Finding Equilibrium #11: Nuclear Fusion, Zero-Knowledge Proofs, and the semiconductor shortage
Hello again everyone,
Welcome to the first issue of 2022! It's already been quite a year. Cool stuff keeps happening, so let's jump back into it with 3 reads, 2 tweets, and 1 listen for the week:
Reads:
🤯 DeepMind trains an AI to control Nuclear Fusion
Nuclear fusion has been touted as the solution to clean energy for decades. Here’s a quick recap on how it works:
Hydrogen atoms are smashed together at insanely high temperatures (greater than 150 million degrees) to energize them into plasma
In this plasma, the hydrogen atoms fuse into helium and release enormous amounts of energy
We capture this energy with 0 pollution worries or energy constraints
This happens easily in the sun – its sheer gravitational forces hold the Hydrogen atoms together. But the Earth weighs 333,000 times less than the sun, so nuclear fusion is a little harder here. We navigate this by running the reaction with magnetic coils in a donut-shaped vessel called a Tokamak . The Tokamak contains the reaction. The magnetic coils maintain the plasma. But you can't just set your magnetic coils and walk away – they need to be continuously adjusted to maintain the plasma. That's the problem DeepMind solved with AI.
DeepMind trained their AI to continuously reconfigure 19 magnetic coils to generate various plasma shapes in a Tokamak. An operator could set objectives and walk away, leaving the AI to handle the rest. This unlocks a fundamental shift from engineering-driven control to AI-driven control: engineers don’t have to worry about testing and hard-coding different cases, Deepmind’s AI will continuously learn and adapt instead. It frees up a critical barrier to scaling nuclear fusion – if it works at scale, we can power the Earth with just ~8000 gallons of water a year (a 10x10x10 foot cube)!
👾 Microsoft and the Metaverse: How software and hardware go opposite ways
Here's a wild metaverse counterpoint: despite all the new web3 companies coming up, Microsoft is best positioned to grab the Metaverse. The argument draws a parallel between Virtual Reality, the metaverse's primary interface, and the PC, the internet's initial interface.
PCs were destinations. You had to go to them – they weren't constantly with you like your phone. That why PCs never started in households. People found them at work, eventually brought them home, and that unlocked widespread use. VR is similar. It's a destination – you're probably not walking to work with VR goggles on! With Microsoft’s multi-year enterprise VR focus, VR could follow the same path as the PC. Its main adoption would be in enterprises, then it’ll make its way back home (properly, this time), making Microsoft the primary “gateway” to the metaverse too.
What jumped out to me here is how software and hardware follow opposite adoption trends. With software, trends start with the consumer (ex. gaming, intuitively designed phone apps) and then move to the enterprise (Ex. making work applications fun and well-designed). With hardware, trends start with the enterprise (Ex. using a PC at work) and then move to consumers (buying one for your home). But given VR’s progress in gaming, there is a world where it could break this trend!
Blockchains have a problem– instead of gaining value with waves of new users, they become slow and expensive. That's why so much crypto discussion is around scaling: it won't get far if every transaction costs $30 and takes 5 minutes.
Zero-Knowledge Proofs (ZKPs) are today’s cutting edge scaling solutions. By using complex math, they let you prove something without sharing information (for example, proving you can make rent without revealing your entire bank details, SSN, etc to your landlord). That means you get privacy and there’s less data to be processed on the blockchain. Because of how they work, they also get cheaper with more transactions!
If you think of a blockchain as having consensus, execution, and data availability layers, ZKPs eliminate the execution bottleneck by making it cheap, fast, and private to run transactions. ZKPs scale blockchain transactions while enhancing privacy. While the technology is cool, I'm still wrapping my head around what big things it unlocks that we can’t do today...let me know what you guys think or find!
Tweets:
This thread zooms us out with some eye-popping stats. There's a potential essay in each tweet, but I’ll pick just one here: we're approaching a shortage of 300k semiconductor workers by 2025.
Semiconductors underpin computing, from AI to crypto to your phones. More technology means more semiconductors. On top of that, individual companies are bringing chip-creation in-house (Ex. Apple with its M1 chips), which only adds to this rising need for semiconductor workers.
It’s hard to train a handful of people on semiconductor manufacturing and design, let alone 300,000. Designing and making chips is an intricate, extensive process. But bottlenecks spur innovation. We’re probably not going to up-skill 300,000 people, but we can use software to supercharge a smaller subset. Google’s already doing deep research in this space, and I bet we’ll see a ton of inventions around designing and manufacturing semiconductors. Have a listen of the episode for a solid primer!
This was a gem in the nuclear fusion rabbit hole. Most of us stop at the energy part, but this thread unravels other things fusion unlocks. If we’re synthesizing an element in the process (Helium from Hydrogen), why not use it to synthesize other heavy elements too? Imagine never having a Lithium shortage or being dependent on uncontrollable conditions to mine and extract materials. It’s probably decades away, but an exciting reminder of what’s to come.
Listen
Cadence: software behind semiconductor design
Semiconductors are a $550B industry today, and they unlock $1T+ in value because you need them in everything from the cloud ($400B) to smartphones ($400B) to PCs ($250B). All of this rests on the ability to do semiconductor design, which is a highly technical and convoluted process that relies on two tools: Cadence and Synopsis. These companies only represent ~$10B in revenue, yet you could argue they’re the bedrock from which most of our digital economy is built from! If that semiconductor stat jumped out you too, this is a great episode to get caught up and learn about Cadence.
As always, please feel free to hit reply with any of your thoughts, comments, and feedback! It always helps me better refine this newsletter and keeps me cranking it out.
Thanks for reading and I’ll see you guys next time,
Aqil