Editor's note
This week, AI moved into the hardware.
Most of the AI we talk about, including what I build, lives in software: models that read data, plan and predict. This week showed the other side of it, AI wired straight into physical kit. A robot arm that can lift a living embryo without crushing it. A grid of sensors that has logged vibration for years and now gets read in real time. The model still matters, but here it is only as good as the hand or the sensor it runs on.
That is a shift worth paying attention to, because this kind of AI has to survive contact with the real world, not just a clean dataset. It will be interesting to see which of these pairings actually holds up once it leaves the pilot.
Axel
WHAT GOT ME THINKING
SweTree and Prevas automate tree-embryo selection with robot vision
SweTree Technologies and Prevas have built a system that finds the most viable tree embryos during somatic embryogenesis and picks each one in under three seconds before placing it in culture. It is SweTree's next step from the lab-grown seedling work in #011, now aimed at faster growth and resilience to drought, disease and insects.
Axel's notes: Somatic embryogenesis is the only realistic way to clone elite conifers at scale, and Swedish forestry has spent years working toward it, with an ambition of something like ten million cloned spruce plants a year. What has held it back is not the biology. It is the cost. A big part of that cost is hands: producing SE plants still leans on skilled people doing slow, delicate work, including picking viable embryos one by one under a microscope. So the interesting part here is not the speed of the robot, it is the price. Take enough manual hours out and the whole method gets cheaper. Cost is what decides whether SE stays a premium niche or becomes ordinary planting stock, and reliable automation is exactly the condition the people scaling this have said they need. If the price keeps falling, cloned and more resilient trees stop being a special case and start being a default choice. I am not ready to call it yet. It will be exciting to see whether the gains in growth and resilience hold up in real trees standing in the ground, not just in a culture vessel.
New Dawn Bio bets on lab-grown wood to replace felling
The Wageningen startup wants to grow real wood in bioreactors instead of cutting down trees, multiplying tree stem cells and steering them to harden into the finished shape, with no sawing or logging. It has raised a €2.1M pre-seed led by CapitalT, with Norrsken Evolve joining. The pitch is premium timber grown to spec, with far less waste than cutting rectangular boards from round logs. Samples are postage-stamp sized today; tabletop scale is the stated next milestone, and structural use is still years out.
Logscom traces timber from stump to mill with in-machine marking
Logscom's system is built into the harvester or forwarder and marks each log automatically as the machine handles it, replacing a manual labelling step and tying every mark to the machine's own measurement data. The Åsele company has sold around 200 systems, now running in SCA's and Sveaskog's machines, with another 30 to 40 in Norway. Its Metricboost tool then follows individual logs downstream to verify the harvester measured them correctly, a first step toward full stem-level traceability between forest and sawmill.
Södra scales Viking Analytics AI across three pulp mills
Condition monitoring is not new at Södra; the mills have logged sensor data for years. What changes now is the analysis layer. After a multi-year pilot at Värö, Södra is rolling Viking Analytics' behaviour-based AI platform MultiViz out across its mills in Värö, Mörrum and Mönsterås, where it reads vibration data from 3,650 sensors to flag failing equipment before it breaks. In testing it cut weekly alarms from around 500 to 20 and caught faults operators had missed.
Treebula recruits its CXO and CTO from outside forestry
Treebula (formerly Virkesbörsen) runs a free digital marketplace where more than 50,000 Swedish forest owners sell timber by collecting competing bids from buyers, alongside a digital forest management plan. It has now hired Elin Ankerblad, co-founder of design agency EY Doberman, as Chief Experience Officer, and Robert Lerner, most recently CTO at Mevisio and earlier at Avanza, as Chief Technology Officer. Both come from design and fintech rather than the forest sector, a sign that forestry tech is starting to pull senior talent that used to head elsewhere.