The State of Classical Computing Hardware
Date: 2025 Oct 09
Words: 1759
Draft: 1 (Most recent)
Someone interested in financial markets recently commissioned me to write about quantum computing. Since quantum computing is at least 10 years away for practical usage, I am going to be doing an overview of current and near-future computer hardware right now before talking about quantum computing. This is meant to be a brief overview, but every sentence in this series is a gross oversimplification and could have its own thesis written about it. Lord help me writing this, for I am not really qualified to speak on the topic, and I don’t know if I am going to stay on topic either.
1. Intro
Right now the US and China are in a race for computational dominance. Whoever gets the best computers faster will be able to make better computers to make better computers to take off and do God knows what. If one nation can build powerful enough quantum computers, it can lead to a “fast takeoff” (even more better computers to do something with), but until then, the war is still being waged on the classical computers front. It is impossible to talk about quantum computing without first being familiar with classical computers, and it is impossible to talk about computing at all without being familiar with the geopolitical connotations, so, I am going to be repeatedly touching back on the geopolitics. There are a few metrics that we measure computing hardware on, and in this list the thing after the comma is which is better
- Classical Computers
- FLOPs on a chip, more
- Node size, smaller
- Power consumption, less
- Parallelization, more
- Quantum Computers
- Qubit count, more
- Coherence time, longer
- Error rates, lower
- Gate fidelity, higher
2. Classical Computers
“Throw everything in NVIDIA”. That has been pretty solid financial advice since about 2020. It made Leopold Aschenbrenner’s fund Situational Awareness worth $1.5B A lot of people believe in this stuff (if quietly), they think they are building a god. Mark Zuckerberg says Meta ‘misspending a couple of hundred billion’ in the U.S. would be ‘unfortunate,’ but ‘the risk is higher on the other side’. (Of course, all this has done so far is create the infinite slop machine) NVIDIA chips do parallel calculations really really fast (“Parallelization, more”). Only simple types of operations though, like matrix multiplication. When you (gross oversimplification) do these billions and billions of times, you get ChatGPT.
If you want to do more complicated algorithms than matrix multiplications, you need multi-core cpus. The domestic American side of CPU manufacturing is actually pretty interesting right now, Intel is dying and is trying to be resuscitated but it is impossible to tell if it actually will be. Leaving just AMD to make CPUs. Both the Biden and Trump administrations have been heavily invested in revitalizing the American chip industry. If the US domestic chip manufacturing industry does not take off again, than the U.S. loses. That is the first big “if”.
To understand the state of the computing industry, it is essential that you know what a few of the ISAs (instruction set architecture) are. Intel has an ISA called x86, which they secondarily share with AMD under a licensing agreement from the 1980s. x86 has thousands of instructions and is backwards compatible with the very first processors Intel made, all the way back to the 1970s. It is a monolith of inconceivable complexity, and only Intel and AMD have the horizontal transfer/“keeping the eternal flame lit” to make it. There have been some prominent signs that x86 is on its way out. Apple stopped using x86 chips in MacBooks in 2020 and switched to its own Apple Silicon chips based on ARM (here). Leading data center provider AWS is increasingly investing in ARM-based chips that consume less power (here). Apple is reconsidering using Intel as a manufacturer again, under pressure from the Trump administration, but if they do there is zero chance they will re-integrate x86 into their ecosystem, and will instead have Intel manufacture Apple Silicon chips (here). It is my personal opinion that x86 is on its way out, and that Intel may even have to throw it overboard if it wants to save itself.
The more prominent instruction sets now are RISC based. RISC (Reduced Instruction Set Computer) has an interesting history worth looking into on its own, but for now all you need to know is that it the first RISC was started at Berkeley in the 1970s, and has gone through a few iterations, with the latest being RISC-V (five) which was started in 2010. RISC until 2020 was American based, but in that year the RISC-V foundation moved to Switzerland, which of course is a historically neutral country, to get around American EAR Commerce Department export controls and collaborate with Chinese companies like Huawei (Berkeley, those communist sympathizers). (here) Worth adding on, ST Microcontrollers is also based in Switzerland and uses RISC-V, and is a non-insignificant player in the chip manufacturing business.
ARM is the name of a company that sells the license to the instruction set of the same name (“Advanced RISC Machines”), founded in 1990 in Cambridge, UK as a joint venture between Apple, Acorn Computers, and VLSI Technology. It has a few different instruction set variants - ARM-A (“application”) for general consumer devices like phones and tablets; ARM-R (“real-time”) for automotive, industrial, and manufacturing; and ARM-M (“microcontroller”) for embedded devices, with a decreasing instruction set count from A to M. (here)
The fact that the both the American and Chinese chips of the future are RISC-V based should be of some note. It puts both countries on an “even footing”, or, neither country has an inherent advantage over the other because of their superior instruction set. It also makes it comparatively easy for one country to copy another’s work. Of course, a RISC-V-based chip can have obfuscation in it, but that is nothing compared to having a whole different instruction set. The Chinese haven’t completely reverse-engineered x86 (they can’t have it because of export controls), but they don’t have to, they are using ARM licenses and are going all-in on RISC-V right now.
To understand the current state of computer manufacturing, it is essential to understand the geopolitical context. So right now the best chips are made on the island of Taiwan which is under the control of the R.O.C., with additional manufacturing in Japan and South Korea. From North to South: Japan, Taiwan, and the Philippines are known as “The First Island Chain”, and are all under the American sphere of influence. The First Island Chain is integral to the U.S. maintaining a global trade system that works in its favor. Less prominently, the U.S. has strong economic ties with the R.O.C.. More prominently, the U.S. has military bases on Japan and the Philippines, which were both previously administered over by the U.S., both by a MacArthur - the Philippines by Arthur MacArthur Jr. from 1900-1901 and Japan by his son Douglas MacArthur from 1945-1952. The blood remembers.
Let’s not try to explain all of Chinese history. I barely know any of it it is its own can of worms and it is millennia of history that doesn’t intersect with our own that much until recent centuries. So we will keep it historically recent and keep it to one paragraph. In 1927 a civil war started between the communist faction lead by Mao Zedong and the Republican faction led by Chaing Kai-shek. In the late 1930s Japan invades, when the Republicans have the upper hand in the war. Both sides agree to stop fighting eachother and instead fight a common enemy, but the Republicans bear the brunt of the war with the Japanese, which gives the Communists time to recover. After the Japanese are defeated in 1945, the civil war resumes in 1946 and the Communists stage a comeback and force the Republicans out of China, and they retreat to the island of Taiwan. The Republicans set up their own government there, and to this day both governments claim to be the only legitimate China and claim control over the other’s territories (the R.O.C actually claims control over China’s former borders before WWII, which includes territories administered by neighboring countries as well).
So now we have two Chinas. The smaller Taiwan needed to find something to do to secure protection from the United States against mainland China, so they got really really good at making chips, which are extremely difficult to make, require a lot of experts, has a complicated supply chain, on and on. And they nailed this. So now the U.S. is next-to forced to defend Taiwan against a potential Chinese invasion. Which they are preparing to do.
Let’s speedrun one of the most complicated supply chains in the world. The P.R.C. (mainland China) is unable to make chips as advanced as Taiwan’s because of export controls on critical manufacturing equipment. The bottleneck: ASML in the Netherlands is the only company that makes EUV lithography machines, which use optics from Zeiss in Germany. These machines cost $200M each and are essential for making cutting-edge chips. ASML is blocked from selling to China due to US pressure on the Dutch government, leaving Chinese chipmaker SMIC stuck at older 7nm technology while TSMC in Taiwan produces 3nm chips.
Here is a more comprehensive diagram, made by Claude:
CHIP FABRICATION SUPPLY CHAIN
=============================
MATERIALS & SUBSTRATES
├── Silicon Wafers
│ ├── Shin-Etsu Chemical (Japan) - #1 globally
│ ├── Sumco (Japan) - #2 globally
│ └── Siltronic (Germany)
│
├── Ultra-Pure Chemicals & Gases
│ ├── Specialty gases: Air Liquide (France), Linde (Germany/US)
│ ├── Photoresists: JSR (Japan), Tokyo Ohka Kogyo (Japan)
│ ├── CMP slurries: Fujimi (Japan), Cabot (US)
│ └── Etchants & developers: BASF (Germany), Merck (Germany)
│
└── Rare Materials
├── Fluorine compounds (Japan)
├── Neon gas (Ukraine - disrupted by war)
└── Photomask blanks: Hoya (Japan), Shin-Etsu (Japan)
LITHOGRAPHY EQUIPMENT (Pattern Transfer)
├── EUV Lithography (most advanced, <7nm nodes)
│ └── ASML (Netherlands) - MONOPOLY
│ ├── EUV light source: Cymer (ASML subsidiary, US)
│ ├── Optics/mirrors: Zeiss (Germany)
│ ├── Laser systems: Trumpf (Germany)
│ └── Control systems: various US suppliers
│
└── DUV Lithography (older nodes, 7nm+)
├── ASML (Netherlands) - dominant
├── Nikon (Japan)
└── Canon (Japan)
DEPOSITION EQUIPMENT (Adding layers)
├── Applied Materials (US) - largest equipment maker
├── Lam Research (US)
├── Tokyo Electron (Japan)
└── SCREEN Holdings (Japan)
ETCHING EQUIPMENT (Removing layers)
├── Lam Research (US) - dominant in etch
├── Tokyo Electron (Japan)
└── Applied Materials (US)
METROLOGY & INSPECTION (Quality control)
├── KLA Corporation (US) - dominant
├── Applied Materials (US)
├── Onto Innovation (US)
└── Hitachi High-Tech (Japan)
ION IMPLANTATION (Doping silicon)
├── Applied Materials (US)
└── Axcelis (US)
CHEMICAL MECHANICAL POLISHING (CMP)
├── Applied Materials (US)
└── Ebara (Japan)
CLEANING SYSTEMS
├── SCREEN Holdings (Japan) - dominant
├── Tokyo Electron (Japan)
└── Lam Research (US)
THERMAL PROCESSING
├── Applied Materials (US)
└── Tokyo Electron (Japan)
TESTING & PACKAGING EQUIPMENT
├── Advantest (Japan)
├── Teradyne (US)
└── KLA (US)
AUTOMATION & ROBOTICS
├── Brooks Automation (US)
├── Daifuku (Japan)
└── Murata Machinery (Japan)
SOFTWARE & DESIGN TOOLS (EDA)
├── Synopsys (US)
├── Cadence (US)
└── Siemens EDA (Germany/US)
↓↓↓
ALL CONVERGE AT:
↓↓↓
╔═══════════════════════════════════╗
║ TSMC FABS (Taiwan, R.O.C.) ║
║ ║
║ • Fab 18 (3nm production) ║
║ • Fab 15 (7nm/5nm) ║
║ • Fab 12 (older nodes) ║
║ • Fab 14 (specialty) ║
║ ║
║ Integrates ALL equipment above ║
║ to produce cutting-edge chips ║
╚═══════════════════════════════════╝
As you can see, it is an extremely complicated supply chain. What happens if China invades? Well, it would basically destroy everything. Taiwan has contingency plans to destroy the fabs if China invades, most of the specialists would refuse to work for the P.R.C. and probably wouldn’t be able to think so clearly and delicately under gunpoint, some of the pro specialists would probably be evacuated by Japan and the U.S., all the equipment at the fabs is extremely sensitive and bombings nearby would destabilize it, if power is cut suddenly it could render permanently defunct some of the machines, most of the international partners would refuse their shipments to a P.R.C.-controlled Taiwan.
But China is still entertaining invading Taiwan because of historical grievances, a need to legitimize itself, and a desire to demonstrate military supremacy. China is also operating under an M.O. of, If China destroys these chips, than the U.S. can’t have them either. Aside - for these reasons, I suggest buying the best computer you can and just holding onto it. They might not be able to make computers these good again for a very long time.
Part two of this article, including parts about quantum computing, will be published a few days from now.