Thunder Compute

Updates: H100s are going live on Thunder Compute, you can now run AI models without leaving your browser, and we continue to improve performance 🚀 Due to popular demand, we're launching H100s from $1.49/hr! Rollout begins this week, DM our team if you need more capacity than we have available. For now, we're limiting supply because these are EXPENSIVE to rent from Google Cloud 💻 You can launch ComfyUI, Ollama, Forge WebUI and more in your browser, without installing anything! Once you've played around with a model in your browser, you can connect to the same instance with the CLI or through VSCode for more control This feature is experimental, please report any bugs 🚴 We've released a few general data transfer improvements which speed up everything. For context, Llama 3 reaches ~20% higher TPS We've also switched over to a compiled version of the CLI which is snappier and easier to install on any OS. We've sped up the web console, and performance gains continue for Tensorflow Follow for more updates next week!

Meet us at NVIDIA GTC 2025 to learn how Thunder Compute's virtualization software can accelerate your enterprise's AI initiatives for 20% of the infrastructure cost. With Thunder Compute's software and NVIDIA's hardware, we are redesigning AI infrastructure from the ground up, designed to be efficient, scalable, and developer-first. What this means is you can switch your developers to Thunder Compute, save 80% off your GPU spend, while accelerating progress. Our offerings include: ⚡ Fully-managed cloud services with the cheapest prices in the world (self serve, no contracts or commitments) ⚡ Drop-in integration with existing VPCs or on-prem GPU deployments ⚡ GPU-attached dev machines for your engineers Reach out to speak with Carl Peterson, Co-founder and CEO, about how we are re-inventing AI infrastructure (and why everyone should care). Calendar link in comment ⬇️

When Brian Model and I left Citadel and Bain & Company to build Thunder Compute, I thought we had escaped corporate politics. I was wrong. At 2AM last Tuesday, as I walked a customer in Mumbai through deploying their AI pipeline on our GPUs, I realized the politics hadn't disappeared—they'd just transformed. At Bain, feedback meant carefully worded emails hinting that slides weren’t pretty enough. Now it comes as blunt LinkedIn comments: "GPU virtualization has been around for decades" and "If it were possible, NVIDIA would have done it already." The stakes feel different too. A mistake at Bain might cost a client millions. At Thunder, one bug can take down an entire AI team's workflow—and with it, our reputation against trillion-dollar competitors. What keeps me going isn't proving critics wrong. It's seeing customers love the product we’ve poured our blood, sweat, and tears into. Last week I spent 14 hours writing test CUDA kernels while Brian debugged memory transfers. Meanwhile, our dashboards showed we were serving 10 more end-users per GPU than any of our competitors. Even better, a startup founder told me that he wanted to move his whole team to use our GPUs. No perfectly formatted slide deck could capture that feeling. Would I go back? Not unless we need another round of edits on board_meeting_vFinal_final_final.pptx.

Transparent pricing, or "book a demo"? In my experience, transparency wins every time. It's tempting for startups to hide pricing behind demos. I've heard all the arguments: "Better conversion," "More tailored experiences," "Higher retention." But let’s be honest: have you ever seen “book a demo” and felt excited? If your product is truly great, you shouldn't need to force someone into a sales call. At Thunder Compute we’ve chosen to maximize transparency. We’re fully self-serve and prices aren't just public—they're the first thing on our landing page. Today, we launched a simple pricing calculator to make costs even clearer. If you have a great product, why hide it?

We just got our VS Code extension published! Our team has been using this nonstop for internal development. It takes two seconds to set up and feels like your computer has a $30,000 data center GPU directly attached. If you are a data scientist or developer and use Microsoft's VS Code or Anysphere's Cursor this is the simplest way to use cloud GPUs. And we're already the cheapest. Not to brag, but using Thunder Compute is starting to look like a no-brainer for teams building with GPUs. And we're just getting started. ⚡

🚀 The AI Revolution's Hidden Truth: Is Our Infrastructure Ready? 🤔   After months of digging into the disconnect between traditional data centers and the demands of AI computing, I’ve uncovered insights that could reshape our understanding of the future of AI infrastructure.   The GPU Paradox 🧩 - Did you know that 80% of GPUs worldwide are sitting idle? 😳 - The industry average utilization hovers around a mere 20%. - Major cloud providers charge an eye-watering $3.06/hour for GPU access, leaving startups struggling to innovate.   This level of inefficiency isn’t just a tech issue; it’s creating a massive market opportunity waiting to be tapped!   The Solution is Here! 🚀 Innovative companies like Thunder Compute (YC-backed) are changing the game with TCP-based GPU pooling technology that: - Boosts utilization from 20% to 80% - Reduces costs by a staggering 75% ($0.92/hour vs. AWS’s $3.06/hour) - Maintains performance while democratizing access to vital resources.   For those contemplating building their own GPU infrastructure, the numbers speak for themselves: - An investment of $100K in GPUs + $15K in servers can yield $490K in annual revenue at 70% utilization with a 2-3 year break-even point! 📈   Beyond Just Pricing: The Evolution of Infrastructure 🌍 It’s not merely about cost savings. AI data centers face unique challenges such as: - 4-8x more power requirements per rack - Extreme cooling necessities - Ultra-fast networking demands - Specialized physical configurations   The future lies in decentralized networks, specialized ASICs, sustainable designs, and edge computing that brings AI closer to data sources.   Why This Matters to You: Whether you’re: - A developer hunting for affordable compute - An entrepreneur diving into AI infrastructure opportunities - An enterprise reevaluating your cloud strategy - An investor on the lookout for the next big play in infrastructure   This transformation is poised to impact your roadmap significantly.    What’s been your experience with AI compute costs? Are you exploring alternatives to the major cloud providers? I’d love to hear your insights in the comments! 💬   Swipe through the slides for a deeper dive into the analysis, or DM me if you’d like to discuss AI infrastructure economics further.   #AI #Infrastructure #GPU #Computing #DataCenters #Startup #Opportunity #AIEconomics 

Hold your ground! Hold your ground! Sons of Gondor, of Rohan, my brothers, I see in your eyes the same fear that would take the heart of me. A day may come when the courage of men fails, when we forsake our friends and break all bonds of fellowship, but it is not this day. An hour of wolves and shattered shields, when the age of men comes crashing down, but it is not this day! This day we launch on Product Hunt!! By all that you hold dear on this good Earth, I bid you support our launch, Men of the West!!!

I discovered an incredible resource for students diving into AI and machine learning. (You won't want to miss this) Today, I learned about Thunder Compute from Carl Peterson (CEO) and Brian Model (CTO), Founders of Thunder Compute Thunder Compute is a Y-combinator-backed startup building GPU virtualization technology used to power a GPU cloud platform. Their insights inspired me to explore Thunder Compute. This platform offers a student-friendly solution that’s truly remarkable: → $20 in free credits equals around 39 hours of A100 GPU time for educational purposes! Here’s what sets it apart: • You can connect your own code editor, like Cursor or VSCode. • A100 GPU access is just $0.57 per hour. That’s 80% cheaper than AWS or GCP! • The command-line interface is so straightforward; you can get started in just 2 minutes. I documented my experience in a blog post, outlining my entire workflow. If you’re tired of Colab timeouts or worried about costs while learning ML, give this a serious look. Thanks to Tejomay Kapadia for hosting this amazing event. Does the idea of connecting your own code editor to a powerful GPU cloud appeal to you? Check out my walkthrough: https://lnkd.in/eTYTYnum #AI #MachineLearning #GPU #CloudComputing #Students #YCombinator #DeepLearning #DataScience #A100 #EdTech

It was a pleasure hosting Carl Peterson and Brian Model of Thunder Compute at the Pace Data Science Club yesterday! In a lively interactive session, more than 75 deep learning enthusiasts learned how to save 70% on their GPU bill using Thunder Compute's tech. Aside from the cost-saving benefits, what really appeals to me is the environmental impact of their solution. I was extremely impressed by Carl and Brian's thorough technical expertise and ability to explain their product intuitively. They're also kind people who patiently answered all our questions and walked us through the setup process. A huge shoutout to Darsh Joshi for bringing this opportunity to the club and for his support during the event! I can't wait for the incredible deep learning projects that will come from the Pace Data Science Club accelerated by Thunder Compute ⚡️

I had the opportunity to attend the "Accelerated Deep Learning with Thunder Compute" event, organized by Thunder Compute and Pace Data Science Club (Tejomay Kapadia Darsh Joshi) Thunder Compute provides a cost-effective solution for running GPU workloads by leveraging its proprietary virtualization technology to maximize GPU efficiency. It was a great experience connecting with Carl Peterson and Brian Model. For anyone looking to cut GPU costs without sacrificing performance, Thunder Compute is a game-changer. #DeepLearning #AI #ThunderCompute #DataScience