somehow missed that this adaptation was going to be split into two parts. So there’s going to be two Denis Villeneuve Dune movies! That’s twice as good!

"I would not agree to make this adaptation of the book with one single movie", says Villeneuve. "The world is too complex. It's a world that takes its power in details.""

Here’s a Vanity Fair piece on the film, if you want to learn more about it. But remember, sometimes knowing less (until after you’ve seen it) can be better…

Q: Can someone explain to me what is going on in Dune?

A: Local teen does drugs, claims to see the future. Billions die.

Stanley Kubrick’s Zoom Call

Image

Source.