560: AI Economic Impact and Bottlenecks, Nvidia's Real Risks, Huawei, TerraVest, TSMC-Intel, Llama 4 SNAFU, Small Biz Apocalypse, Musk Bored of Tesla, Ovitz, and The Last of Us

"The Californianess of everything was turned to eleven!"

There is no hope unmingled with fear, and no fear unmingled with hope.

—Benedictus de Spinoza

📸 Photo I took on a hike 💚 🥃

🌊☀️🕶️🏄‍♂️ Last week, I went to California for the first time. San Diego, to be more precise.

It was almost *too California!* The Californianess of everything was turned to eleven! 🌴

All the clichés that have been accumulating in my head since childhood were present: perfect weather, laid-back people, a lovely beach with great surf waves, scenic hiking trails, hills full of beautiful houses in the distance, palm trees, F-35s and F/A-18 Hornets flying overhead from Marine Corps Air Station (MCAS) Miramar (where they shot the first Top Gun film and where the real Top Gun school used to be before it moved to Nevada).

I had a friend (Jimmy Soni 💚 🥃, actually) take a portrait of me in front of a huge cactus 🌵 because why not?

On the way in, I got this shot out of the plane window of a bunch of carriers and destroyers from the U.S. Navy’s Pacific Fleet 📸:

I can also now say I’ve seen the Grand Canyon with my own eyes, even if it was from up high. I snapped this shot 📸:

🛀💭🪞 At bedtime the other night, my youngest boy asked my wife and me how mirrors are made. As I explained silvered glass and aluminum coatings, I started thinking about the world before cheap and ubiquitous mirrors (mid-1800s), and how this simple innovation likely had a profound impact on humanity.

How was people’s self-image different pre-mirror? When I think of myself, I have a pretty good idea of what I look like, and that’s integrated into my mental model of myself, but that wouldn’t have been the case for most of humanity's history.

I’m sure this had second-order effects on how people thought and behaved, and not just in the obvious ways (vanity, fashion, etc) 🤔

📖 2️⃣0️⃣2️⃣5️⃣ 🗓️ Will the word of the year for 2025 be Dirigisme?

---

🏦 💰 Liberty Capital 💳 💴

🤖🏭🏗️💰🚀 Thinking about the Likely Economic Impact of AI and Systems-Level Bottlenecks 🤔💭

Great discussion hosted by friend-of-the-show Dwarkesh Patel, who interviewed Ege Erdil and Tamay Besiroglu of Epoch AI on, among other things, the potential bottlenecks to AI’s economic impact on the world’s economy.

It’s particularly interesting because they have variant views and longer timelines than many big names in AI. They believe that even if AI keeps progressing very fast, we may not see incredibly fast double-digit GDP growth, at least for a long while.

I recommend the whole thing, it’s very thought-provoking:

• 🎧 AGI is Still 30 Years Away — Ege Erdil & Tamay Besiroglu

Here are some highlights:

On the creativity of even frontier reasoning models:

if you look at the way they approach problems, the reasoning models, they are usually not creative. They are very effectively able to leverage the knowledge they have, which is extremely vast. And that makes them very effective in a bunch of ways. But you might ask the question, has a reasoning model ever come up with a math concept that even seems slightly interesting to a human mathematician? And I’ve never seen that. [...]

just think about the sheer scale of knowledge that these models have. It’s enormous from a human point of view. So it is actually quite remarkable that there is no interesting recombination, no interesting, “oh, this thing in this field looks kind of like this thing in this other field”. There’s no innovation that comes out of that. And it doesn’t have to be a big math concept, it could be just a small thing that maybe you could add to a Sunday magazine on math that people used to have. But there isn’t even an example of that.

They also note that, in humans, certain talents cluster together, whereas in AI they can be surprisingly decoupled.

The way we test humans on certain things usually correlates with their capability to do other things that may be useful for R&D or software engineering or scientific research, but a model might crack super-hard math puzzles yet struggle on many of the tasks that drive real‑world progress.

On bottlenecks to growth, Tamay Besiroglu argues that focusing solely on AI "intelligence" misses the bigger picture, much like focusing only on "horsepower" during the Industrial Revolution ignored other crucial changes:

It isn’t the case that the world today is totally bottlenecked by not having enough good reasoning, that’s not really what’s bottlenecking the world’s ability to grow much more substantially [...]

we have a different view as to how this acceleration happens, that it’s not just having a bunch of really good reasoners that give you this technology that then accelerates things very drastically. Because that alone is not sufficient. You need kind of complementary innovations in other industries. You need the economy as a whole growing and supporting the development of these various technologies. You need the various supply chains to be upgraded. You might need demand for the various products that are being built.

And so we have this view where actually this very broad upgrading of your technology and your economy is important rather than just having very good reasoners and very, very, very good reasoning tokens that gives us this acceleration.

That riff reminds me of Tyler Cowen’s recent line: “Most of sub-Saharan Africa still does not have reliable clean water. The intelligence required for that is not scarce.”

In other words, even if you increase the supply of intelligence a lot, other parts of the system may still choke progress.

👨🏻‍🏭 TerraVest’s Large Army & Navy Contracts 💰💰💰

Quick update because I wrote a few times about TerraVest recently, and if you follow them, you may find this interesting:

• Two days ago, the U.S. Navy awarded an almost $50m contract for tactical trailers to Heil Trailer, a division of Entrans, which TerraVest recently acquired. Details here.

• Last month, Heil Trailer won a MASSIVE five-year contract with the U.S. Army for ‘Tactical Fuel Distribution Systems’. The total value is $588.6m (!). It’s a mix of “firm-fixed-price and fixed-price with economic price adjustment” so the final number may vary a bit (especially if they have to pass on some tariff costs on raw materials...). Details here.

Add it up: almost $640 million in new orders booked in <30 days. Not bad!

h/t friend-of-the-show C.J. (🇨🇦 💚 🥃)

👀🇨🇳 Nvidia & China: Short Term Pain vs The REAL Long Term Risk ⚠️

It may be tempting to view the recent restriction on Nvidia selling the hobbled H20 in China (it’s inferior for training, but still very strong for inference) as a band-aid being ripped off — short term pain, but at least now the company is less exposed to China’s uncertainty and things should be more predictable going forward.

Of course, there are downsides, like creating stronger incentives for Huawei (more on that below) to get to scale because more Chinese buyers will be forced to use its chips. And CUDA’s moat may shrink because alternatives will likely be developed, some of which may be open-sourced and used more widely…

Is there a silver lining? ☁️ In the same way that it was painful for Google and Meta to be banned from China back in the day, but in hindsight, there are some benefits to not having that Sword of Damocles 🗡️ hanging above their heads… something that Apple and Tesla are finding out!

BUT

That’s still not zooming out enough. What is the *real* big risk?