520: Intel the Next BlackBerry?, Gavin Baker Redux, Music Streaming is Dumb, Game Design Lessons, Palmer Luckey, and Cultural Tutor

"Today, they don’t have the best of anything"

Kinser's Law: Around the time you finish doing something, you know enough to start.

🏄‍♂️🧠 I bought a balancing board.

It’s a very clever design. There are stoppers on the underside that attach with magnets. You can move them around to increase or decrease the difficulty level, and it comes with both a roller cylinder and a ball.

I figured that it could help with some of what is missing from my workouts. While I do strength training with weights and running for cardio, I’ve been neglecting stability work. This board seemed like a fun way to challenge my nervous system + all those little stabilizer muscles.

🎮🤖🔮 This is a follow-up to Edition #519 where I told you about the “hallucinated” Doom game engine. This video goes deeper into what is going on and why it matters beyond the 1993 game:

📚📖🚮🗑️ Abandoning bad books is good training for dropping bad ideas more generally and avoiding the sunk cost fallacy. The net value of this practice is higher than just the recovered opportunity cost.

Start more books AND stop more books! The upshot is that you’ll read more AND be happier with what you do read.

🛀💭 What amount would someone have to pay you to give up Internet access entirely for a month? 💰

What if it’s a recurring deal that you can take again every month, but it stops as soon as you refuse once — how long do you think you would last? ⏳

I’m not sure what my answer would be, but just thinking about it is a good reminder of how much value I’m getting from being online, especially compared to how much access costs.

💚 🥃 🚢⚓️ You recognize that ideas are often undervalued, even among those whose job it is to make decisions, solve problems, or have the right idea at the right time.

Consider this: Even just one good idea per year can be INCREDIBLY valuable.

The impact of ideas extends far beyond the financial: Making better decisions with your spouse or about your kids or health can have huge lifelong positive effects. 💪

This newsletter provides exploration as a service. We can’t know in advance what will be useful, but we can have fun learning and thinking things through together.

Your next favorite thing might be just one newsletter away. Thank you for your support! 💡

---

🏦 💰 Business & Investing 💳 💴

🛑💣🏚️🙈 Is Intel the next BlackBerry? 🤔😬

Or RIM, as they used to be called. You know what I mean…

It has been reported that Intel is working with bankers to explore “strategic” options to help the company recover from its tailspin (with the ground rapidly approaching…).

Recently, they’ve cut the dividend, laid off 15,000 employees (out of about 125k), reduced 2025 capex plans by 17%, and they’re apparently considering selling their Altera FPGA unit along with other assets along with further reducing capital spending plans.

For anyone who has studied the history of Silicon Valley and the tech industry, this is a significant fall from grace. Intel used to have the best chips, the best R&D, the best manufacturing, the best management, the best distribution, the best brand, the best profits…

Today, they don’t have the best of anything that I can think of.

Not mobile chips, not desktop CPUs, not gaming GPUs, not data-center x86 chips, not R&D, not manufacturing… and certainly not profits!

Even as AMD is eating its lunch in x86, new competitors are encroaching on x86 with ARM-based chips and GPUs, reducing the relative importance of their bread & butter.

Gelsinger is a smart guy, but he may be too late. The aircraft carrier was set in this direction a while ago and changing course takes time — something they don’t have much of.

When MBI (🇧🇩🇺🇸) and I had a conversation about the semiconductors industry in July, we came up with the idea of a consortium of big tech to try to kickstart a US-based alternative to TSMC. I really wish something like this would happen.

These players all have the most to lose if something were ever to happen to TSMC, and they have the resources to underwrite such a venture — in fact, it would be a cheap insurance policy for them and they’d do a better job than the government could:

🤔 Mag7 Semiconductor Insurance Policy Proposal 💰💰💰💰💰💰💰🏗️🛠️🐜🐜🐜🐜

TSMC is the most mission-critical company on Earth.

If we imagine an inverted pyramid with TSMC at the bottom, there is so much value in the world resting on it. In other words, if it were to disappear, pretty much everybody would be hurt.

The US Big Tech companies — known as the 𝓜𝓪𝓰𝓷𝓲𝓯𝓲𝓬𝓮𝓷𝓽 7 these days — have the most to lose because they are *heavy* users of semiconductor and they are incredibly profitable.

Because these trillion-dollar companies touch so many other parts of the economy, if they’re hurt, there will be plenty of pain to go around.

So you have trillions and trillions of dollars in market cap (and I mean that in the broadest sense of the term, also applying to private companies and governments, national security, etc) resting on one company.

It seemed to us that it should be a no-brainer for the Mag7 to create a consortium to try to create an alternative to TSMC. They could do a better job than the government trying to sprinkle subsidies and tax breaks.

The quickest path may be for Intel’s foundry business to be spun off and maybe merged with GlobalFoundries (AMD’s old fabs) to get more of that third-party customer service DNA injected in the culture.

Mag7 consortium members could start by chipping in $5-10 billions each and lobby the government for a streamlined and fast-tracked regulatory regime on national security grounds. Stuff needs to move quickly, not take 10 years in paperwork review.

Semiconductor talent from around the world could be attracted with higher-than-standard compensation and equity in the new startup that would have the potential to be the next TSMC.

That’s NOT an expensive insurance policy when you consider how many trillions of dollars would be impacted by a long-tail event taking TSMC offline (and sadly, that’s not a far-fetched scenario. China’s leadership frequently discusses the invasion of Taiwan).

The opportunity cost wouldn’t be prohibitively high since all these companies have net cash on their balance sheet earning a few percent.

Back to BlackBerry.

Intel was bigger and more dominant than BlackBerry ever was, but the metaphor I’m going for is that if something radical isn’t done, Intel appears to be on a course to become a living fossil like BlackBerry: A company that still exists but has lost relevance and is but a shadow, a ghost of its former self. 👻

Maybe they deserve it, but *we* don’t.

It’s too many eggs in one basket to have the entire world economy be dependent on a single company located in an unstable part of the world. 🧺🥚🥚🥚🥚🥚🥚

We need more diversified sources of cutting-edge semiconductors.

🔮🤖 Gavin Baker Follow-Up on the Future of AI 🐜

This is a follow-up to what I wrote in Edition #519.

Gavin writes:

the ROI on AI has been positive thus far. This is undeniable. ROIC at the largest spenders on AI has gone up *significantly* since they ramped their datacenter/GPU capex spend last year. Most of the ROI on AI thus far has come from improved advertising targeting and creative leading to higher ROAS for customers. The higher customer ROAS has driven increased spend and thereby higher revenue growth for the largest internet advertising companies. AI underpins both Performance Max and Advantage Plus, which have really ramped over the last year.

When thinking about AI workloads, it’s easy to focus on what’s most visible — AI startups and consumer-facing products — but a lot of that capex has gone to less-visible but very profitable parts of the stack.

Any improvement to Google and Meta’s ad targeting and content recommendation/engagement moves the needle! 🪡

scaling laws may eventually collide with economics *if* they hold. And no one on Planet Earth *knows* if they will hold. Almost everyone closest to AI believes they will. It sounds like Kevin Scott has seen early checkpoints of GPT5 which confirm this, but no one knows. If scaling laws hold and there are no fundamental technology breakthroughs, then the models trained in 2026/2027 are likely to require clusters that cost well over $100 billion. Multiple players are discussing 5-10 gigawatt clusters - that is easily into the hundreds of billions of cost for a single cluster. At this point, I think scaling laws may collide with economics, ROIC *may* begin declining and there might be an ROI on AI debate that is underpinned by reality rather than ridiculous speculation.

5-10 gigawatts!

Reliable electricity is likely to be one of the main bottlenecks to scaling. 🔌

NVLink is the single most important technology and product at Nvidia

Giant training clusters with 100k+ GPUs (growing to millions in a few years) need to be tied together with incredibly fast networking that can handle ridiculous amounts of data being transferred.

If networking can’t keep up with GPUs, that’s another likely bottleneck (or if not a complete show-stopper, it would make things even more expensive by reducing the efficiency of training).

the Xai 100k cluster is the first true test of scaling laws since 20-30k H100 clusters were stood up roughly a year ago. This should lead to a super impressive Grok3 if scaling laws hold and Xai is likely several months ahead of other players who were waiting for Blackwell but are now almost certainly frantically trying to figure out how to make 100,000 H100/200s coherent. Really hard to do btw, but the attempts might buffer any Blackwell delay. I think it is highly likely that scaling laws hold and Grok3 is the best model in the world by a wide margin barring a "strawberry" miracle.

Wouldn’t that be something if Musk was able to leapfrog ahead of the competition (again) by having outsized ambition, risk tolerance, and ability to execute rapidly?

If he does, it’ll be interesting to see if he can monetize it, and if having the best model is enough of a pull that they see significant adoption of Grok3 during the period until someone else catches up.

One obvious thing to do for them would be to uncouple Grok from Twitter/X, but that may not be something Elon wants to do if he’s trying to salvage some of his investment in Twitter (the last mark from Fidelity values it at $15/share vs the $54.20 that he paid, or down 72%).

🔊🎶 Music Streaming Platforms are Too Dumb 🎸🎤🎺🤖🧠

Music streaming services often seem like they are built by people who aren’t music fans. A lot of obvious features don’t exist.

I’ll shout some of my ideas into the void. Hopefully, someone in the bowels of Spotify or Apple Music (or some startup that will be acquired by them) is already working on this.

For context, here’s what I used to do 15 years ago in iTunes.

I would create a smart playlist that updates in real-time. Using the built-in boolean logic, I would tell it to shuffle tracks that:

• I have rated 5/5 stars ⭐️⭐️⭐️⭐️⭐️

• That I HAVEN’T listened to in at least 3 weeks 📆

• I would also filter out a few genres that I generally don’t want to have pop up randomly, like death metal and grindcore and harsh industrial music or whatever

The result was that I had a personalized “radio station” that only played music that I really liked and that I hadn’t listened to in a while. If you have a large library (thousands of albums), it’s an amazing way to resurface things you wouldn’t spontaneously think of playing.

I get it, few people used iTunes boolean filtering for playlists, so they haven’t exactly been developed since (in fact, they got worse over time). But with today’s technology, there’s no excuse. We should have much better tools to slice and dice the abundance of music that streaming brings, and because of natural language processing, we can make it so easy anyone can do it.

Spotify and Apple Music have endless playlists and algo-curated channels based on your library and what tracks and albums you “like”, but these are crude. They do an ok job, but leave a lot of great music experiences on the table.

If I spend a day writing while listening to instrumental music to concentrate (Mammal Hands and Portico Quartet, f.ex.), then when I fire up my personalized algo-station later, it’ll suffer from recency bias and mostly give me more of that. But maybe it’s the evening or weekend and I’m in a different context and am looking for something else.

There’s currently no easy way to tell it what I want. But in the post-ChatGPT world, it should be possible for me to tell my music service: “Hey, I want upbeat music from my library, but only stuff I haven’t listened to in 6 months, and no instrumental music” or “Give me things that have a Nick Cave vibe, but don’t play Nick Cave”.

I should be able to create smart playlists by just *describing* what I want and have an LLM parse that and create my unique playlist.

I should also be able to iterate on the fly with the algo and tell it when something is just a temporary signal and when it’s a deeper preference (by that I mean that when you thumb up or thumb down 👍/👎 a track, it can mean “not right now” or “I hate this, never play it again”).

There’s always the danger that you skip a track because you’re not in the mood, but the algo decides that you don’t like that kind of music and gives you less of it in the future against your actual preferences. That’s a failure, and it would be great to say “I’m not in the mood for that tonight, but I still like that artist” and have the algo understand.

Because of our crude algos, millions of people are missing out on music they’d really like just because their Spotify profile got painted into a corner for some random reason…

---

🧪🔬 Science & Technology 🧬 🔭

🎮 Learning from Game Design 👾⛳️

There’s so much to learn from how games are designed.

The video above is ostensibly about one feature of first-person shooters (aim down sights), but the *way to reason and think through* whether this feature is good or bad, or appropriate or not for certain specific games, and what makes it work or not depending on context, is the kind of thinking that we should all be doing more of in our lives.

🥽 Palmer Luckey: From Underground Videp Game Vaults to Defense Tech 🏰

This profile of Luckey has a bunch of interesting facts about his backstory and interests.

To be clear, the most interesting things about him are what he’s doing at Anduril and how that’s changing the defense ecosystem in the US, but I thought the human-interest anecdotes also had something to teach:

Luckey is the owner of the world’s largest video game collection, which he keeps buried 200 feet underground in a decommissioned U.S. Air Force nuclear missile base

Or this one:

Luckey tried his hand at building a nonprofit private prison chain that only gets paid when ex-prisoners stay out of prison. After he decided that would require too much lobbying work, he attempted to solve the obesity epidemic by making food out of petroleum products centrifuged out of the sewer system—a perfectly delicious and low-calorie idea, he maintains, which he only ditched because of the “marketing nightmare” of persuading people to eat remanufactured sewage.

Y’know when they tell you that founders may seem a bit eccentric or have strange ideas or nerdy passions, this is the kind of stuff they mean.

“These days they’d probably say I had ADD,” Luckey told me at his home in Newport Beach, sitting at his makeshift Dungeons & Dragons table littered with Sonic condiment packets, beneath the 6,500-gallon coldwater tank filled with local predatory fish he built into his white and teal living room. [...]

When he wasn’t doing his schoolwork, Luckey liked reading Jules Verne, Neal Stephenson, and Anne McCaffrey novels, playing video games, and educating himself on electrical engineering, mechanical engineering, gas and solid-state lasers, and high-voltage power systems.

Once people become successful, all this gets filtered through the Halo Effect — biographers tend to keep just enough references to this stuff to make it seem lovingly quirky — but in real-time, it often seems pretty out there and many will dismiss people who don’t fit the mold.

winning first place in the Texas Renaissance Festival’s costume contest with historically meticulous renderings of Henry VIII and Anne Boleyn sewn and stitched by his wife, Nicole—who’s been at his side for 16 of his 31 years on earth—

Luckey recently built a bypass for his peripheral nervous system to experiment with giving himself superhuman reflexes; vestibular implants to pipe sounds into his skull so that instead of having to call him and wait for him to pick up, Anduril employees could just pick up a designated Palmer Phone and talk straight into his head

Don’t go too far in the other direction — not all quirky nerds with uncommon obsessions will be successful founders. Most won’t. But among successful founders, it seems rather common to have very spiky personalities.

I also liked this part (don’t think about the next step, think about the *last* step and work backward from there):

Luckey said the first smart thought he ever had came to him one night in the trailer after realizing that making consoles more portable could only take his desires so far.

“My ideological framework had been, ‘What’s the next step? How do I make my gaming PC better? How do I upgrade it?’” he told me. “But then I just had this light bulb moment where I said, ‘Next step doesn’t matter. What’s the last step?’ And that flipped my thinking upside down, because it allowed me to think in just a totally different way about the problem. And I immediately concluded, ‘Oh, it’s virtual reality. It’s the ability to literally feel like you’re inside of a game, as real as the real world. That’s the real purpose of all of this. The next step might be eight monitors instead of six, but the last step is virtual reality. That’s what I’m going to do.’”

♠️♥️ Interview: Nate Silver on the Art of Poker (probabilities, psychology, lifestyle, etc) ♣️♦️

I enjoyed this one. Didn’t know what to expect going in, and it turned out to be a bit of an atypical episode for Rick Rubin. He’s a lot less conversational than in most other eps, but it works:

• 🎧 Rick Rubin: Nate Silver

Silver is very good at explaining things and goes fairly deep into the world of Poker, which I’ve always been fascinated by at the meta level (ie. I don’t play, but I like learning about how people become good at the game — I recommend this book about how Andy Beal went against a bunch of Vegas pros in high-stakes games).

🦙🤖 Meta AI has 185m Weekly Active Users and Llama passed 350m Downloads

Speaking of AI distribution…

Zuck writes:

Llama is growing even faster than I expected: almost 350M downloads (>20M in the last month!) and a 10x jump in monthly usage since the start of the year. Excited to share the next set of updates and Llama models soon. Thanks to everyone building with us 💪

And more good news: Meta AI now has more than 400M monthly actives and 185M weekly actives across our products! Growing quickly, and we haven't even rolled out in UK, Brazil, or EU yet.

What does this mean exactly? It’s hard to tell.

When something is this new, a lot of usage may be people checking it out and not coming back, or Meta putting hooks everywhere to get people to go over there and “interact” in some way so that KPIs look good (I’m seeing a ton of Threads stuff in Instagram, and I’ve had Meta AI pushed on me in Messenger a bunch..).

OpenAI has said that ChatGPT has “200 million weekly active users”, though since they’ve been around longer that stat is probably a bit more “firm” than Meta’s.

It’ll take longer to see if people truly make a habit of using it and add it to their lives or if it’s just the initial curiosity spike… But considering how useful these models are, and how fast they get better, I wouldn’t be surprised if usage just keeps going up for a long time — I just don’t know *which* services will get the bulk of that usage.

---

🎨 🎭 The Arts & History 👩‍🎨 🎥

✍️ The Cultural Tutor on Maximalism vs Minimalism 🏺🧶

I really enjoyed this conversation between friend-of-the-show David Perell (✍️📚📖) and the man known as The Cultural Tutor (I had the pleasure of talking to him via video call yesterday — such a smart, creative, curious, and generative guy. Keep an eye out for his future projects!).

This is their second episode together, and they focus on the concepts of maximalism vs minimalism, which makes for a very thought-provoking ideastorm.

It’s something I’ve been struggling with when I write — the push and pull of wanting to get straight to the point and avoid anything unnecessary AND all the good that can come from spending more time on an idea, exploring it from more angles, going for depth and style and playfulness…

Our modern world very much leans in the direction of minimalism, so it tends to come more naturally to us, or feel more “right”.

When we think of maximalism, we tend to think of the downsides first (something that is very busy, cluttered, dense, opaque, hard to understand, impenetrable, etc). I think it’s a valuable exercise to inverse things and spend extra neuron cycles on trying to avoid the downsides of minimalism and to try to cherry-pick the good things about maximalism when it makes sense.

What feels best is having a toolbox that includes techniques from both styles. It’s a broader palette with which to paint the thought bubbles floating in my head (is that too maximalist?).