Technology, Connected - Give Quantum Computers These Jobs First

Episode Date: March 12, 2026

Quantum computing for materials is moving closer to practical use because quantum computers, GPUs, CPUs, and AI coding tools are beginning to work together. Pranav Gokhale explains how future battery design could depend on simulating electrons, splitting materials problems between GPU workflows and quantum subroutines, and using Hamiltonian simulation where classical computers fall short. The conversation connects logical qubits, Nvidia, quantum-GPU orchestration, material science, chemistry, drug discovery, and why 2028 could be an important threshold for early quantum applications.--Other ways to connect with us:⁠Listen to every podcast⁠Follow us on ⁠Instagram⁠Follow us on ⁠X⁠Follow Mark on ⁠LinkedIn⁠Follow Jeremy on ⁠LinkedIn⁠Read our ⁠Substack⁠Email: [e-mail]

Quantum computing has always been five years away. Until now. In this short, inflection CTO Pranav Gokal explains why 2028 is the year everything changes for quantum. And if you want to check out the longer episode, you can find it at thinkingonpaper.xyZ or where you're listening to this. Let me make it one step even more specific to now I'm going to make Mark a material science researcher. And so one of the things that... I want to build a space elevator. But you're building a space elevator.

Perfect. Awesome. So for that space elevator, you're probably going to want batteries because that space elevator is sometimes going to be on the other side of the earth way from the sun.

It's not always going to have solar power. And so you want to be able to store the electric inputs, the solar inputs for 24 hours before the sun comes back or 12 hours. And so one of the frontiers of where quantum computing and GPU are coming together

is for designing batteries that last much longer in a much more small form factor. And the reason that right now you can't just computationally simulate that is because when you really boil down to the computational problem, it involves simulation of electrons.

So really, what Mark is going to have is a description at the level of electrons not so dissimilar from your high school chemistry class when you learned about carbon and other atoms having a certain number of valence electrons, you're going to have that picture in front of you. But for right now, you're kind of stuck. You can't really do anything computationally with it because simulating those electrons needs a quantum computer.

And so what Mark is going to be able to do with this emerging workflow of quantum computers and GPU and CPU coming together is to take that problem of there's this potential material for a battery that lost 10 times longer, et cetera. pass it off to an application software front end, and that application software front end is going to be able to break this problem into two pieces. Now, this is getting into a little bit of jargon, but there's something called GW theory,

which is basically how these structures of molecules interact with each other, and some of that doesn't need quantum mechanics because it abstracts the way those electrons. And so that runs on existing GPU software stacks, hardware stacks that are really good at maybe that 95% resolution of that problem. But unfortunately, you need that 5%, or fortunately for us, you need that 5% extra. And that's where specifically there's a subroutine of quantum computing called Hamiltonian simulation.

I promise it's the most jargon. I'll give you at that level. But Hamiltonian simulation is kind of like a key primitive for quantum computing. It's sort of like Google search is based on page rank or the Netflix sorting algorithms. has an algorithmic primitive behind it to show you the best content on Netflix. That algorithmic primitive is going to run on a quantum computer, and then you're going to have a orchestrator workflow that performs consistency between the two. So it's maybe slightly more technical response than necessary,

but the high level is there's specific applications in material science, chemistry, drug discovery, etc., where there's known primitives that work really fast in a quantum computer, And everything else is already really good and good enough on the world's biggest GPU clusters, et cetera. And then finally, the thing I'd add is consistent with what Sam said. In the past, we might have needed to have six months of application engineers building out very specific, like drag and drop menus and software tool flows for building all this slicing and dicing of the problem. A lot of this is now as easy as running opening eye codex or cloud code, which under the hood, of course, runs on NVIDIA's stack. so that we're able to reach more applications than ever feasily thought before

in the click of a button or a few prompts, which is changing everything about how we're programming these applications. Could that be tomorrow? Could that be in 20 years? How long before I can build my space elevator with a new material that Quantum helps discover? Historically, one of the critiques of Quantum is that it's always been five years away.

I generally think it's different this time, and the reason is that for the first time ever, with credit to Google for being the pioneer here. We have as an industry stumbled upon, not just stumbled, engineered our way to what is called a logical qubit. And I guess the quick summary of what is a logical qubit is it's kind of like when you're using Wi-Fi or 5G, you get packets into your phone or your laptop or router,

but those packets have error. And that's kind of like the cubits that we had pre-20204. You had them, but they were error-prone, and you couldn't just use them raw. And of course, Wi-Fi and 5G end up giving you pristine actual information, 5 megabits per second of download speeds, et cetera, 10 megabits per second.

And that's the actual data that is pristine that you get out at the end of the day. And so in 2024, the entire industry of quantum computing went from packets to data. They went from noisy information to actually useful information. And so today, very few companies have crossed this chasm.

Inflection is one of them. And in fact, the first time we did it was this paper that we put out with NVIDIA showing two logical qubits. Two is not a lot, but it's a lot more than zero, which was what we had for the first 35 years of quantum computing. And so last year in September, we've now hit 12 logical qubits. Our publicly announced roadmap has us getting to 30 logical qubits this year in 2026 and 100 logical qubits in year 2008. So to answer your question head on, we think that at 100 logical qubits, you reach this tipping point towards some of the first applications. It's not going to be every application,

but some of the applications, especially to fields like material science. And so around the 2028 time frame is when we start to think that this flips from purely being kind of researchers testing this to understand how to maximize the juice of the future machines to actually getting, juicing those future machines, and that's about year 2028. And the last thing I'll say is there's upside surprise that has emerged in our field because of AI because things that used to take six months, honestly for me, just to figure out how to clone a repository of code with Git or GitHub, that's now the press of button. And so there might be room for upside surprise too, but in general, we think that 2028 is a realistic

timeframe.