493: Texas Instruments Deep Dive, Adversity, Google Waking Up, Non-Compete, Silicon Graffiti, Microsoft Phi-Mini, and Geothermal

"Agency is the key."

If you shut up truth, and bury it underground, it will but grow.

—Émile Zola

🛀💭 The reality of facing adversity tends to be pretty different from the heroic scenarios that come to mind when thinking about it.

When things are going well, it’s easy to imagine yourself being 𝕕𝕖𝕥𝕖𝕣𝕞𝕚𝕟𝕖𝕕 and 𝕡𝕖𝕣𝕤𝕚𝕤𝕥𝕖𝕟𝕥, showing 𝕣𝕖𝕤𝕚𝕝𝕚𝕖𝕟𝕔𝕖 and 𝕘𝕣𝕚𝕥! Facing challenges head-on!

We daydream of ourselves as David Goggins running on a broken leg or whatever.

🤘 NEVER GIVE UP! NEVER SURRENDER! 🤘

But actual adversity is rarely cinematic. There’s no audience cheering. It seldom feels romantic in the moment and usually seems trivial to others, because most problems are pretty mundane — but it matters to you because it’s YOUR problem.

Adversity usually comes out of nowhere when you least expect it — it can be a combination of small things that, when taken together, test your resolve and trigger moments of self-pity.

You’re not sleeping well, your basement gets flooded, you’re sick, your kids are going through something, your car breaks down, whatever. It can be the absence of things: feeling like you lack options or human connection, etc.

But this too shall pass.

You have to power through. Take action. Agency is the key.

It’ll feel good on the other side.

🤫 🌳🚶🏻‍♂️🌳 A friend recently said something that reminded me of something I wrote back in the Neoarchean era of Edition #113.

She was talking about her own practice of taking daily walks in the woods without listening to anything (podcasts, audiobooks), and of making an effort generally to reduce inputs to allow for time to think, process things, and be more creative.

I had come to a similar conclusion about myself, but I haven’t been very good at implementing it. I wrote:

I’m trying to be *more* bored lately. I'm not very good at it, partly because I don't have enough free time to do anywhere close to everything I want to do, so it's easy to always have something to fill any downtime.

But I've started to think of “no input” time as "processing time", and that makes it easier to seek out rather than try to avoid.

We can't constantly have stimulus.

Sometimes we need to be bored a bit, both to get used to the hard task of focusing on a single thing without jumping to pleasant distractions as soon as the going gets hard, but also to have space for your own thoughts, and to process what we’ve learned and/or need to find answers to.

It’s probably why every time I'm peeling potatoes or taking a shower I have plenty of ideas.

Update: I actively worked on this. Usually when I take walks in the woods, I listen to podcasts, but yesterday I didn’t.

I could really see the difference. I was thinking through things and processing ideas and coming up with all kinds of stuff (some of which are in my notes and will make it to future editions).

It works.

I won’t stop listening to podcasts — I get a lot from them too — but I think I’ll try to alternate more and reserve some “processing” time.

💚 🥃 🚢⚓️ If you’re getting value from this newsletter, it would mean the world to me if you become a supporter to help me to keep writing it.

If you think that you’re not making a difference, that’s incorrect. Only 2% of readers are supporters (so far — you can help change that), so you make a big difference.

You’ll get access to paid editions and the private Discord. If you only get one good idea per year, it’ll more than pay for itself!

---

A Word from our Sponsor: 💰 Brinker Advisor 💰

Are you ready to take control of your investments and secure your financial future? Our premium subscription is designed to empower you with the tools and insights you need to make informed decisions and maximize your returns.

🔑🔓 When you become a paid subscriber, you'll gain exclusive access to:

• Marketimer Model Portfolios I, II, and III, Income Portfolio, Active/Passive Portfolio 📊📈🏦

• BFIA Model Portfolios Aggressive, Moderate, Conservative, Tax-Exempt, Exchange-Traded Fund Portfolio 🧭

• Recommended List of No-Load Mutual Funds (updated daily) 📑🕵️‍♂️

• Full Access to All Archived Posts 🗄️📜

All this along with our favorite investment ideas and our views on the economy, monetary policy, and related topics.

⭐️ Sign up here and get 20% off your first year with the coupon code “Liberty20” (offer good through June 30th, 2024). ⭐️

---

🏦 💰 Business & Investing 💳 💴

🩻🔍 Deep Dive: Texas Instruments (and Analog Devices, kind of) 🐜🐜

Texas Instruments aka TI is a company that I’ve been following for 7-8 years.

Even within the semiconductor sector, it always stood out as particularly interesting. Maybe because analog chips are so different from the digital chips that get most of the spotlight — they process signals from sensors that interface with the outside world, like temperature, light, pressure, movement, etc, unlike the 1s and 0s of purely digital chips.

Making them is a specialized skill and analog semi engineers are doing all kinds of black magic that digital semi engineers are probably glad they don’t have to deal with. 🧙‍♂️

TI has been around forever and a lot of semi history is connected to it. f.ex. Morris Chang, the founder of TSMC, used to work at TI before going back to Taiwan and it’s where Jack Kilby invented Integrated Circuits and the handheld calculator.

But even though I’ve been following the company for a while, I still learned things from my friend MBI (🇧🇩🇺🇸):

• 📄 Texas Instruments: Leading the Analog Chips Industry

For example, I didn’t know that TI was founded as Geophysical Service, Inc. (GSI) and focused on reflection seismology for the oil industry.

Management is particularly clear-thinking when it comes to capital allocation and shareholder communications. MBI writes: “This is my 46th Deep Dive, and I think I haven’t come across any other company with a greater clarity on capital allocation than TI’s.”.

The company transformed itself based on a very clear strategy:

over the course of 2010s TI went through a bit of a transformation. Back in 2013, nearly half of TI’s revenue came from personal electronics and communications equipment. Today, that number has come down to only 20%. TI also did their last major acquisition: National Semiconductor (NSC) in 2011. At the time of the acquisition, TI was generating 22% revenue from industrials whereas 46% of NSC’s revenue came from industrials. NSC had higher gross margin (~69%) vs TI (~54%) at that time and contributed 12k analog SKUs to TI’s ~30k products at that time.

Now, industrials and automotive contributes nearly three-fourth of overall revenue of TI. With ~$20 Bn revenue, TI is the leading analog chips company in the world.

The number of products that they sell is ginormous:

TI has almost 80k products that accomplish many different things, such as converting and amplifying signals, interfacing with other devices, managing and distributing power, processing data, canceling noise and improving signal resolution. [...]

TI sells all these products across 100k customers

This is very different from Intel or AMD or Nvidia, which have a few hero SKUs that they sell for relatively high prices.

TI and Analog Devices sell tons of inexpensvie chips that get designed in all kinds of things, and once they’re in, they tend to stay around for many years since reliability tends to matter more than cutting-edge performance for these applications (does this remind you of Heico and Transdigm?).

This means that these analog chips tend to have a much longer shelf-life than hero digital chips that rapidly become obsolete and that it would take forever for a new entrant to replicate the catalog.

It’s a very profitable model (if cyclical):

In the last 10 years, TI’s gross profit increased by ~$5 Bn. During the same time, their SG&A expense was basically flat and R&D was up only by $340 Mn!! I am not sure I have ever seen almost a decade of near flat opex and yet the company was able to grow gross profit by ~$5 Bn. While TI typically expects 75% of gross profit to fall through to operating profit, over the course of the last decade 97% of incremental gross profit actually turned into operating profit! To be clear, I don’t think that is quite replicable going forward and management thinks ~70-75% of gross profit falling through to operating profit is still a reasonable assumption going forward.

I gotta pace myself, I kind of want to quote half the deep dive. You should definitely read the whole thing (link above)!

MBI contrasts the strategies of ADI and TI, which I love because it makes clearer what trade-offs each is making.

ADI made multiple large acquisitions in recent years, including some in software, while TI has been growing organically. Another difference is with manufacturing: TI does most of its fabbing and is building new 300mm wafer fabs to further lower its costs per chip while ADI fabs about half its chips and outsources the rest.

On capital allocation:

looking at the last 20 years of capital allocation substantiates that TI walks the talk. In the last 20 years, TI generated $100 Bn cumulative Operating Cash Flow (OCF). Only ~$5 Bn of that OCF was Stock Based Compensation (SBC), ~25% of it was spent on capex, ~6% on acquisitions, and kept the balance sheet largely closer to cash neutral. As a result, they were able to return ~$86 Bn cash back to shareholders through dividends and repurchases. Buyback intensity also seemed to have been rather opportunistic in the past which also helped decrease shares outstanding by ~47% during the last 20 years!

👋

Is Google Waking Up? 🥱 💰💰💰💰💰💰

One of the biggest challenges for highly successful companies is to stay lean and hungry. 🍽️

For some years, it has felt like Google was in losing that hunger and nimbleness — financial results were still great, but you could feel the company slowing down, losing focus, becoming more bureaucratic, inward-facing, and entering its dad bod era.

The rapid rise of AI and the Gemini SNAFU may have shaken management enough to force some changes, possibly because Sundar Pichai fears for his job.

As an example of this, Google Search head honcho Prabhakar Raghavan told his employees:

“I think we can agree that things are not like they were 15-20 years ago, things have changed,” Raghavan said, according to audio of the event obtained by CNBC. [...]

“It’s not like life is going to be hunky-dory, forever,” he said.

“If there’s a clear and present market reality, we need to twitch faster, like the athletes twitch faster,” he said. [...]

“People come to us because we are trusted,” Raghavan said. “They may have a new gizmo out there that people like to play with but they still come to Google to verify what they see there because it is the trusted source and it becomes more critical in this era of generative AI.”

Raghavan had some tangible changes to announce. He said the company plans to build teams closer to users in key markets, including India and Brazil, and revealed that he’s shortening the amount of time that his reports have to complete certain projects in an effort to move faster.

“There is something to be learned from that faster-twitch, shorter wavelength execution,” he said. [...]

Raghavan said Google has to address its “systemic” challenges and build “new muscles that maybe we have let fall off for a bit.” 

He praised the teams working on Gemini, the company’s main group of AI models. He said they’ve stepped up from working 100 hours a week to 120 hours to correct Google’s image recognition tool in a timely manner. That helped the team fix roughly 80% of the issues in just 10 days, he said. [...]

“The number of agreements and approvals it takes to bring a good idea to market — that’s not the Google way,” Raghavan said. “That’s not the way we should be functioning.”

Words are words, let’s see if actions follow them.

✋👩🏻‍⚖️ Federal Trade Commission Bans Non-Compete Agreements in U.S. 🛑🇺🇸

Yesterday, the FTC voted 3-2 to approve a final rule banning most non-compete agreements nationwide:

Under the FTC’s new rule, existing noncompetes for the vast majority of workers will no longer be enforceable after the rule’s effective date. 

Existing noncompetes for senior executives - who represent less than 0.75% of workers - can remain in force under the FTC’s final rule, but employers are banned from entering into or attempting to enforce any new noncompetes, even if they involve senior executives.

It’s estimated that 30 million Americans, which is nearly one-fifth of the workforce, are under noncompete agreements.

The FTC has a bunch of estimates on how this new rule would boost new business formation, create jobs, increase worker earnings and lower healthcare costs.

The specific numbers are just made up (as always), but it feels intuitive that increasing competition, labor mobility, and the rate of diffusion of knowledge across companies would result in a more dynamic economy.

Of course, this is all theoretical for now, as this rule will no doubt be challenged in court and may not survive.

---

🧪🔬 Science & Technology 🧬 🔭

🔬 Silicon Graffiti! 🐜

Ok, this is cool. Talk about a unique art for the modern age!

Microchip graffiti, also known as chip art, silicon art, or silicon doodling, is a form of microscopic artwork integrated into integrated circuits (ICs) or chips. It has been a part of the semiconductor industry for decades.

It started when over-caffeinated and/or bored engineers and chip designers began sneaking small doodles, initials, and complex drawings into the unused spaces of chips.

Before 1984, it served a real purpose, similar to how cartographers put small inaccuracies on their maps so that if these showed up in a competitor’s maps, you could prove that someone copied your stuff using this microscopic art.

The images include everything from chip designers’ names, renderings of favorite pets, cartoon characters like Dilbert, and planes, trains, and automobiles. [...]

Announcing your existence with images visible only through a microscope may seem an obscure route to fame. In fact, though, according to several chip graffiti artists, enlarged images are often displayed on the walls of people’s offices and cubicles as portions of huge printed plots of the IC. They are also visible to the designers who work on revisions of the IC [...]

Many of the images are insider jokes that only other designers or engineers would get, like puns on the names of chips or their functions. The first one that former HP chip designer Willy McAllister did was a visual pun on the adder logic circuits on an arithmetic chip around 1980. “We did full adders and half adders, little snakes with full tummies and little snakes cut in half—it’s pretty juvenile humor”

Microchip graffiti has been on the decline, as after 1984 chip masks became copyrightable and circuit design is increasingly automated, product cycles are tighter, and the risk of screwing something up is probably taken more seriously today than when the industry was young and small.

If you want to learn more, there are more details here.

🤖 Microsoft Phi-3 Mini (3.8bn): Small is the New Big

While very large frontier models are the most exciting because they expand capabilities (everyone is waiting to see what GPT-5 can do), smaller models that perform as well as previous bigger models are another extremely cool development in the world of AI these days.

They have lower inference costs (10x or more than large models) and if they’re small enough, they can run locally on laptops and mobile devices (Phi-3 Mini can run on 1.8gb of RAM).

Data quality is the main factor when it comes to making small models perform well. What’s interesting is how circular things get once you start using synthetic data.

As the name implies, it isn’t “real world” data that you find on the internet. LLMs are used to generate a ton of text (Microsoft mentions bootstrapping the model with generated children’s stories based on 3000 words, and then going up the complexity ladder). Then you filter that for quality and use the best stuff to train on.

You end up with a model being largely trained on the output of another model:

As long as the syn data is similar enough to what humans could generate, it still works — potentially avoiding a lot of low-quality data from the web, and allowing you to train on more tokens than are publicly available.

How far can this scale is an open question.

Eric Boyd, corporate vice president of Microsoft Azure AI Platform, said that Phi-3 Mini is as capable as LLMs like GPT-3.5 “just in a smaller form factor.”

How long until we have a GPT-4 level model that can run on a phone?

I’m sure Microsoft is working very hard on reducing its reliance on OpenAI, so I wouldn’t be surprised if they didn’t just stick to small models and eventually came out with a monster (Phi Mega?).

🕳️🔥🔌⚡️ U.S. Bureau of Land Management adopts “categorical exclusions to expedite geothermal energy permitting” 🎉

This is great! There’s a sun beneath our feet and we too often forget about it. ☀️

I’ve written a bunch about how I think we should do a lot more to develop deep enhanced geothermal, as it can potentially be one of the pillars of the grids of the future, providing clean baseload power around the world (if we can only get it to be cost-effective).

“Geothermal energy is one of the technologies that can move our country toward a clean energy future,” said BLM Director Tracy Stone-Manning. “It only makes sense to use the same streamlined processes for permitting geothermal exploration that other government agencies have proven can work.”

Under the National Environmental Policy Act (NEPA), categorical exclusions define activities that as a whole have been determined to pose no significant risk to the quality of the human environment. Used appropriately, these exclusions alleviate the need to prepare an environmental assessment as part of the permitting process.

The categorical exclusions adopted today apply only to geothermal exploration operations on public land. Subsequent development of a geothermal resource would require additional NEPA analysis.

It’s a good start, but NEPA as a whole probably needs to be reformed.

It makes everything take a lot longer and be a lot more expensive, which is slowing the transition away from polluting sources... having the opposite effect of what a regulation aiming to protect the environment should do.

BLM has authority to manage geothermal leasing on approximately 245 million acres of public lands (and another 104 million acres of U.S. Forest Service lands)

---

🎨 🎭 The Arts & History 👩‍🎨 🎥

🎼 Recomposing (better that than decomposing…) 🎶🤔

This is such a good album.

I wish more super-talented composers like Richter would "recompose" other greats (in this case, Vivaldi).

There's a time and place for originality, but there's also lots of greatness that can come from riffing on existing ideas and exploring the idea space of some brilliant foundation.

Jazz musicians understand that better than anyone. Most of what they do is reinterpretation of the greats and improvising around standards.