Scaling Robotics is starting to look a lot like scaling software. Over the past months we went through more iterations than we’d like to admit, all trying to answer one question: how does hardware get funded, get to market, and survive long enough to ship? Software can launch from a laptop. Anything that has to physically exist needs money (and lots of it), specialized talent, lab space, testing, manufacturing, distribution, the right connections, and time before it can prove anything. Early on, one thing was clear: we did not want to build another launchpad. We wanted a home for robotics onchain. That meant finding the right partners, getting the tokenomics right, and giving early-stage builders everything they need to go from idea to production. A few years ago, none of this would have worked. Why now? To answer that, we have to go back to 2020, when @OpenAI released GPT-3. Most people outside tech didn’t care yet, but if you were playing with it, it was hard to ignore, because a language model could suddenly learn a new task from just one or two examples. There was no retraining involved, you just showed it what you wanted and it figured out the rest. It felt familiar for a reason. Humans have a remarkable ability to pick up new skills from very little. Watch someone tie a knot once and you can probably do it. A new barista watches one latte pour, and they’re halfway there. In 2020, for the first time, software started learning the same way. From there, it took off. Within a couple of years, that same ability was writing code, building apps and designing games. Anyone with an idea could ship something on a screen. But it all stayed on the screen. Fast forward to today, and the same shift is now happening in the physical world. Robots are starting to learn the way GPT-3 did. The clearest example is @GeneralistAI’s GEN-1.5. Show it a few seconds of someone opening a jar or unzipping a pouch, and the robot does it. No new code, no weeks of training. It learns a physical skill the way the barista learned the latte. The hardware is getting cheaper just as fast. In 2023 a two-arm research rig cost close to $20,000 USD. Right now, an open-source robot arm costs about $230. That’s why we, along with roboticists across the globe, believe we’re very close to a ChatGPT moment for robotics. Where teaching a machine means showing it, and the parts cost less than a phone, building a robot stops being something only labs can do. So what’s stopping someone from starting a hardware company? Money. A prototype has never been cheaper, but turning it into a scalable product still needs manufacturing, certification, distribution and a team, long before the first unit sells. It’s not that nobody is funding robotics. Robotics startups raised $18.8B in the first half of this year, more than in all of 2025 and more than the 2021 record. But most of it went to a few giant rounds, like @SkildAI’s $1.4B in January. Across venture, over 40% of seed and Series A money now goes into $100M+ rounds. Right as the tools reach everyone, the money is moving further away from them. Nobody is writing a $100M check for a garage prototype. That’s the gap MachineFun is built for. Our vision is that with greater efficiency in hardware scaling, small start ups, prototypes, and passionate entrepreneur now have the ability to create value at the micro level. By providing simulations tooling like SimArena or verifying your robot through peaqOS, founders can now have early backers follow their journey in public while demonstrating evidence through the provided tooling. By giving founders a platform at the foundational level, we see this as a key stepping stone to reaching mass scalability. The raise is only the start. We also support with what comes after it: getting to market, logistics, and introductions to the manufacturers and partners who can help you build at scale. If you have a machine, we want to make it fun.

Scaling past a trillion parameters requires inference engineers to solve 3 issues: - Width: Routing between hundreds of sparse experts - Depth: Mixing global and local linear attention - Length: Processing 1M-token sequences I asked Ali and Bryce to draw it up. They delivered.

Scaling agents shouldn't mean mystery token bills or shared human credentials. Natasha Balasubramanian unpacks Google Cloud’s agent controls: 1. Smart Routing + Cloud Billing spend caps 2. Agent Registry & cryptographic Identity 3. Sandbox VMs, Agent Gateway & Observability

Scaling up carbon dioxide removal is among four major environmental ambitions that have fallen short, in a world that’s blowing past targets for limiting global temperature rise

Scaling an AI agent fleet? We provide full-stack agentic solutions, and now we're expanding our portfolio with Filestore agent volumes: fully managed, elastic file storage purpose-built to give agents isolated, economical, high-performance workspaces →

Scaling live video shouldn't come with an "integration tax." See how Layercake revolutionized sports streaming with Cloudflare, slashing operational costs by 85%, cutting setup time by 95% (down to 10 seconds), and seamlessly orchestrating 300+ live events at once.