The only story that you should be paying attention to right now is the United States vs China situation.
Here's a primer on what's going on:
For decades, the United States and China had a mutually beneficial partnership.
The US was a hub of innovation and deep technical expertise. China had a billion people to throw and any labor intensive problem.
But now, a new wave of trade restrictions is changing everything.
The U.S. is cutting off access to GPUs which are essential to training artificial intelligence.
AI is a classic "dual-use" technology. AI powers social media ad placement, but it can also be used in military weapons.
The new rules set a hard cap on what China can import.
This all really started heating up in 2019 when Trump blocked Huawei. Backdoors in 5G towers would have given China the ability to monitor global internet traffic.
Trump was notorious for getting "tough on China", but here's the weird thing.
When Biden took office in 2021, he didn't change course.
Even though Biden and Trump were painted by the media as polar opposites, on the China issue, they agreed.
It had become a bi-partisan issue, so Biden doubled down.
He passed the CHIPS act, giving $52 Billion dollars to American semiconductor companies.
This was important because:
- New chip startups weren't getting funded.
- Big chip companies weren't investing in capex.
- America was falling behind in STEM graduates.
But he went further:
In September, Biden restricted the sale of Nvidia A100s and H100.
They have lots of memory and bandwidth. Great for training larger models.
Every AI engineer is trying to get their hands on these right now.
You can even use them to make custom AI models like this:
But specific graphics cards are just one piece of the puzzle. There's a whole supply chain.
TSMC makes the actual chips inside the cards.
ASML makes the machines used at TSMC.
ASML needs companies like TRUMPF to supply mirrors and other parts.
So Biden restricted all of this.
The latest bans cut China off from the semiconductor supply chain, so now China has to figure out how to make everything by themselves.
They are already working on it with their own versions of TSMC and ASML (SMIC and SMEE), but it's hard to say exactly how long it will take.
The deeper you go into this story, the more you'll realize what a big deal this is. Tensions have never been this high, but fortunately, it's not quite boiling over yet.
Biden and Xi Jinping were still able to meet in person face-to-face to discuss things.
The bottom line is that the United States is ahead in artificial intelligence and wants to maintain that lead.
This will be the defining geopolitical story over the next few years.
If you'd like to know more, just watch this video:
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Of course that’s your contention. You just heard about DeepSeek two days ago. Just got done watching some 40-minute deep dive—Deirdre Bosa, probably. You’re going to be talking about how this complicates things for the hyperscalers and how NVIDIA stock traded down 15%.
And how this is bad for GPU demand. You’re going to be convinced of that until next month when you learn about how Jevons’ paradox predicts that cheaper AI models will actually increase demand for semiconductors.
That’ll last a year or so, then you’ll be reading SemiAnalysis and Jeffrey Emanuel, talking about how the training cost might have actually been under $6 million, how they used FP8, and how the DeepSeek team is really omega cracked.
What to do when a journalist emails you and they are clearly writing a hit piece.
These pics show what “the good ending” looks like. A true flawless victory.
🧵 Thread of 10 anti-hit strategies, mostly stolen from @micsolana and @lulumeservey
1: Do not give an interview.
It never turns out well and it makes it harder for other people to circle the wagons. Don’t respond or say “no comment.”
2: Tell your friends
It’s highly likely someone in your immediate friend group has been through something similar. And someone in your group chat is going to have a good dunk tweet.