I have a six-month-old habit I am slightly embarrassed about.
When I finish a Johnny Harris video and my head is full of half-answered questions, I open Claude and we go three hours deep on whatever it was. War. Borders. Trade routes. The slow death of empires. I do not do this for work. I do it because I want to be a less ignorant person, and Claude turned out to be a better teacher than the textbooks I never read.
I want to tell you about one specific night, because I think it explains why this habit has changed my brain more than anything else I have done with AI.
The topic was the US-China chip war.
The video left me with the wrong questions
The Harris video covered the visible part — Taiwan, TSMC, US export controls, Huawei, the Chips Act. I watched it and felt informed for about an hour. Then I noticed that almost every question I wanted to ask next, the video had not even gestured at.
Why exactly can a chip not just be made somewhere else? Everyone makes phones. Why does this one company in Taiwan make almost all the advanced ones? Why has the US not solved this by throwing money at it? They threw $52 billion at it three years ago. What happened to the money? Why is the Dutch company in the middle of all this? Who decided one company in the Netherlands gets to control the entire industry? Why did China not just buy the machines before the ban?
I opened Claude at maybe 11pm.
"I just watched a video about TSMC and the chip war. I have a hundred follow-up questions and the video did not have time to answer them. Can you help me actually understand this, like I am a moderately educated person who has heard the words 'lithography' and 'EUV' but does not really know what they mean? Push back if I get something wrong. Start with the part that the geopolitics videos all skip — the actual physics of why this is hard."
What followed was the most engaged three hours of learning I have had since university. I want to tell you what I came away with, because it is darker and more interesting than the public narrative.
Part one: it is not about Taiwan, it is about ASML
The first thing Claude made me sit with was this: the bottleneck of the entire global chip industry is not Taiwan. It is a single Dutch company called ASML.
ASML makes the machines that print the patterns onto the silicon wafer. There is no second-place vendor. They have roughly 90% of the EUV (extreme ultraviolet) lithography market and 100% of the high-NA EUV market. Each machine costs around $200 million for standard EUV, $400 million for high-NA. They take years to build. The backlog is multi-year.
TSMC, Samsung, Intel — every leading-edge chipmaker on the planet — is downstream of ASML. If ASML stopped shipping tomorrow, the bleeding edge of the industry would freeze in place. There is no Plan B.
When I first understood this, I pushed back. "Surely someone else can build these machines? The Chinese, the Japanese, anyone with enough money?"
Claude walked me through why no, actually, no one can. EUV uses light at a 13.5 nanometer wavelength, which is so far into the ultraviolet that no normal optics work. ASML's solution involves vaporizing droplets of molten tin with a high-powered laser, twice per droplet, fifty thousand times per second, to generate plasma that emits EUV light, which is then bounced off mirrors so perfect that if you scaled them up to the size of Germany, the largest bump would be a millimeter. The mirrors are made by Zeiss. The laser comes from Trumpf. The whole supply chain has maybe fifteen companies in it, mostly European, almost none replaceable.
This is not a "money can solve it" problem. This is a "two decades of accumulated tacit knowledge that lives in the hands and heads of specific German optics engineers" problem.
That was the first thing I had completely wrong before this conversation.
Part two: the Chips Act is not what you think it is
I had absorbed the political narrative that the Chips Act was the US "reshoring" semiconductor manufacturing. I thought this meant the US would soon be making its own advanced chips and the dependency on Taiwan would shrink.
Claude untangled this for me, and the picture is more humiliating than the political narrative suggests.
TSMC's Arizona project — the flagship of the entire reshoring story — is a $165 billion buildout. Six fabs eventually. As of mid-2026, Fab 1 is running 4nm in production. Fab 2 just finished construction, with 3nm volume production projected for the second half of 2027. The third fab broke ground last year and is targeting 2nm and A16 nodes, probably starting in 2028.
Sounds great until you understand TSMC's internal policy, which Claude explained to me and I had to verify because it sounded made up: TSMC has an explicit "N-2" rule for overseas fabs. Taiwan stays two generations ahead. Always. By design. By the time Arizona runs 2nm in 2028, Taiwan will be doing A14 or smaller. The most advanced node in the world will, by TSMC's deliberate strategy, always be in Hsinchu.
The Arizona fabs are not reshoring. They are a hedge. A very expensive hedge. And TSMC is the one deciding what generation gets hedged.
This is when I started to feel a little sick. The strategic story the US has been telling itself — that money plus political will can rebuild the supply chain — runs straight into a Taiwanese company's internal policy that explicitly prevents it.
Part three: China is doing something nobody expected
I had absorbed the other dominant narrative — that China is years behind, sanctions are working, and SMIC and Huawei are stuck.
Claude said this was partly true and importantly wrong.
What China actually did, after being denied EUV machines starting in 2019, was take older DUV (deep ultraviolet) machines — which they could still buy because export controls had a loophole — and apply a technique called multi-patterning. You expose the wafer three or four times for every layer, instead of once, and you can squeeze chips out of equipment that was not designed for it.
SMIC used this approach to produce the Huawei Kirin 9000S at a roughly 7nm-equivalent node. The yield was bad. The cost per wafer was probably three to four times TSMC's equivalent. The methodology was, by industry standards, slightly absurd.
But it worked. China shipped 7nm-class chips at scale despite being officially years behind.
And there is now a credible report of a Chinese EUV prototype out of Shenzhen, born from what Chinese state media has called their "Manhattan Project for chips." The timeline to mass production is anyone's guess — maybe five years, maybe fifteen. The point is that the assumption that sanctions would freeze China in place is provably wrong, and the US is actively scrambling to close the DUV loophole through legislation like the MATCH Act.
Part four: the part that broke my brain
The thing that I sat with longest, after closing the laptop at 3am, was this:
No single government can solve this problem in under a decade.
Not the US, because the supply chain runs through Europe and Asia and the talent is not American.
Not China, because EUV is a twenty-year accumulated R&D project even with infinite money.
Not Taiwan, because their entire strategic survival depends on staying the indispensable node, which means they actively work against full diversification.
Not the EU, because ASML is theirs but the demand-side concentration is in Asia and America.
Every actor is rational. Every actor is constrained. Every actor is doing the thing that makes the situation harder to resolve, because the alternative is worse for them individually.
This is what Johnny Harris does not have time to say in a 22-minute video. The chip war is not a story with a villain or a solution. It is a story about how the most strategically important technology in the world ended up structurally impossible to redistribute, because every party that would have to cooperate is locked into a position that prevents cooperation.
I went to bed that night feeling like I understood something I had been pretending to understand for a year. Not because the news got smarter. Because I finally sat with the actual mechanics.
What I think Dev.to readers will find interesting about this
Most posts about "I use AI to learn things" stop at the praise. I want to say a more specific thing.
The reason this conversation worked is not that Claude is smart. It is that I was willing to be embarrassed.
I asked questions that were obvious. I admitted I did not know what EUV stood for. I pushed back when something sounded wrong, and was corrected. I asked Claude to repeat things in simpler language. I said "I do not understand" probably twenty times across three hours. I would never have done this with a human teacher because the social cost is too high. I do not even do this with Google because Google does not actually answer follow-up questions.
What Claude unlocked is not knowledge. Knowledge has been free for thirty years. What Claude unlocked is the ability to be a beginner without flinching. That, it turns out, is the actual scarce resource for learning anything as an adult.
We are all building software in a world where every chip we depend on is made by a Taiwanese company that uses Dutch machines printed with German mirrors. I do not think most engineers I know could explain why. I could not, six months ago. Now I can, badly, but well enough to follow the news with comprehension instead of vibes.
That is the gift. Not productivity. Not output. The slow rebuilding of comprehension in a world that mostly trains you to consume summaries.
A few patterns that work for me
If you want to try this — and I think you should — three things that worked across many late-night sessions:
Start with what the video or article skipped. The best prompts begin with "the documentary did not have time to explain X." That is where the real learning lives.
Tell Claude to push back. "Correct me when I am wrong. Do not just agree with me." Claude will. This is the difference between learning and being flattered.
Ask for the unglamorous mechanics. Not "why is the chip war important," but "what does a lithography machine physically do." The mechanics are where the strategic story makes sense.
I do not have a manifesto-style ending for this. The honest closing is: I am six months into a habit that has changed the shape of my attention. I am less afraid to be confused. I read the news with more context. I understand fewer things, more deeply, instead of many things, shallowly.
If you have a rabbit hole that broke your brain in a useful way, leave it in the comments. I am looking for the next one.
Thanks for reading.
— Su
Su writes from Southeast Asia. AI obsessed, quantum and space curious, occasional Johnny Harris evangelist. Find me on X @suda8866.
Top comments (0)