VC10X Pulse - Etched Just Doubled Its Valuation to $21B in 1 Month — Here's What They're Building

One month ago, Edged was valued at about ten billion dollars. Today, it's worth twenty-one billion dollars. That's more than doubling in less than a month, and the company has just raised another seven hundred million dollars led by Jane Street. But here's what's really interesting. Edged isn't trying to build another Nvidia. It's trying to solve a very specific problem in AI, inference. And if Edged is right about where AI computing is going, this could become one of the most important semiconductor stories. Let's dig in. What is Edged actually building? The easiest way to understand Edged is it's building specialized AI infrastructure designed specifically for running AI models, not training them. When you ask ChatGPT a question, that's inference. When millions of users interact with an AI model simultaneously, you're doing enormous amounts of inference. And as AI agents, voice AI, and real-time applications grow, inference could become a massive portion of total AI compute. Edged believe general purpose GPUs aren't necessarily the optimal architecture for this future. So it's building custom silicon and complete rack scale inference systems optimized around the characteristics of modern AI models. But why not just use Nvidia? This is the central question. Nvidia GPUs are incredibly flexible. They can train models, run inference, support different architectures, and they are backed by CUDA and an enormous software ecosystem. But flexibility comes with overhead. Edged thesis is essentially if you know what exactly your workload is, why pay for all that flexibility? Instead of building a chip that can do everything, build a system optimized for the workloads that matter the most. That's the ASIC argument, and we have seen this before. Google built TPUs, Amazon built Trainium, Broadcom builds custom AI accelerators. The difference is that Edged is going after inference with a particularly aggressive specialization strategy. Now let's talk about the original Edged bet, which was extremely narrow. The company's original Soho architecture was designed around transformers. That's an enormous bet on the continuing dominance of transformer-based AI. The logic was simple. ChatGPT uses transformers. Claude uses transformers. Gemini uses transformers. Most major frontier models have been built around transformer architectures. If transformers remain dominant, specialized hardware could potentially deliver dramatically better performance per dollar and per watt. But there's an obvious risk. What if the architecture changes? That's one of the biggest things investors need to understand. A GPU can adapt. A specialized ASIC is much less flexible. So Edged is making a very concentrated technological bet. But Edged is now building more than just a chip. This is an important evolution. Edged now describes its products as frontier inference clusters. That means the company isn't simply selling a processor, it's co-designing chips, memory, packages, PCBs, interconnects, cooling, software, and entire rack scale systems. The company's argument is that the unit of AI computing is increasingly moving from the chip to the rack and eventually the cluster. That's