Arena

Thrilled to have Catherine Kauber join us in the Arena, driving forward our deployments of Atlas! Interested in accelerating humanity’s rate of hardware innovation? We’re hiring across roles - learn more at arena-ai.com/careers.

When software eats the world, what happens to the physical one? There’s been a surge of new tools aimed at physical product development across industries like aerospace, defense, robotics, and consumer hardware. Interest and funding in the space are way up, but it's still hard to tell where many of these tools fit or what exactly they’re trying to replace. (The tools are better, but is the work actually happening faster?) At Hardware FYI, we mapped the full development cycle from requirements to production ramp and plotted where each of these startups fit. Full list of companies sorted by development stage and category linked here: https://tally.so/r/wv8avQ

Thrilled to have you join us in the Arena, Tommaso Dreossi!

Every rocket scrub, EV recall, or satellite blackout shares a hidden culprit: stray electromagnetic (EM) fields. Over the next decade they’ll quietly vaporize $1 Trillion. Here’s why—and how AI is turning the tide. Modern hardware – rockets, satellites, drones and data centers – relies on one thing—pushing electrons, photons, or plasma exactly where we want. Each is just a small packet of electromagnetism. 🔎 The hidden cost • ≈ $200 B/year on EM sims + lab tests (half are reruns) • Schedule slips, respins & scrap add $800 B+ • 10 year total: >$1 T quietly lost ⚠️ Why complexity keeps winning A 3 nm chip moves 6 × 10¹⁷ electrons per ms through traces thinner than a virus, feeling every Wi-Fi burst, engine spark, and solar flare. One unmodeled ripple and satellites drop links or rockets miss windows. 🚀 What changes next 1️⃣ Expertise → Digital copilots — scarce EEs pair with AI that reads schematics, writes test plans and analyzes test results in minutes, not days. Tribal knowledge is encoded, and new grads climb the curve faster than ever before. 2️⃣ Closed-loop optimization — AI tunes firmware, layout, even material stacks to squeeze more performance per watt, and more resilience, with fewer (and in many cases zero) humans in the loop. 3️⃣ AI-discovered basis sets — automatically learned embeddings become the new bases for modeling physics. Generating a new basis set used to be rate limited by the appearance of a once-in-a-generation genius like Fourier or Chebyshev. Auto-learned bases could accelerate applied physics enormously. Teams across the industry—ours at Arena included—are racing to unlock this. If stray fields dictate your schedule, let’s trade notes. Here's a fuller writeup for the curious: https://lnkd.in/gTZ7Gtfp

Congratulations Bausch + Lomb and Arena on their partnership! Two best-in-class teams coming together to elevate what's possible. Wishing you both continued success!

Congratulations to Arena's Distinguished Researcher Siddharth Garg for winning Best Paper at the 62nd DAC, The Chips to Systems Conference! Verigen is the first comprehensive dataset and LLM fine-tuned model for Verilog code generation, driving new research in LLM-based hardware design, including the VeriThoughts model family, which possesses advanced reasoning capabilities.

The most sophisticated ‘AI’ in last week’s B-2 strike on Iran wasn’t an LLM. It was a 60-year-old math algorithm hard-wired into a specialized chip the size of a postage stamp. Last week, seven B-2 Spirits just dropped 12x 30,000-lb GBU-57 bunker-busters on Iranian nuclear tunnels. The guidance computer wasn’t running a large-language model—it ran a 60-year-old algorithm that still rules real-world autonomy. 1️⃣ What’s inside the bomb? The Kalman filter, invented for the Apollo program, fuses gyro, GPS and radar data—updating dozens of times every second—to keep a 30k lb warhead inside a < 2 m target window. 2️⃣ Why does it have to think so fast? The bomb drops at Mach 1 ≈ 340 m/s → so it travels 0.34 m every ms. A 2m precision requirement determines the max possible drift. So, 2 m ÷ 0.34 m/ms ≈ 6 ms Which means we need 170 updates per second or you miss the tunnel. 3️⃣ Why can’t we use a CPU or GPU? General-purpose processors can stall for > 10 ms under normal load —> mission failure. Guidance engineers hard-wire the Kalman filter math into radiation-hardened FPGA / ASIC silicon (e.g. AMD Xilinx). A typical FPGA can finish a full update in 240 µs. 4️⃣ The talent bottleneck Faster and more precise systems demand more innovation for systems like this. But, there's a talent gap. Almost 1 in 3 U.S. engineering roles sit vacant—and electrical-control seats are the hardest to fill. Re-industrializing the economy isn’t just welders and CNCs; it’s EE brains that master nanosecond-tight loops. 5️⃣ The only scalable fix is AI Rebuilding a deep bench of veteran EEs will take a decade—too slow for today’s great-power competition and supply-chain shocks. AI-augmented engineers are the path forward. At Arena we’re building Atlas, an AI Electrical Engineer that can auto-tune Kalman filters validate power rails, optimize RF layouts Bring your learnings from the world of bits to the realm of atoms. Come build with us. 🚀🔧

The future is self-healing. Learn more about our open roles at arena-ai.com/careers.

Few projects in the world mirror the ambition and potential for transformative human benefit of ITER. So as a physicist I was particularly excited to contribute some thoughts and potential ways our AI Engineer ATLAS can speed up HW development, not only at ITER but at HW companies across the globe. Thank you ITER Organization, Laban Coblentz for inviting Arena, and ComputerWeekly.com and Pat Brans for covering our talk at the 2025 ITER Organization Private Sector Workshop in "Fusion and AI: How private sector tech is powering progress at ITER". Read more here: https://lnkd.in/e5mu9v-9

At the ITER Organization Private Sector Fusion Workshop in Cadarache in April, three organisations – Microsoft Research, Arena and Brigantium Engineering – shared how artificial intelligence (AI), already transforming everything from language models to logistics, is now stepping into a new role: helping humanity to unlock the power of nuclear fusion. Kenji Takeda, Michael Frei, Lynton Sutton, Laban Coblentz