On a grey November morning in Hoogstraten, a tomato grower sits down before sunrise with coffee and the climate computer. Screens up or half down, how hard to dose CO2, when to crack the vents against the wind forecast: dozens of setpoints, adjusted from experience, that decide whether the week's crop gains or loses. Two announcements in October made it concrete that software will soon make some of those calls in European glasshouses — not in a research trial, but under commercial contracts, starting this winter.
Two October deals, one direction
The first is an acquisition. On 9 October, Source.ag — the Amsterdam company whose software monitors crops and forecasts harvests in more than 500 greenhouses across 25 countries — announced it had acquired Optimal, a London startup whose AI directly steers heating, ventilation, screens, lighting, CO2 dosing and irrigation through the grower's existing installations. According to the announcement, reported by trade outlets such as Hortidaily, the first autonomous deployments at Source customers in Europe and North America go live this winter. Agro Care, one of Europe's largest greenhouse tomato producers, says it intends to be among the first to switch the system on.
The second is a funding round. Also in October, Berlin-based hexafarms raised €4.8 million, led by Ananda Impact Ventures and Green Generation Fund, to scale its AI monitoring platform: wireless sensors and cameras that track temperature, humidity, CO2, light and substrate conditions, forecast harvest volumes and flag pests and disease early, with models covering about 50 cultivars. More than 50 growers in 13 countries already use it, including the growers' cooperatives Royal ZON and Elo, and the company names the Benelux as a priority for expansion.
Neither is a lab experiment. Both are commercial bets on the same idea: the glasshouse, the most measured and most controllable environment in agriculture, is where AI takes over day-to-day decisions first.
What does an autonomous greenhouse actually do?
It closes the loop between measuring and acting. Most growers already have sensors and a climate computer, but a person still translates data into setpoints. An autonomous system does that translation itself: it combines a physics model of the greenhouse with machine-learning models of the crop, plans the climate strategy days ahead, and then adjusts heating, screens, vents, light and CO2 in real time as the weather changes. The grower sets the goals and the limits; the software handles the hour-by-hour steering.
The systems reaching the market do not decide what to grow, when to plant or what price to accept. They work inside boundaries a human defines — closer to an autopilot with a pilot in the seat than to a pilotless aircraft.
Do the published numbers hold up?
Treat them with attention and caution at the same time. In a commercial-scale trial in a Dutch greenhouse, Optimal's system reportedly produced 13% more yield with 27% less energy than a benchmark Dutch grower. Those figures come from the company, as does hexafarms' statement that customers managed more than €300 million of produce through its platform this year. None of it has been independently audited.
The direction is plausible, though. Energy is one of the largest cost lines of a heated glasshouse in Belgium or the Netherlands, and climate steering is exactly the kind of continuous, multi-variable optimisation where software beats periodic human adjustment. A system that trims even a tenth off gas consumption pays for a lot of subscription fees. The honest reading: the technology is past proof-of-concept, and the numbers on your crop, your greenhouse and your energy contract are still yours to verify — ideally in a measured pilot on one compartment before any site-wide commitment.
The quiet driver: there are not enough growers
Behind both deals sits a labour problem. A good head grower takes years to train, and across Europe fewer young people enter the profession than leave it. Optimal's founder has argued that autonomous control exists precisely to deal with this scarcity, and the logic extends naturally: one experienced grower supervising several sites through software, instead of one grower per site.
For cooperatives the logic is especially interesting. When a cooperative like Royal ZON puts a monitoring platform in front of its members, the smallest member gets forecasting tools that previously only the largest operations could staff for. The pattern travels beyond horticulture too: pooled data, shared tooling, and one negotiation on data ownership and pricing instead of every member negotiating alone.
When this isn't the right move
If your glasshouse is cold or lightly steered, the savings shrink. The economics rest on energy and yield in intensive cultivation of tomatoes, peppers, cucumbers and similar crops. An unheated tunnel or a low-tech ornamental operation gives an algorithm far less to optimise.
If your cultivar is niche, the models may not know it. Fifty cultivars is a lot, and also not that many. Ask for references on your specific crop before signing.
If the contract is vague about data and exit, wait. Some platforms own the hardware and sell access by subscription, which lowers the entry cost. But ask what happens to your climate and crop data if you leave, whether the system works with your existing climate computer, and who carries liability when a wrong setpoint costs a crop. These are contract questions rather than technology questions, and they are where a smaller business wins or loses the deal.
Frequently asked questions
Will this replace my head grower? No. The systems going live this winter work within targets and limits a grower defines, and can be overridden at any time. What changes is the ratio: one experienced grower can supervise more hectares. Vendors say scarcity of skilled growers, not their replacement, is driving demand.
Do I need to rip out my climate computer? Usually not. Optimal's control works through existing installations, and hexafarms installs and maintains its own wireless sensors under subscription. The practical questions are integration with your specific setup and who pays when hardware fails; put both in the contract.
What does the EU AI Act mean here? Greenhouse climate control is not a high-risk application under the Act, so the heavier obligations fall mostly on the vendors. Businesses using AI systems do need to ensure basic AI literacy among the staff who work with them — think a training afternoon, not a compliance department.
Can a small grower afford this? Subscription pricing has lowered the entry point, and cooperative agreements lower it further. The real question is payback against your energy bill and crop value, which is why a one-compartment pilot with a measured baseline beats any brochure.
The greenhouses switching to autonomous control this winter will not make headlines next spring; they will simply burn less gas and lose fewer crops to late decisions. That is how useful AI tends to arrive: a narrow system doing one costly job well, under human constraints. Finding that job in your own operation — climate steering, harvest forecasting, or the administration around them — is the real work, and it is the work Cresly does from Brussels with farms, cooperatives and food producers. If you are weighing where AI would actually pay off in your greenhouse, packing line or cooperative, write to [email protected].