Latest / The Edge Computing Podcast with Fexingo: Local Compute, CDNs, and Distributed Infrastructure / How Edge Computing Is Changing EV Charging Infrastructure
Transcript
- Lucas: You pull into a fast-charging station for your electric vehicle. You plug in, tap your card on the terminal — and nothing happens. The screen says 'connecting to network' and then just hangs. That's not a hardware problem. That's a cloud problem. Luna: I've had that exact thing happen. Sat there for two minutes while the charger tried to phone home. And when it finally timed out, I had to unplug and find a different station. Lucas: Exactly. And the frustrating part is, the charger itself works fine. It can deliver 350 kilowatts of DC power. The battery management system in the car is ready. The cable is connected. The bottleneck is just a round trip to a cloud server that might be hundreds of miles away. Luna: So the solution is to bring the compute closer. Local edge servers at the charging site. Lucas: Right. And some networks are already doing it. A large European charging operator I've been following — they run about 8,000 fast-charging points across Germany, France, and the UK — they deployed edge nodes at about 200 of their busiest stations. The results are striking. Luna: What kind of improvement? Lucas: Before edge, their transaction failure rate was around 8 percent at peak hours. After edge, it dropped to under half a percent. The edge node handles the payment authorization locally — encrypts the card data, validates the session, then forwards the record to the cloud when the connection is available. If the cloud is down, the charger still works. Luna: So the edge node acts as a local cache for the cloud service. That makes sense. But what about the actual charging session? Does the edge node control the power flow too? Lucas: That's the next layer. The edge node isn't just a payment proxy. It also runs the charging session logic — the handshake with the car, the start and stop commands, the safety checks. And it can do load balancing across multiple chargers at the same site. If you have six 350-kilowatt chargers all drawing on the same grid connection, the edge node can dynamically cap each one so the total doesn't blow the transformer. Luna: That load balancing is actually critical for the grid operator too, right? If all six chargers try to pull full power at once, you could trip a local substation. Lucas: Exactly. And some grid operators are starting to require that fast-charging stations have that local intelligence. They want the station to respond to curtailment signals within milliseconds — not seconds. A cloud-based load management system introduces too much latency. By the time the message arrives, the surge has already happened. Luna: So the edge node is doing real-time power management. That's a very different use case from just caching web content. Lucas: It's a different class of edge altogether. This isn't a content delivery network node. This is a hardened industrial computer that speaks the charging protocol — the Combined Charging System, or CCS — and the grid protocol simultaneously. It has to be reliable at -20 degrees Celsius and in direct sunlight. Luna: And it has to handle software updates without taking the charger offline for hours. Lucas: That's another edge advantage. The operators I talked to use a containerized deployment model. They push updates to the edge node, run them in a sandboxed environment, and if the update crashes, the node rolls back to the previous version automatically. The charger never goes offline. Luna: That's smart. Because a charger that's offline during a software update is a charger that's not generating revenue. Lucas: And it's not serving drivers who need to get somewhere. Which brings up the broader point: the reliability of the charging network is becoming a public infrastructure concern. In the U.S., the Joint Office of Energy and Transportation has funded over $7 billion in charging infrastructure, and they're starting to include edge compute requirements in their grant criteria. Luna: So the federal government is essentially mandating local intelligence. Lucas: Not mandating yet, but strongly encouraging. The most recent request for proposals I saw from the California Energy Commission specifically asked bidders to describe how their stations would maintain functionality during a cloud outage. If you answer 'we have a backup cellular link to the cloud,' they'll ask 'what happens when the cloud itself is down?' Luna: And that's where a local edge node becomes the answer. It's the only way to guarantee operation without a cloud dependency. Lucas: Right. And the economics of edge have shifted enough that it's not a huge cost adder. A capable edge node for a charging station — something with a quad-core ARM processor, 8 gigs of RAM, a TPM chip for security, and a solid-state drive — that hardware costs around $300 to $400 in volume. On a station that costs $100,000 to install, that's a rounding error. Luna: Especially when you consider the revenue loss from transaction failures. An 8 percent failure rate on a busy station could mean thousands of dollars in lost charging sessions per month. Lucas: Exactly. So the ROI is clear. And there's another emerging use case that makes edge even more compelling: vehicle to grid, or V2G, where the car can discharge power back to the grid. Luna: V2G requires very tight latency, doesn't it? The grid needs to signal when it needs power, and the car needs to respond almost instantly. Lucas: Yes. The standard for V2G communication, ISO 15118, specifies response times under 100 milliseconds. If you route that signal through the cloud, you're adding 30 to 50 milliseconds just for network round-trip time, plus processing time on the server. With an edge node on-site, the signal stays local. The grid operator sends a request to the edge node, the edge node talks to the car, and the car starts discharging in under 50 milliseconds. Luna: So edge compute is essentially an enabler for V2G to work at scale. Without it, the latency is too high for the grid to rely on. Lucas: That's the argument some grid operators are making. They want thousands of EVs to act as a virtual power plant, but they need the control loop to be deterministic. Cloud introduces too much jitter. Edge gives them a predictable, low-latency control point. Luna: And that control point also becomes a security boundary. If each station has a local edge node, a compromise at one station doesn't spread to the cloud. Lucas: Right. The edge node can have a hardware root of trust — a dedicated security chip that stores the private keys for authentication. Even if someone physically accesses the node, they can't extract the keys. That's a much stronger security posture than a cloud-only model. Luna: I want to go back to something you said earlier about the European operator. How many stations did they deploy edge nodes on? Lucas: About 200 of their busiest stations, out of 8,000 total. They focused on sites where failure rates were highest — typically highway rest stops with high throughput. And they told me the hardware has been rock solid. Zero node failures in six months of operation. Luna: That's impressive. And it's a great proof point that edge compute for EV charging isn't theoretical. It's deployed, it's working, and it's improving reliability measurably. Lucas: Absolutely. And the interesting thing is that this is just the beginning. As charging speeds increase — we're already seeing 500-kilowatt chargers being tested — the demands on local intelligence will only grow. You can't have a 500-kilowatt charging session fail because of a cloud timeout. That's just not acceptable. Luna: And with more cars on the road that can accept those speeds, the stakes get higher. A single station might be serving hundreds of vehicles per day. Lucas: Exactly. So the edge compute model for charging infrastructure is going to become table stakes. Every new station built today should probably have local compute, just like every new station has a transformer and a cable. Luna: It becomes part of the standard bill of materials. Lucas: Right. And that's where the industry is heading. I think within three years, most new fast-charging stations will ship with an edge compute module as standard. It'll be as common as the payment terminal. Luna: If today's conversation gave you something usable — a new angle on edge compute, a concrete example to think about — I want to mention something quickly. This show is entirely listener-supported. No ads, no sponsors, no corporate backing. A small group of listeners already chips in monthly through buy me a coffee dot com slash fexingo, and that's what keeps these episodes coming. Lucas: Yeah, and we really mean it. This isn't a side project for us — we produce 300 shows across the network, and the model only works because of that direct support. If you find value in what we do, that's the only ask. Luna: And we're back to the topic. Lucas, you mentioned V2G — are there any other edge applications for charging stations that we haven't covered? Lucas: One I'm watching is predictive maintenance. The edge node collects data from the charger's internal sensors — temperature, voltage ripple, contactor wear — and runs a local model to predict failures before they happen. That way, the operator can dispatch a technician to replace a component during off-peak hours, rather than waiting for the charger to fail. Luna: That's a great example of edge AI in action. The model runs locally, so it doesn't need a cloud connection to make predictions. Lucas: Exactly. And the data stays on-site, which addresses some privacy concerns. The operator doesn't need to send detailed operational data to the cloud to get value from the model. Luna: So edge compute is really becoming the backbone of smart charging infrastructure. From payment to power management to predictive maintenance. Lucas: It is. And I think we'll look back in five years and wonder why we ever built fast-charging stations without local intelligence. It's such an obvious application for edge computing. Luna: Agreed. And it's one of those rare cases where the hardware cost is tiny compared to the operational benefit. Lucas: That's the edge computing sweet spot — a problem where latency, reliability, or security make the cloud the wrong answer, and the hardware is cheap enough to deploy locally. EV charging hits all three. Luna: Great episode. Thanks for walking through it. Lucas: Thanks, Luna. And thanks to everyone listening. We'll be back next week with another edge computing angle.