The AI Revolutionaries: Ex-OpenAI Brains Building a Leaner, Meaner Open-Source Future

Imagine leaving the most talked-about AI company on the planet to build something even better, something that challenges the very giants you once helped create. That’s precisely the audacious path taken by Yash Patil, Rhythm Garg, and Linden Li. These three young, brilliant minds, all under 25, are the co-founders of Applied Compute, a startup that’s rapidly making waves by offering enterprises a refreshingly cost-effective and highly customized alternative to the big-name proprietary AI models dominating the market. Their mission? To democratize powerful AI, making it more accessible and tailored for businesses hungry for innovation without breaking the bank.

Their story, really, begins with a pivotal moment in early 2025. You see, the tech world was buzzing when a relatively small Chinese lab, DeepSeek, unveiled its R1 open-source model. It wasn't just another release; it "stunned Silicon Valley," prompting influential figures like Marc Andreessen to declare it a "Sputnik moment" for the AI industry. It was a clear "wake-up call" that open-source could truly compete, even thrive. For Patil, Garg, and Li, then still at OpenAI, this was more than just news—it was an affirmation, a spark that ignited their entrepreneurial spirit. Just five months after DeepSeek R1's debut, they decided to strike out on their own, fueled by a belief that their deep expertise, honed at OpenAI, could be leveraged to optimize and deploy these emerging open-source models for real-world enterprise needs.

So, what exactly does Applied Compute do? Well, they're not just reselling existing models. Instead, they’re masters of customization. They work intimately with large enterprises, consulting closely with their engineers to train bespoke, open-source AI models. Think of it: taking powerful foundational models, perhaps like Alibaba's Qwen or Moonshot's Kimi, and then meticulously fine-tuning them with reinforcement learning on a client's own internal, proprietary data. This ensures the AI isn't just smart, it's smart for that specific business. Once trained, these specialized models run seamlessly on Applied Compute's own cloud infrastructure. And for those clients who prefer a more hands-on approach, they've even released AC2, a platform that empowers companies to utilize Applied Compute's internal training tools themselves, offering an even more budget-friendly option than their consulting services. It's about giving businesses choice and control.

The core appeal, beyond the customization, is undeniably the economics. In an era where AI expenses can quickly spiral, Applied Compute offers a breath of fresh air. We’re talking about significant cost savings—Yash Patil himself suggests their models can be up to 40 times cheaper per million tokens than some proprietary alternatives, or at least 10 times cheaper in many scenarios. But it's not just about saving money; it's about superior performance where it counts. Take DoorDash, for instance: Applied Compute built a specialized model for them that actually outperformed some of the more general "frontier" models on a very specific, critical task. Another impressive case? Harvey, a legal AI startup, needed an agent. Their head of applied research, Nico Grupen, noted that building it with Applied Compute took less than two months—a process previously estimated to take "several months to years." That's not just efficiency; that's a game-changer.

It's no wonder, then, that Applied Compute's trajectory has been nothing short of meteoric. They secured a solid $20 million seed round back in June 2025, which, let's be honest, is a hefty sum for a seed. But things really kicked into high gear in 2026. By June, their valuation had already soared to an impressive $1.3 billion. And get this: Microsoft CEO Satya Nadella himself saw fit to interview Yash Patil that very month, a clear sign of the industry's keen interest. Fast forward to September 2026, and the company is reportedly in talks to raise a whopping $350 million, a deal that would catapult their valuation to an eye-watering $3.25 billion. They're clearly attracting big names, not just in funding, but as clients too, boasting partnerships with DoorDash, Nvidia, Microsoft, Cognition, Mercor, and even the legal AI innovator Harvey. Eli Gil, a lead investor, rightfully highlights the founders' unique expertise from OpenAI as a critical factor in their ability to so effectively optimize open-source models.

What Applied Compute is doing isn't just building a company; they're helping to redefine the future of enterprise AI. As businesses increasingly look for specialized, cost-effective solutions that can be tailored precisely to their needs, the appeal of open-source models—especially those expertly fine-tuned by some of the sharpest minds from the AI frontier—is undeniable. These young founders, having emerged from the heart of the AI revolution, are now showing the world that innovation doesn't always mean bigger, more expensive proprietary systems. Sometimes, it means smarter, more agile, and openly accessible alternatives that truly empower businesses. It's a fascinating shift, and Applied Compute is right at the vanguard.