Thunder Compute enables users to deploy AI or machine learning models rapidly while paying only for active GPU usage.

Thunder Compute

Thunder Compute (YC S24) | 3,251 followers on LinkedIn. Creating technology to serve more users with the world&#39;s GPUs | Thunder Compute creates virtual GPU-over-TCP technology to decouple CPUs from GPUs. This allows developers to share a pool of GPUs across thousands of machines rather than reserving individual GPUs for each machine.

Join our community: https://discord.gg/nwuETS9jJK

Investor at WaveAI Investor at Plato Investor at Jugemu.ai

Daisuke Minamide

Founder at Thunder Compute (YC S24)

Brian Model

Founder at Thunder Compute (YC S24) | Building what's next | I like trying new things

Carl Peterson

Founding Engineer at Thunder Compute (YC S24)

Jack Decker

Thunder Compute (YC S24)

Exciting times for Thunder Compute (YC S24) ! We’re hiring, launching new features each week, and now… we have a billboard in Times Square. It’s surreal to see an idea that started with Brian Model and I in a college dorm now on display in the heart of New York City. Our vision for data scientists to 'never pay for idle GPUs' is becoming possible thanks to our growing team and support from our early adopters. I am so grateful for every one of you. Thank you to our partner Brex for the billboard—check back next year to see the second version come to life 😏

Your coworker Jim racked up a $10,000 AWS bill by leaving his GPUs on for a month. You might know Jim. He might even be on your team. There’s a chance you are Jim (shame on you). If you’re Jim’s manager, you have probably invented a technique to convince him to turn off his GPUs. I’ve heard anything from office-wide emails of shame exposing Jim for his misbehavior to (jokingly?) threatening to fire him. What if I told you that you could stop Jim in his tracks? Brian Model and I can help. We invented Jim-proof technology to automatically disconnect idle GPUs. Don’t be a Jim. Use Thunder Compute (YC S24) .

When Brian and I launched Thunder Compute (YC S24) , everything broke. The chaos started after we made a post on HackerNews, a forum for engineers. Within 24 hours we had our first 150 users. Initially we were thrilled, this seemed like the first real sign of traction—until things started to fall apart. At first, users couldn't log in. We scrambled to release a fix and soon discovered that Windows users were shut out. Once we solved this, and people were finally able to run code, it crashed...again and again. In other words, our entire product was broken. We thought everyone would leave, but instead of abandoning our product, users reported their bugs and waited patiently as we fixed them. To this day, our main focus is still making users happy, even those who always seem to find bugs on Fridays at 11 pm and urgently need our help (you know who you are 😉)

We are so proud to have reached this point. We are looking for the best of the best to join us as employees #1 and #2.

Do you think grinding LeetCode is a waste of time? Did you work at a hedge fund or in big tech and find yourself wanting more? Do you dread the thought of spending your life building internal tools for ad tracking? If this resonates with you, we are hiring Thunder Compute 's first employee. As a founding engineer: You will immediately impact customers and touch every part of our product. You will build a first-of-its-kind GPU virtualization technology. Your work could eventually touch every accelerator in the world. This isn’t the kind of job where you can coast. You will likely deal with the gnarliest bugs you have ever seen. It will not be a 9-5. That said, this may be the most interesting problem you ever work on, with the highest impact. I can promise you won’t get bored. If this scares you away, I understand. If it excites you, reach out. Your future self will thank you. #hiring #startup #YCombinator

Meet Carl Peterson : Co-Founder and CEO of Thunder Compute Q: Where are you from? A: I was born and raised in Warren, New Jersey, although by this point, I would consider Atlanta my home having lived here for 6 years. Q: What are your hobbies? A: I have always loved music. I played trumpet for 15 years and still play as much as I can. Many are surprised to hear that when picking a college, I nearly majored in Trumpet performance—I auditioned at several programs and earned a few music scholarships. At Georgia Tech, my free time was split between marching band and ultimate frisbee. More recently, though, I have taken up rock climbing. Q: If you could live in any other country, which would it be? A: I daydream about this often. It would have to be somewhere with great beaches and scuba diving. Of the places I’ve been, Okinawa, Japan or Bali, Indonesia top the list. Q: What are you currently reading? A: I’m a pretty stereotypical business nerd. I’m currently reading Start with Why by Simon Sinek. My reading list is about half self help and half fantasy. I am still looking for ways we can bring magic to Thunder Compute—open to any suggestions. Q: When work gets tough, how do you de-stress? A: I love taking walks to a nearby coffee shop or rock climbing if I have more time. Q: What is your favorite part of being a founder? A: I love waking up to monstrous, unexpected bugs… Kidding, my favorite part is the constant opportunity to build things. As a systems engineer at heart, getting to build a system from scratch and enhance it every day is my dream job. Q: What’s your favorite thing about working with Brian? A: Brian is the best cofounder I could have asked for. Beyond his entire background being perfect for this company and likely being the smartest person I have ever met, he also happened to be one of my closest friends throughout college. So, getting to work with him has made the ups and downs much better. Q: How do you define Thunder Compute’s success? A: I’m a pretty big dreamer—I am convinced that our GPU virtualization tech will become the default way that every developer uses GPUs. I quit Bain with this vision six months before we even confirmed that the technology was possible, not having any semblance of a prototype. Today, with dozens of developers using our technology for actual workloads, I’m even more confident that this future outcome will be reality. At a minimum, I would consider Thunder Compute a success if 50% of the world’s AI developers exclusively used network-virtualized GPUs.

Thunder Compute has joined the NVIDIA Inception Program! We are excited to work with NVIDIA and the other startups in the program to bring Thunder Compute’s GPU virtualization software to the mainstream. You probably haven’t heard of NVIDIA—they’re a small hardware company based out of Santa Clara making a new type of computer chip. Jokes aside, as a co-founder of a startup building GPU technology, I am so proud for our product to be recognized by the company that literally invented the GPU. We are creating a new way to connect developers to GPUs. Our technology is built on top of NVIDIA CUDA, the market leading API for AI development. NVIDIA GPUs are the fuel for the accelerating growth of AI. Thunder Compute’s virtualization technology makes it easier for developers to use GPUs while allowing each chip to do more. Thank you NVIDIA for Startups , we are thrilled for the future of NVIDIA x Thunder Compute! #NVIDIAInception

At Thunder Compute, we are building a new type of GPU virtualization that we believe is going to revolutionize AI development. In most forms, today’s GPU virtualization is confined to a single server, either by dividing a GPU among several virtual machines or by dedicating a GPU to one. These methods improve efficiency but are limited by the physical constraints of a single server. As a result, GPU availability remains scarce and developers desperate for capacity commit to multi-year reservations. Thunder Compute takes a different approach. Our network-based GPU virtualization allows any computer to access GPUs across multiple servers. This effectively pools GPU resources across a data center, enabling dynamic allocation so that GPUs never sit idle. The result? We expect to be able to serve over 5x more developers with the same GPUs. Our technology bypasses the need for physical connections, using a network-aware interface to make remote GPUs behave as if they’re locally attached. Through horizontal scaling—in which we add more servers to expand GPU capacity—we ensure every GPU is used to its full potential by attaching it where it’s needed most. Though performance challenges remain, the rapid advancements in network-based GPU virtualization suggest a future where the speed and reliability of this technology rival those of physical connections. We believe network GPU virtualization is how developers will use GPUs in the future—and we are building the technology to prove it.

Sharing a bunk bed with two $20 mattresses isn’t exactly what most envision for themselves as a consultant at Bain. But it soon became my reality—a self-imposed sacrifice to achieve my goal: founding a startup. Instead of enjoying the perks of having a steady income, I operated knowing I would be unemployed in a year—and that my exciting, new income would soon become zero. I can’t say it was easy. My friends asked me to go on trips; my inner-foodie wanted to try new restaurants; and after long days at the office, I would’ve loved to opt for a mechanic changing my car’s oil instead of doing it myself. Safe to say, I had a lot of practice saying “no.” When it was time to rent an apartment, I took budgeting to an extreme, embracing my inner college freshman. I settled on a $600/month apartment which meant sleeping on a twin-sized bunk bed with four guys crammed into two bedrooms. The only thing luxurious about the place was the $20 used mattress I snagged off of Facebook Marketplace (it was actually amazing). I cut back in other ways too: I biked to work every day to save on office parking, opted for nights of coding instead of going out, used the office gym instead of buying a membership, and packed lunches made with Walmart groceries and love. So, when I finally quit my job and cut off my only source of income, nothing in my life needed to change—I was already embracing the unemployed lifestyle. The most valuable thing I gained from my budgeting wasn’t the savings but rather a new mentality: one of sacrificing near-term pleasure for a larger goal. This mentality prepared me for the realities of a startup—spending sleepless nights debugging, working during vacations, and missing out on weekends. Suffice to say, I will still be packing my lunches and biking around—but I think I’ll upgrade to a $50 mattress.

Carl and Brian are working hard to keep bringing new features to Thunder Compute! Latest improvements: 📈 Performance optimizations (Stable Diffusion is over 2x faster than it was last week!!) ⏩ Rapid response to bug reports for our early adopters (we love y'all) ✅ Improved instance management and onboarding for new users On deck: ⚡ Even faster performance ❓ Your suggestions Here's a quick demo of what we could do a week ago. It's only getting better.

Thunder Compute offers a platform for deploying AI and machine learning models with a focus on efficient GPU utilization, allowing users to pay only when the GPUs are active. It enables quick setup and scaling for developers and enterprises, simplifying access to virtual GPUs.

THUN

COM:THUNDERCOMPUTE

Thunder Compute

Filings Not Available