538: Google's Dick Move, Intel Fires Gelsinger, USSMC, Meta's 4GW of Nuclear, Bluesky, Apple M4, Morris Chang + Jensen Huang, US vs Europe, and Paul McCartney

"Without leading-edge chips, everything turns to mush"

Many don’t have an opinion until they’re asked for it, at which point they cobble together a viewpoint from whim & half-remembered hearsay, before deciding that this 2-minute-old makeshift opinion will be their new hill to die on.

—Gurwinder's Theory of Bespoke Bullshit

🏡🏋️‍♂️ Home gym update:

I’m still rocking the Nüobell adjustable dumbbells (from 5 to 80 lbs each) and the balance board, but I’ve recently added a power rack to my basement gym.

After researching various models for quite some time, I pulled the trigger when the month-long Black “Friday” sales started, ordering a REP PR-4000. I grabbed a separate deal from Bells of Steel for some bumper plates, a powerlifting barbell, and an EZ Curl barbell (not pictured).

Though I've only used it a handful of times, I'm already loving it.

I'm taking it slow and focusing on form, so I’ve been lifting a touch lighter than what I was doing with the dumbbells, gradually working my way back up.

🦃🥧🥔🌽🍽️🇺🇸🏠👨‍👩‍👧‍👦🙏💝 If you’re American, I hope you had a *great* Thanksgiving last week!

While we have Thanksgiving here in Canada (in October), your version seems a lot more fun than ours. Thanksgiving week is an interesting time for another reason: the internet becomes very quiet — except for photos of delicious-looking food on social media.

It’s a good time to be productive. I’ve been working on multiple projects (you’ll see!).

But it’s also the perfect time to reflect and be grateful.

I’m truly grateful for you, and the 25,999 others on this steamboat. ⚓️🚢 💚 🥃

There are two ways to help me: ✨ Become a paid sub ✨ and tell a friend about it.

---

🏦 💰 Liberty Capital 💳 💴

🏴‍☠️ Google Wants to Insert Links/Ads in Everyone’s Work Without Permission 🤦‍♀️🤬

This is the kind of crap that awakens my inner Hulk:

The Google App for iOS can automatically convert some website text into links to Google Search results [...]

Google calls the feature “Page Annotations” and says it “extracts interesting entities from the webpage and highlights them in line,” which can then take you out to Search results when you tap them — inserting links without explicitly asking the user or the site owner first.

Not only is this not opt-in, the opt-out process is deliberately sloooow and burdensome (for those who are even aware that Google is doing this to their work in the first place):

• Publishers must fill out a special form (everyone’s favorite!)

• They must list all variations of their site (www, non-www, http, https, and subdomains)

• They must wait up to 30 days for changes to take effect

30 days?

What is this, 1995!? Do we have to fax them something too?

The Problem with ⭐️Non-Consensual Content Modification⭐️

I think it’s incredibly disrespectful for Google to go into someone else’s house and rearrange their furniture without permission.

And let's be clear: the endgame is increased ad revenue. A lot of these links will go to keywords that are valuable to Google from an advertising POV — if they don’t at first, you can be sure that over time they will move in that direction. It's a clever/evil way to monetize other people's content without their consent.

What a scummy move.

If it was opt-in it would be fine.

But consider these analogies:

1. It's like a bookstore adding advertising stickers throughout a novel without the author's permission

2. It's equivalent to a movie theater using AI to insert brand logos onto actors' clothing during a film

3. Imagine if a gallery decided to add QR codes next to every element in an artist's painting

In each case, the principle is the same: degrading creative work without permission while inserting themselves as middlemen for profit.

If this feature truly added value for users and publishers, they wouldn't need to force it on everyone. Pubishers and readers would demand it. Google could make it super obvious in the Google app with a popup that says "you can now do this! click here to turn it on!"

Links are content. Links are editorial decisions. Links affect how pages look and flow. Modifying that without the author's permission or even the reader's permission feels like just more enshitification of the web, like GDPR popups and all that crap.

Writers like John Gruber and Ben Thompson (💚 🥃 🎩) are pouring their souls into their work. To have some algorithm come in and add links to car dealerships because they wrote about EVs or to insurance companies because they wrote about Berkshire feels like it cheapens their work and undermines their editorial self-determination.

Take Byrne Hobart's writing (💚💚💚💚💚 🥃), his carefully selected links are integral to the content. If I click and get a random Google search, that dilutes Byrne’s work.

This practice could also create link fatigue and trains readers to be less likely to click if the overall quality of links goes down. This means that when you actually want to show something to people and send them somewhere, they will be less likely to do it.

Even if they're good links, doing it without permission is a bad idea.

Thought experiment: it may also be valuable for Google to rewrite someone’s prose to “improve” grammar and to summarize long paragraphs, but that would also be a dick move if done without permission or user input.

I care deeply about the written word. For a company that was largely built on it and the open web's creative forces, Google should know better than this.

Maybe they need to bring back “Don’t be evil”.

🦅🇺🇸 U.S. vs Europe Tech Companies 🇪🇺🤏

The graph is self-explanatory…

You can see a list of tech companies ranked by market cap here.

The source of the graph is Andrew McAfee.

ASML, a spin-off/joint venture of Philips, is not included because the graph is only for companies “from scratch”: a company from its inception as a new entity, rather than through mergers, acquisitions or spinoffs from established firms.

🥾 Intel Board Hits “Eject” Button on Pat Gelsinger 🚪 + The Case for a ‘USSMC’ Big Tech Consortium 🐜🦅🇺🇸

After four years of attempting to right the ship, Gelsinger is out. It looks like he was “retired” after a board meeting that got spicy 🌶️

You rarely see the CEO of such a large (well, increasingly small, but still important) company get a press-release’d on a Sunday with a retroactive retirement date the preceding day. Safe to bet he didn’t get a gold watch 😬

Bloomberg:

The clash came to a head last week when Gelsinger met with the board about the company’s progress on winning back market share and narrowing the gap with Nvidia Corp., according to people familiar with the matter. He was given the option to retire or be removed, and chose to announce the end of his career at Intel, said the people, who asked not to be identified because the proceedings weren’t made public.

Reading between the lines, this likely signals we’re about to see Intel go in a pretty different direction — possibly broking it up.

Is this you, Pat?

Gelsinger wasn’t the guy to do that. He likely has an Intel tattoo somewhere.

In his interview with Ben Thomspon, he made it clear that his strategy was to keep the band together as a single company (even if with more autonomy for the foundry division).

Looking ahead, I wouldn’t be surprised if we saw them sell Alterra to raise cash, then split the foundry from the design side and go fabless like AMD and Nvidia. This would immediately make the fabless side more competitive, freed from the manufacturing anchor around its neck… but!

BUT

The foundry remains the scarce asset and critical to national security. Without leading-edge chips, everything turns to mush in both the civilian economy and the military.

I remain hopeful that the 👉 Big Tech Consortium USSMC plan I outlined previously 👈 is given a shot. It seems like the best way to create a national champion that can compete with TSMC (eventually, after substantial development and investment). Or at least serve as a solid Plan B if TSMC is ever taken off the board (Thor forbid).

🏗️ ⚛️ ⚛️ ⚛️ ⚛️ Meta Requests Up to 4GW of New Nuclear ⚡️

It’s happening!

At Meta, we believe nuclear energy will play a pivotal role in the transition to a cleaner, more reliable, and diversified electric grid. That is why today we announced that we will be releasing a request for proposals (RFP) to identify nuclear energy developers to help us meet our AI and sustainability objectives.

Our aim is to add 1-4 GW of new nuclear generation capacity in the U.S. to be delivered starting in the early 2030s. We are looking to identify developers that can help accelerate the availability of new nuclear generators and create sufficient scale to achieve material cost reductions by deploying multiple units, both to provide for Meta’s future energy needs and to advance broader industry decarbonization. We believe working with partners who will ultimately permit, design, engineer, finance, construct, and operate these power plants will ensure the long-term thinking necessary to accelerate nuclear technology.

This is exactly what Mark Nelson and I discussed in our recent podcast about Big Tech’s need for nuclear power if they want to keep AI scaling (stay tuned for part 2 very soon!).

The most significant aspect here is the potential multiplier effect: We're likely to see other Big Tech players follow suit with their own requests for proposals for new builds.

The napkin math on Amazon, Microsoft, Google, and maybe Apple and Oracle, each funding 1-4GW gets us in the 6-25GW range.

Now we’re starting to get somewhere. Time to build (and not just write an essay about it)! 👷‍♂️🛠️🏗️

🦋 🕰️🔄 🐦 The Bluesky Bunker Experiment (Twitter’s Back to the Future?)

Bluesky usage stats 📈

When Mastodon, Threads, and Bluesky first came out, I created accounts on each to form first-hand opinions.

Initially, Threads seemed the most promising Twitter alternative, while Bluesky left me underwhelmed. I rapidly abandoned Mastodon and Bluesky, only checking Threads occasionally to keep a finger on the pulse there.

Fast-forward to today, and Threads (🧵🪡) seems a lot less interesting to me. It’s more like Instagram or TikTok than Twitter in spirit, even if the superficial mechanics resemble Twitter.

The platform is not designed to follow real-time events, and conversations don’t emerge as organically because the algo is mostly showing engagement bait (memes, short-form videos, and Instagram-like curated lifestyle photos).

The spiritual successor to Twitter’s peak era (2010-2015) appears to be Bluesky.

There are fewer people, but whenever I log in, I end up seeing what the people I follow post. Conversations emerge more naturally. I see links being recommended — Twitter used to be the best place in the world for discovery until they shadow-banned links.

Things are very customizable, so if you prefer an algorithmic feed or want to see what is popular with your follow-graph, you can also see that.

While Twitter/X's network effect will likely maintain its dominance, having Bluesky as a backup platform is reassuring (remember: build the bunker before you need it).

In a hypothetical world where we solved the coordination problem and everybody decided to move over there, I think the platform mechanics would lead to a healthier “town square” than the current incarnation of Twitter, which is trying to be more like TikTok/Reddit than 2013 Twitter.

• If you want to follow me on 🦋: liberty.bsky.social

See you over there, in the bunker! 👋

🇺🇸 🚚🚚🚚📦 🇨🇦🇨🇳🇲🇽 U.S. imports of goods from Canada, China, and Mexico

Import/export numbers are always tricky because most people don’t look under the surface

Is an iPhone imported from China? Well, it’s assembled there, many parts are from there, but most of the value comes from the US, Taiwan, South-Korea, Japan, etc…

Source.

---

🧪🔬 Liberty Labs 🧬 🔭

🏎️ Apple’s M4 Chip: Very impressive! 🍎🐜

Let's rewind to 2020. Apple's M1 chips didn't just enter the market – they redefined it.

They were almost too good.

It did it all: Extremely power efficient AND extremely fast. Gobs of memory bandwidth, strong GPUs, a fast TPU for machine learning/AI tasks, and multiple versions that scaled from 8 CPU cores (4 efficiency cores + 4 powerful cores) all the way to 20 cores with the Ultra.

The M1 family of chips put Apple’s laptops years ahead of the competition.

The M2 and M3 were more incremental upgrades, and I was starting to think that maybe the M1 jump forward was turning out to be one of those once-a-decade breakthroughs when everything lines up.

So when the M4 was announced, my expectations were modest... Just another iterative update 🥱

But it turns out that the M4 has some fire in its belly:

The graph above shows performance per Ghz. The M1, M2, and M3 are pretty similar, so most of the increase in performance from one generation to the next came from faster clock speeds and more cores. But the M4 breaks this pattern.

• The M1 had a max clock rate of 3.2Ghz

• The M2 peaked at 3.49Ghz

• The M3 peaked at 4.05Ghz

• The M4 seems to peak at 4.3GHz (with some reports that the Max version can go to 4.5GHz)

So the M4 runs at a faster clock rate AND does more per cycle.

This double-dipping allows it to show the biggest inter-generation improvement so far (around 22%) while also doing better on multi-core benchmarks because it also has more cores (10 cores for the M4, 14 cores for the M4 Pro, and 16 cores for the M4 Max, but this one doubles the GPU cores of the Pro from 20 to 40).

A lot of this comes from being fabbed on TSMC’s second-generation 3nm process. This first 3nm process was a technical dead end and didn’t have very good yields, so it was fairly quickly abandoned (that’s what the relatively short-lived M3 was fabbed on).

To appreciate the M4's efficiency, consider this:

The AMD Ryzen Threadripper PRO 7995WX scores 25,365 on Geekbench multi-core using:

• 96 CPU cores

• 2.5GHz base / 5.1GHz boost

• Up to 350 watts under load

The M4 Pro achieves 22,527 on Geekbench with:

• Just 14 CPU cores

• Only ~40 watts of power

Imagine an M4 SuperUltra with 96 cores and a 350-watt power budget. The performance would be bonkers 🤯

🇹🇼🐜 Morris Chang Offered TSMC CEO Role to Jensen Huang 😎

There’s a new memoir by Morris Chang (I can’t wait for the English translation) that contains this fun anecdote:

In the book, Chang said that while searching for a successor in 2013, he viewed Huang as an ideal candidate because of his character, professional experience, and semiconductor expertise.

"I spent about 10 minutes concisely explaining my deep expectations for TSMC," Chang recalled. He said Huang listened patiently but responded, "I already have a job."

A few weeks later, Chang tried again, but Huang was resolute.

"Jensen’s response to me was honest: He 'already had a job!' That job was to elevate Nvidia to what it is today, 11 years later," wrote Chang.

Imagining what may have happened if Jensen had accepted is quite the alternate timeline “what if” 🤔 (but I don’t think there was any real chance he would accept)

TIL About “Heavenbanning” 🔐☁️

I hadn’t heard of this concept. Sounds like something out of Black Mirror:

Heavenbanning is a theoretical social media moderation concept where problematic users are placed in an artificial environment without their knowledge. In this system, the banned user can still post content, but their posts are only visible to AI-generated accounts that provide positive feedback and engagement.

How It Works

- The heavenbanned user experiences a carefully crafted illusion where:
Their posts receive enthusiastic engagement and praise from seemingly real users

- They gain followers and receive supportive comments

- Any disagreements they encounter result in the opposition eventually conceding to their viewpoint

However, all of these interactions are generated by AI language models, and no real human users can see or interact with the heavenbanned person's content.

Technical Implementation

While heavenbanning remains largely theoretical, it could be implemented using:

- Large language models to generate convincing responses and interactions

- AI systems trained to provide personalized, narcissistically satisfying experiences

- Content generation algorithms tailored to match the user's confirmation bias

Behind the Scenes of Nuclear Diplomacy & Non-Proliferation 💣✋📝

Great podcast interview of Australian diplomat Richard Butler by friend-of-the-show Joe Walker (🇦🇺):

• 🎧 Richard Butler — Nuclear Diplomacy at the End of History (#160)

Butler played a crucial role in both the permanent extension of the Nuclear Non-Proliferation Treaty in 1995 and the adoption of the Comprehensive Nuclear-Test-Ban Treaty in 1996. His work helped establish the framework through which we still manage nuclear weapons risks today.

Some of the anecdotes and tactics used to get deals would make for great film scenes!

---

🎨 🎭 Liberty Studio 👩‍🎨 🎥

🎶 Paul McCartney talks about how he wrote 10 iconic songs

I love this kind of stuff. The creative process is endlessly fascinating!

h/t my friend Dylan (🇮🇪)