Latest / The Edge Computing Podcast with Fexingo: Local Compute, CDNs, and Distributed Infrastructure / How Edge Computing Is Powering Smart Agriculture
Transcript
- Lucas: So there's a vineyard in California's Central Valley — about 120 acres of Cabernet Sauvignon — that's been running an edge computing experiment for the past two growing seasons. And the headline number is a 35 percent reduction in water usage with no drop in yield. Luna: Thirty-five percent is huge. How are they getting that kind of efficiency? Lucas: It's a combination of sensors and local compute. They've buried soil moisture probes at three different depths every 50 feet across the vineyard — about 500 nodes total. Each node connects to a LoRaWAN gateway, which feeds into an arm based edge gateway mounted on a pole near the center of the property. Luna: So the edge gateway is doing the processing locally rather than sending everything to the cloud? Lucas: Exactly. The gateway runs a lightweight neural network that takes in soil moisture, temperature, humidity, and wind speed — plus imagery from a drone that flies over twice a week. It predicts which vines need water and opens the drip irrigation valves for exactly the right duration. Luna: And that decision has to happen fast. If you send the drone image to the cloud and wait for a response, the conditions might already have changed. Lucas: Right. The farmer told me that even with a fast cloud connection, round-trip latency is around 200 milliseconds. Their edge gateway does inference in under 10 milliseconds. That twentyfold difference matters when you're adjusting valves for hundreds of individual drip lines. Luna: I imagine the savings add up quickly. Water in California isn't cheap. Lucas: It's not. And the vineyard is in a region where groundwater allocations have been cut by 20 percent since the Sustainable Groundwater Management Act kicked in. So this isn't just a cost play — it's a compliance play. They needed to do more with less water. Luna: What kind of hardware are we talking about? Is this a ruggedized server or something more like a Raspberry Pi? Lucas: Closer to a Raspberry Pi, but industrial-grade. The gateways are from a company called ADLINK, using an NXP i.MX 8M processor. They're fanless, rated for outdoor temperatures up to 60 degrees Celsius, and draw about 15 watts at peak. The whole setup — gateways, LoRaWAN concentrators, and the sensors — cost around 80,000 dollars for the full 120 acres. Luna: So about 670 dollars per acre. That's not trivial, but if you're saving 35 percent on water and maintaining yield, the payback period must be reasonable. Lucas: The owner calculated a 2.3-year payback based on water savings alone. And that doesn't include the secondary benefits — less fertilizer runoff because they're not overwatering, and fewer fungal diseases because the vines aren't staying wet. Luna: That's a pretty compelling business case. But I wonder about scalability. A 120-acre vineyard is one thing. What about a 10,000-acre corn operation in Iowa? Lucas: Good question. The economics shift when you scale. For large row-crop farms, the sensor density is lower — you might have one soil probe per 10 acres instead of per vine. So the per-acre hardware cost drops. But you also need more edge gateways because the latency requirements for irrigation are less stringent for corn than for premium wine grapes. Luna: So the edge deployment is really tailored to the crop and the water stress profile. Lucas: Exactly. And this is where I think edge computing has a real advantage over one-size-fits-all cloud solutions. You can tune the model for a specific field, even a specific block of vines, and update it over the air without touching the hardware. Luna: Speaking of over-the-air updates — how do they handle connectivity in the field? Cellular coverage in the Central Valley can be spotty. Lucas: They use a combination of cellular and satellite backhaul. The edge gateway stores data locally for up to 72 hours and only syncs when a connection is available. The inference runs entirely on the gateway, so even if the backhaul goes down, the irrigation schedule continues based on the latest sensor readings. Luna: That's the beauty of edge — it doesn't break when the internet does. Lucas: Exactly. And it's not just irrigation. Some farms are starting to use edge for real-time pest detection. You mount a camera on a tractor, run object detection on the edge, and spray only where insects are present. That can cut pesticide use by 90 percent. Luna: I've seen that with weed detection too. There's a startup called Blue River that does 'see and spray' — it uses computer vision on the implement itself. Lucas: Blue River was actually acquired by John Deere a few years back. They're now integrating edge compute into the tractor itself. Deere's latest sprayers have an NVIDIA Jetson module on board that processes camera feeds at 30 frames per second. Luna: So we're moving from a model where the farm sends data to the cloud and waits for instructions, to one where the equipment makes decisions in real time. Lucas: That's the direction. And it's being driven by cost reductions in edge hardware. The Jetson module in that sprayer costs about 500 dollars. A few years ago, a comparable GPU would have been several thousand. Luna: And the cloud costs add up too. If you're sending high-resolution video from every tractor, you're paying for bandwidth and storage. Lucas: Right. And the latency issue again. For pest detection, you need to trigger the spray nozzle within 100 milliseconds of the camera seeing the weed. Cloud round-trip is too slow at any reasonable cost. Luna: I want to pivot slightly — there's a funding angle here. I've been tracking agtech venture capital, and edge-focused startups raised about 1.2 billion dollars globally last year. Lucas: That's up from around 800 million in 2024. A lot of that is going to companies building edge infrastructure for precision agriculture. Luna: One that caught my eye is a company called Arable. They make solar-powered edge nodes that combine weather, soil, and plant measurements into a single device. They raised a 40-million-dollar Series B in March. Lucas: Arable is interesting because they're targeting the middle market — farms that are too big for manual monitoring but too small to hire a full-time agronomist. Their edge node does all the data fusion locally and only pushes summaries to the cloud. Luna: That's exactly the kind of product that makes edge accessible. You don't need a data science team to interpret the numbers. Lucas: And that's the broader theme, I think. Edge computing in agriculture isn't about deploying the most advanced AI — it's about making reliable, low-latency decisions with minimal infrastructure. The vineyard we started with proves that you can do it on a modest budget and see real ROI. Luna: I'm curious about the limitations, though. What happens when a sensor fails or a gateway goes down? Is there enough redundancy? Lucas: That's the weak spot. Most edge deployments today are single point of failure designs. If that one gateway in the vineyard dies, the irrigation schedule falls back to a static plan until someone replaces it. Some operators are starting to deploy dual gateways with automatic failover, but that doubles the hardware cost. Luna: So there's a reliability vs. cost trade-off that still needs to be solved. Lucas: Exactly. And that's probably the next frontier for edge in agtech — building affordable, fault-tolerant systems that can run unattended for months at a time. Luna: It's also a reminder that this stuff doesn't build itself. The hardware, the software, the installation — it all takes investment. And honestly, I think it's worth pausing to appreciate that shows like this can dig into topics like gateway failover and LoRaWAN sensor networks without having to answer to advertisers. Lucas: Yeah, that's a good point. The reason we can spend an episode on the minutiae of vineyard irrigation is that a small group of listeners chips in monthly through buy me a coffee dot com slash fexingo. It's not a huge number, but it's enough to keep the show ad-free and focused on substance. Luna: Exactly. No pressure, but if you get value from these deep dives, that's one way to keep them coming. And it really does make a difference. Lucas: Alright, back to the tech. I mentioned the Jetson module in John Deere's sprayer. One thing I find fascinating is how the edge model gets trained. The initial model is trained in the cloud on thousands of labeled images, but then it's fine-tuned on the edge device using data from that specific farm. Luna: So the edge model gets smarter over time without needing to send raw data anywhere. Lucas: Exactly. That's called federated learning, and it's becoming more common in agriculture. Each tractor on a farm runs its own local training loop, and only the model updates — not the data — get sent back to the cloud. That preserves privacy and reduces bandwidth. Luna: And it means the model adapts to local conditions. A weed in Iowa looks different from a weed in Texas, even if it's the same species. Lucas: Right. Soil type, light angle, humidity — they all change how the camera sees the plant. Local fine-tuning can improve detection accuracy by 10 to 15 percent in field tests. Luna: That's a meaningful improvement when you're trying to minimize herbicide use. Lucas: Exactly. And it all comes back to edge compute. Without local inference and local training, you're either spraying too much or missing weeds. The edge makes the whole system smarter and more efficient. Luna: So where do you see this going in the next five years? Are we going to see edge nodes on every farm? Lucas: I think we'll see a tiered approach. Large corporate farms will have edge gateways on every piece of equipment, with full redundancy. Small family farms will probably adopt a shared model — maybe a cooperative buys a few edge nodes that get moved between fields. The technology is there; it's the business model that needs to adapt. Luna: I like that framing. It's not just a tech story — it's an economic story about who gets access to the tools. Lucas: And that's the conversation we should keep having. The vineyard example shows what's possible when you combine edge compute with domain expertise. The challenge is making it affordable and reliable for everyone. Luna: Well, I think we've given our listeners plenty to chew on. Thanks for digging into this one. Lucas: Yeah, it's a great example of how edge computing is quietly transforming industries most people don't associate with high tech. Luna: And we'll be back next week with another angle. Lucas: See you then.