Axelera AI

If you train YOLO models, there's a one-line path from @Ultralytics straight to a Metis-ready file: model.export(format="axelera") Custom weights work exactly the same way. From there, a handful of lines of Python takes you from an image to detections. This landed in the Voyager 1.6 docs as part of the new Pipeline Builder. Worth a look if Ultralytics is already in your stack. Write-up and links on the community: eu1.hubs.ly/H0wKK1K0 #EdgeAI #YOLO

The instinct in computer vision is to build one clever, general-purpose model that handles everything. WG Tech Solutions did the opposite, and the reasoning is genuinely useful. Conventional wisdom says build one clever general model. WG Tech did the opposite: ~45 task-specific @ultralytics YOLO models on Axelera Metis, each tuned to its own camera angle and lighting. For safety-critical work, that's the smarter call 👇 eu1.hubs.ly/H0wFSGY0 #EdgeAI #ComputerVision #IndustrialAI

@bjbaumann2014 @furrier @theCUBE @GemmaAllenSays thecube.net/events/nyse/ra…

Heading to Paris. Our own Fabrizio Del Maffeo joins the Raise Summit 26 stage next week to talk purpose-built AI hardware and what it takes to run inference at scale. Thank you @bjbaumann2014 @Furrier @theCUBE @GemmaAllenSays for hosting. Link below. #AIInference #RaiseSummit26 #AIInfrastructure

Drone operations have a scaling problem: 1 drone usually equals 1 operator. DroneStar AI is changing that math. Read the full story: eu1.hubs.ly/H0wB7nR0 By embedding the Axelera Metis AIPU onboard, a single operator can now supervise an entire pack of 5 to 10 autonomous drones. Here is how they are doing it: - Real-Time Edge AI: Each aircraft runs real-time object detection and tracking on 1280×1080 video at 20 FPS in flight. - Uncompromising Efficiency: The Metis M.2 module delivers the heavy computation required for full autonomy within strict drone size, weight, and power constraints. - Traceable Hardware: A fully transparent supply chain ensures compliance as they expand into critical infrastructure, security, and search & rescue. From scaling mountain rescue operations to automated wildfire detection, we are proud to power DroneStar AI's mission. #Drones #AIaccelerators #searchandrescue #AIPU

A full voice assistant, running with no internet connection at all, on a tiny, self-contained device you can put anywhere. This is the new Axelera AI Mini PC running Llama 3.2 1B as the language model, with separate speech-to-text and text-to-speech models alongside it, all on device. Ask it a question out loud and it answers out loud. Nothing leaves the box. Metis M.2 Max plus Intel Core Ultra, no need for a GPU, 25+ simultaneous 1080p streams for vision work, and 32GB DDR5 with a 256GB NVMe SSD ready to run out of the box. The point is this isn't only for vision. Although it's great for that too! Post your projects or ask us anything on the community. Register now to join in. 👇 eu1.hubs.ly/H0wvMtW0 #EdgeAI #LocalLLM #ComputerVision

We just launched the Axelera AI Mini PC. Powered by the Metis M.2 Max accelerator. 25+ simultaneous video streams. 15 TOPS/W. No GPU. Production-grade AI vision inference in a form factor that fits anywhere. Order today, deploy this week. eu1.hubs.ly/H0wsjgZ0 #EdgeAI #ComputerVision #AIHardware #MiniPC #AxeleraAI

Deploying AI models to embedded targets usually means a full SDK on a device that doesn't necessarily have room for it. So Voyager 1.6 splits that workflow in two. You compile your model on a development machine using axelera-devkit, which produces an .axm file. Then you run it on the target with axelera-rt, a lightweight pip-installable runtime. The devkit doesn't need to be anywhere near the deployment hardware. For anyone putting Metis into space-constrained or awkward-to-access targets, that's a meaningful change. Compile on your beefy dev box, copy the .axm across, run. Find out how to get this up and running in no time: eu1.hubs.ly/H0vZ5-j0 #EdgeAI #VoyagerSDK

Container yard operations demand speed, accuracy, and airtight data security. This video shows real-time Vision AI at the edge: arrival tracking, structural damage detection, and space optimization, all running locally on Metis and SMARTSTATION hardware. No cloud dependency. Near-zero latency. Complete on-prem data privacy. eu1.hubs.ly/H0wc79z0 #EdgeAI #VisionAI #AxeleraAI

Click-and-collect ANPR on a Metis M.2 has never been more achievable 🚗 YOLO finds the plate, LPRNet reads it. LPRNet hits 9,581 FPS on M.2, so plate recognition is basically free and the card sits idle for whatever else you want to run on the same device. Pipeline layout here + two other POS builds for anyone in smart-retail: eu1.hubs.ly/H0vNgkw0 #EdgeAI #ANPR

The "banana trick" at self-checkout works by ringing up a £4 avocado as a 40p onion 🍌 The fix is a simple AI cascade. @Ultralytics YOLO detects each item, a classifier (EfficientNet, MobileNetV4, etc) tells you what it actually is, and then cross-checks against the POS scan log. What used to be a difficult task now runs easily on a Metis M.2: yolov8s-coco: 531 FPS yolo26s-coco: 394 FPS Full build options right here: eu1.hubs.ly/H0vNkqw0 #EdgeAI #ComputerVision

Watch the full video here: youtu.be/T8X8DwkWP7E?si…

Small form factor. Serious AI performance. SMARTSTATION CEO Manish Dave joins Axelera CMO Alexis Crowell to discuss how pairing compact, ruggedized hardware with the Metis AIPU enables scalable edge deployments across manufacturing floors, vehicles, and harsh environments. #EdgeAI #AxeleraAI #HardwareEngineering

Last week our CFO Marta Ostroumoff joined a stellar panel at #HelloTomorrow Summit 2026 in Amsterdam: “From 1st‑of‑a‑kind to Nth‑of‑a‑kind.” She shared how Axelera AI is scaling from breakthrough prototypes to repeatable, industrial‑grade edge AI deployments, across manufacturing, capital, and teams. On to the next N‑of‑a‑kind wins. #DeepTech #EdgeAI #AIInference #Semiconductors #ScaleUp #EuropeanTech

Edge AI should be powerful enough for industrial workloads and compact enough to deploy anywhere. We have partnered with SMARTSTATION to integrate our Metis AIPU into next-gen AI PCs. Real-time computer vision made accessible, without cloud dependencies. Link: eu1.hubs.ly/H0w70g40 #EdgeAI #ComputerVision #IndustrialAI

⚽ Scrappy living-room goal-line tech, built on @AxeleraAI Metis: - Stock COCO @ultralytics yolo26s, ball = "sports ball" class - ~250 fps detection on Metis (was the bottleneck on laptop, now it's OpenCV drawing) - Goal only counts if it crosses the line between the posts A full-on AI referee at home, next stop the #WorldCup! Full build:eu1.hubs.ly/H0w4Y1w0 #EdgeAI #ComputerVision #FIFA #Football #Soccer

Our CEO Fabrizio Del Maffeo takes the stage at VivaTech Paris this Wednesday for a founder's view on what it takes to find product-market fit. Wednesday June 17, 4:45 PM, Founders Hall. #AxeleraAI #VivaTech2026 #Founders #ProductMarketFit

A checkout transaction lasts one or two seconds. If item-verification has to round-trip to a cloud GPU and back, the moment is gone before the answer arrives. It's the same reason POS itself moved to the edge years ago. The network drops, but the till can NOT. Vision AI inherits exactly that constraint: streaming raw video off-site for inference falls over on cost, latency, and the days the uplink has a wobble. So the brief writes itself. Local inference, low watts, works with the cameras already up there, cheap enough per lane that the rollout actually escapes the pilot phase. Three reference builds (loss prevention, shelf analytics, click-and-collect) for anyone building in the smart-retail space right now 👇 eu1.hubs.ly/H0vMDCw0 #EdgeAI #Retail

This excellent case study by @WGTech and @Ultralytics demonstrates 28% fewer safety violations, which sounds like an AI accuracy story. It more than that, though. The model isn't catching MORE incidents. It's preventing them. Which means workers are changing behaviour, and that's a feedback loop story rather a detection story. The loop only works because inference happens at the edge. Send the video to the cloud, wait a few seconds, the unsafe moment is already over and the chance to correct it is gone. Local processing is what makes the behaviour changes stick. Great write-up from on this exciting deployment in India, so don't miss it 👇 eu1.hubs.ly/H0vQTQk0 #EdgeAI #ComputerVision

We are at Bosch Connected World in Berlin this week. Booth 16, Future Tech. June 10 to 11. Also joining the programme: a keynote from our VP Strategy on Europe's moment in Physical AI, and a panel on integrating AI acceleration into next generation mechatronic systems. #AxeleraAI #BoschConnectedWorld #EdgeAI #PhysicalAI