Edge Computing & Inference
Small language models are landing on HMI panels and air-gapped edge boxes
Rockwell Automation's FactoryTalk Design Studio Copilot now runs Nemotron-Nano-9B-v2, a 9-billion-parameter open-source model from NVIDIA, directly on HMI panels and air-gapped edge appliances on factory floors.
On-prem AI with no cloud dependency — the security objection to shop-floor AI is quietly disappearing.
IIoT & Connected Manufacturing
Agentic AI in manufacturing: 171% average ROI, against a cautious forecast
Manufacturers that ship AI agents into production report an average 171 per cent return on investment and a 34 per cent gain in production efficiency, according to 2025-2026 enterprise deployment data. The piece sets those numbers against a Gartner forecast.
Treat vendor ROI figures as a ceiling, not a plan — scope one agent against one measurable loss before buying a platform.
Robotics & Cobots
Hirebotics debuts an explosion-proof, no-code painting cobot
Hirebotics has debuted an explosion-proof painting cobot that combines FANUC's CRX-10iA/L paint hardware with its Beacon platform, eliminating the need for dedicated paint cells and complex programming.
Paint is one of the last hard-to-automate SME jobs — no dedicated cell and no programmer changes the payback math.
Process Technology
METTLER TOLEDO targets four-week delivery on product inspection systems
METTLER TOLEDO's LightningBuild program uses standardized engineering, inventory management and production planning to shorten equipment delivery times, targeting four-week delivery on product inspection systems.
Lead time is becoming a spec line — if a vendor quotes 20 weeks, treat that as a negotiating position, not a fact.
Automation & Controls
Orama.AOI and BrainChip put low-power defect detection on the line
Orama.AOI is working with BrainChip to retrain and optimize BrainChip's Akida models on Orama.AOI's industrial inspection datasets. The collaboration reports robust defect-detection performance and positions Akida for inspection at the edge.
Low-power inference on the line means visual inspection without a GPU box or a network round-trip.
Smart Manufacturing & Digital Twin
Tesla's Optimus: 1,000+ units deployed, zero external sales, no uptime data
Tesla's Optimus program holds an estimated 1,000 to 1,200 humanoid robot units deployed across Fremont and Giga Texas as of mid-2026, yet the company reports zero external sales and publishes no uptime figures.
Deployed unit count is not proof of production value — ask any humanoid vendor for uptime and cycle data instead.