Lior Susan

Super excited to share the news that Ursa Major is going public! Congrats to the entire team and everyone who's made this success possible. To some, this might seem like the end of the road, for Ursa this is just the beginning! Thanks for having us as a partner all these years. 🫡🇺🇸

👀 How do you build a machine vision system that can operate reliably in real-world assembly environments? 🤖 Tori Colthurst, Robot Perception Engineer at Bright Machines, discusses the challenges of robot perception within the Bright Factory system, and how Beckhoff’s EtherCAT technology enables the high-speed communication, low-latency data exchange, and real-time control required for accurate navigation, inspection, and automated assembly processes. 🚀 EtherCAT allows engineers to process massive amounts of data, scale automation systems, accelerate deployment, and quickly diagnose issues in production environments. 📚 Read the full case study to discover how Beckhoff and EtherCAT help Bright Machines push the boundaries of what's possible, not only in assembly, but across the entire manufacturing landscape: https://loom.ly/svBAiuY #EtherCAT | #TwinCAT | #AI | #Assembly | #Beckhoff | #BrightMachines

We're excited to lead the round in Reframe Systems. Reframe stood out to us for applying a unique combination of Physical AI and manufacturing discipline to one of the largest and least productive sectors of the economy. The company is helping create an era of housing abundance by making it possible to build more homes, more quickly, and more affordably. We believe that combination of mission and execution positions the company to become a category leader in housing production. https://lnkd.in/gkGQXjNz

Reframe Systems is industrializing homebuilding and  expanding across North America, building resilient homes in communities people want to live. We just closed $40M in financing led by Energy Impact Partners (EIP), with co-investments from Counterpart Ventures, E12 Ventures, Global Brain Corporation, Thin Line Capital, UP.Partners and Los Angeles Cleantech Incubator (LACI), and continued support from existing investors, including Eclipse, VoLo Earth Ventures, Cubit Capital, RA Capital Management (Planetary Health), MassMutual Catalyst Funds and Nor'easter Ventures. The U.S. is short an estimated 4.5 million homes, and the construction industry is short more than 500,000 skilled workers. We started Reframe because we believe robotics and physical AI can actually solve this, not just make the same process marginally faster. Homes should be affordable and quick to build in the communities where people want to live. Our microfactories are delivering homes 3x faster while reducing costs by more than 20% compared with traditional construction, and this expansion is how we meet the demand across more communities. We have 10 homes built, and another 114 expected this year, while FAB1, our New Massachusetts microfactory, will produce up to 500 multifamily units or 250 single-family homes annually. Grateful to our investors for backing this vision, and huge congrats to the entire Reframe team building the engine of housing abundance and moving towards our goal of 1M homes by 2040. Thank you to Rocket Drew at The Information for the exclusive on this milestone. Story linked in the comments. Aaron Small Felipe Polido Justin Hersh Keith Feldman, CFA, CPA Stacey Harris Amy Villeneuve Charly M. Kareem Dabbagh Anil Achyuta Lior Susan Aidan Madigan-Curtis Michael Campos Aaron Fyke Philip Carson Jason Allen, JD Kathryn Meng Jinya Kurita Paulina Szyzdek Patrick Eggen Aria Cyrus Safar

At Cerebras, we were among the first chip companies to use water cooling in data centers. When I first proposed it, it sounded like heresy. But while Cerebras was still in stealth, we knew that water cooling would be essential. A chip the size of a dinner plate dissipates an enormous amount of power. There was no practical way to cool the wafer directly with air. Water had to carry the heat away from the wafer. That heat could then be transferred either to facility water or to the surrounding air through a large radiator. But even assuming no redundancy, removing that heat with air would have required a full rack—48 RU of radiators, heatsinks, and fans—to cool a single system. The only practical solution was to transfer the heat directly into the facility-water system. We designed a cold plate with water circulating through internal channels. Heat flows from the silicon, through a thermal-interface material, into the cold plate, and finally into the water. Sounds simple. It wasn't. Silicon and metal expand at different rates. Bond the wafer rigidly to the cold plate and the resulting stresses can crack the silicon. The interface had to conduct heat efficiently while allowing microscopic relative motion. That meant compression. A conventional chip uses four spring-loaded screws—one at each corner—to maintain the required force despite manufacturing tolerances and temperature changes. That works because the span between the screws is small. At wafer scale, the mechanics change completely. With screws only around the perimeter, both the board and the cold plate bow between attachment points. Some regions are over-compressed. Others barely touch. Even tiny variations in thermal-interface thickness increase thermal resistance and create hot spots. The solution wasn't to eliminate the screws. It was to multiply them. We distributed an array of spring-loaded screws across the entire wafer-scale assembly. Each screw shortened the unsupported span and controlled the compression locally. Instead of asking four corner screws to keep a dinner-plate-sized surface flat, we divided the interface into many much smaller mechanically controlled regions. That introduced another set of challenges. The screw locations had to coexist with the wafer, the board, and the cooling channels. The springs had to provide enough force for excellent thermal contact—but never enough to damage the silicon. The cold plate also had to remain stiff and flat while distributing water uniformly across the wafer. There was no off-the-shelf solution. So we designed the cold plate, the thermal-interface material, the spring-loaded compression system, the mechanical stack, and the water paths as one integrated assembly. Because of the Cerebras team's persistence and ownership, it worked. Today, direct liquid cooling is becoming standard for high-performance AI systems. But getting there wasn't quick. Innovation rarely is.

From Store to Operating System: What Kroger’s 500th Robot Could Signal for the Future of Retail In our opinion, Kroger’s deployment of its 500th Simbe Tally robot is more than a robotics milestone, it is a signal that autonomous shelf intelligence is crossing the threshold from experimentation to core operating infrastructure. The strategic implication is significant: as real-time shelf visibility becomes embedded into daily store operations, retailers are not simply automating tasks; they are beginning to fundamentally reshape the economics, productivity and management model of the store. That distinction is critical. Kroger is not automating the store, it is making the store visible in real time. That is a far more consequential shift. Hundreds of locations are gaining the ability to continuously detect what legacy retail systems routinely miss: lost availability, misplaced inventory, pricing errors, failed promotions and demand that disappears at the shelf. In-store robotics based real-time shelf intelligence closes those shelf gaps and creates three strategic consequences. First, availability shifts from an operational problem to a managed economic outcome. Out-of-stocks are no longer treated as an inevitable cost of doing business; they become visible, measurable exceptions, with an owner, a response time and a quantifiable revenue impact. That changes the management equation: retailers can direct labor and intervention toward the availability gaps that matter most, protecting sales, strengthening customer trust and improving the economics of every shelf. Second, labor shifts from inspection to intervention. Associates spend less time searching the store for problems and more time resolving the issues that matter, restoring availability, improving fresh execution and serving shoppers. The economic prize is not simply fewer labor hours; it is a step-change in productivity, with every store hour deployed against higher-value work. Third, the retailer-supplier collaborative relationship gains a shared source of operational truth. Both parties can see, with far greater precision, whether products were available, promotions were executed and planograms were followed. That shifts the conversation from debating what happened to jointly addressing why it happened, reducing friction, accelerating corrective action and improving the economics of execution for both sides. Kroger’s 500th robot should be viewed as an industry warning shot. The instrumented store is no longer theoretical; it is becoming the new operating standard. The winners will not be defined by store count and/or fleet size, but by how quickly they convert continuous shelf intelligence into better decisions, faster execution, and stronger economics. In our opinion, the mandate is clear: instrument the store, redesign the operating model, and move now. In real-time retail, speed is becoming a structural advantage, and waiting will only widen the gap.

At the start of this year, the gates of Buckingham Palace opened to one of Wayve’s driverless cars carrying Alex Kendall and his New Zealand family. In the grandiose Throne Room, beefeaters on guard, Princess Anne presented him with an OBE for services to AI. From writing the first code, to hiring a team, to finding customers and investors, to commercialising the business, Kendall’s role has changed wildly since his early Cambridge days. “Now the interesting challenge for me is leading a global executive team, managing world-class leaders at a global scale, and I love that,” he said. He speaks of creating a “generational company”, of seeing Wayve’s technology going beyond driverless cars, into robotics more broadly, in factories and in the home. Almost a decade after Wayve was founded, as he turns 34 this August, Kendall said: “Physical AI is about to have its moment. I feel privileged to be able to play a role in that.” Read more 👇 https://lnkd.in/eGKqfAYZ

CS4 is just the beginning of ultrafast inference. Check out the Cerebras product roadmap to see how we're pushing inference speeds even faster in future releases.

Rare look under the hood of the new CS-4 at Cerebras Supernova 2026. From rack plumbing to the wafer cold plates, wafer-scale engineering built entirely around the goal of ultra-low latency inference. ⚡️ Congratulations to Andrew Feldman, Andy Hock and the DevRel team on this exciting event. #Compute #Hardware #AI #FastInference #LLM #Supernova2026 #GPU