More Quantum

More Quantum

I became interested in quantum computing earlier this year when it became clear that AI was reaching its physical limits. You can read past posts on quantum here and here. AI growth today is constrained by total NVidia chip supply and total electrical grid capacity, not to mention the tens of billions of dollars needed to build and power these models. Under the current configuration, we may be able to 10x our AI capabilities with more spending, but 1000x improvement seems more dependent on breakthroughs than additional capex spending. Quantum computing appealed to me because it may be one of the breakthroughs that unlock massive AI gains, but other profound benefits extend beyond AI that I want to cover today.

I find it challenging to get clear answers on quantum computing from Google, ChatGPT, or even most conversations, so I’m going back to the basics in this article because I think it’s helpful to define it from the ground up, both for readers and for myself. Classical computers use bits, 1s and 0s. Bits are extremely stable, meaning the values of 1 or 0 will reliably read and write as a 1 or a 0 every time you use them.

Quantum computers are made of qubits instead of bits. Qubits store far more data than bits, but they are highly unstable. We can’t easily read and write them and they don’t retain information easily. To fix this issue, quantum computers use error-correcting algorithms to combine unreliable qubits to make something called “logical qubits.” There are a lot of complex challenges in quantum computing that make it difficult to explain and understand. I think a focus on logical qubits is fair for this article.

Microsoft’s primary quantum partner is a division of Honeywell called Quantinuum. Quantinuum appears to be the first company to have produced a computing platform with 12 logical qubits. These 12 logical qubits required 56 underlying qubits, raising the question of how many qubits will be needed to produce logical qubits at scale. Quantum computers today exceed the capabilities of classical computers. We can no longer compare the outputs with a classical computer.

Encryption and chemistry simulation are the primary applications for a working quantum computer.

Encryption

Quantum computers dramatically outperform classical computers when it comes to encryption. Post-quantum encryption services are already a significant market. Companies need to be investing in protection from quantum cryptography today. I think of this area in terms of military applications. Governments will want access to quantum cryptography tools as soon as they are available for defense and national security. I wrote about this before. Governments can collect data today to store for de-encryption once the tools are available. Realistically, no one reading this should care about this issue personally. Bitcoin records have been used to catch criminals after the fact, as bad actors previously misunderstood the privacy of the public blockchain and ended up being caught once the public records were analyzed. No private information is private forever. The cryptography topic in quantum computing feels somewhat zero-sum, like an unfortunate but necessary arms race that will be expensive for both sides.

Chemistry

The more interesting quantum application is in chemistry. One of the primary examples has to do with ammonia production, the synthetic fertilizer we use to grow enough food globally to feed eight billion people. Nature is more efficient at nitrogen fixation than our industrial processes. As soon as it’s available, we will use quantum computing to simulate the chemistry behind nitrogen fixation at the electron level. Classical computers cannot handle this calculation. The nitrogen fixation calculation will be a massive breakthrough. Biotech, steel manufacturers, aerospace companies, energy producers and so on will be able to run simulations to test and discover new materials and processes.

Total Logical Qubits Needed

Classical computers broke the standard website and banking encryption standard RSA-768 in 2009. Many financial transactions today are encrypted with the upgraded RSA-2048 standard. Classical computers have not broken RSA-2048. I can’t seem to find consistent estimates for how many logical qubits a quantum computer would need to break RSA-2048 encryption. The estimates range from 4,000 to 20 million. For nitrogen fixation, it’s estimated that we will need between 2,000 to 4 million qubits. The range of estimates is ridiculous, but it’s fair to assume that efficiency continues and the number of required logical qubits will ultimately fall closer to the lower end of the range. This isn’t a magic breakthrough machine, but it will enable previously impossible calculations that can lead to eventual breakthroughs.

The Market

It’s clear that quantum encryption services will be a large and persistent market. All sensitive industries need to take the quantum threat seriously, and it doesn’t seem this issue will ever be permanently solved.

The bigger hurdle with the quantum innovation timeline is that even after we scale up the number of logical qubits, we still need highly sophisticated people to operate these machines intelligently. Quantum-as-a-service is a likely future industry. In that scenario, large industrial companies would allocate an annual budget to quantum R&D to work on problems related to their field. Companies don’t want to manage their data centers and won’t want to operate in-house quantum computers either.

It’s challenging to invest in the quantum industry in public markets. There are pockets of quality but there aren’t many public companies purely focused on quantum. We are looking at private markets for quantum opportunities. In general, I think there is likely to be more than one approach to quantum computing that succeeds. I also think the timelines can be surprisingly fast. AI had its ChatGPT moment in 2022, and fusion had a breakthrough a few years ago thanks to Commonwealth Fusion’s new magnet technology. Eventually, challenging technical problems get their breakthrough.