A significant market signal from Nvidia

Computing is moving out of the cloud onto drones, hospital equipment, and your car's dashboard. The shift is worth billions.

Only “crazy” people question the returns of AI, which might be “insane,” Nvidia’s Jensen Huang insisted. Despite growing concerns about its hefty price tag, AI promises too much for businesses to ignore. Bain&Company’s report perfectly sums up the trend.

“Every year, boards approve bigger automation budgets. Every year, CEOs sign off on the next wave — robotic process automation (RPA), then machine learning, then generative AI, now agents. And every year, the savings fall short. Not catastrophically, not enough to kill the programs, but consistently, quietly, and by a margin that should be making executives uncomfortable,” the report reads.

Nvidia, naturally, has a huge impact on the market. Aside from overblown quotes, even Nvidia’s reporting, put under scrutiny, can influence the market and give us an idea of where it’s headed.

In its Q1 FY2027 earnings release, Nvidia made a clear distinction between data centers and edge computing. As edge computing now has its own forecasting line, let’s briefly look at what it is, its use cases, and its main benefits.

What is edge computing?

In simple terms, edge computing is computing done as close to the data source as possible. For example, data can be processed on a device such as a mobile phone, drone, car, or on-premises server instead of being sent to the cloud for processing.

Nvidia’s edge category included PCs, game consoles, workstations, AI-RAN base stations, robotics, and automotive.

“That breadth reflects the opportunity for accelerated computing systems and new software layers to distribute outward from cloud and centralized data centers to real-time edge processing and data gathering,” Mike Shapiro, founder and co-president of iodyne, told Cybernews.

Edge computing is most beneficial in robotics, defense, automotive, manufacturing, and healthcare, among other industries. It enables real-time decision-making by eliminating the delay caused by data traveling back and forth to the cloud.

Industries like defense prefer edge computing because it enhances data security.

How big is edge computing?

Nvidia reported total revenue of $81,6 billion, with the majority coming from the Data Center segment. The edge category accounted for $6,4 billion, representing 29% growth compared with the previous year.

As per Grand View Research, the global edge computing market size was estimated at $23.65 billion in 2024 and is expected to reach $ 327.79 billion in 2033.

“The increasing demand for faster data processing, reduced latency, and real-time insights is significantly accelerating the adoption of edge computing across industries,” the report reads.

What signal does Nvidia send?

According to Grand View Research, the following companies are leading players in the edge computing market, dictating industry trends:

• Amazon Web Services, Inc.
• Microsoft Corporation
• Google LLC
• Cisco Systems, Inc.
• Hewlett-Packard Enterprise Development
• Intel Corporation
• Huawei Technologies Co., Ltd.
• Schneider Electric
• Siemens
• General Electric Company

As you can see, Nvidia didn’t make the top 10 list. And still, since it’s the most valuable company in the world, its every breath is put under scrutiny.

“When a company the size of Nvidia formally separates the edge from the data center in how it communicates its own business to investors, it's signaling something about where it believes a next meaningful battleground actually exists to drive growth. The AI data center story has been dominant for the last several years, and rightfully so, but this reorganization suggests accelerated edge computing is graduating to Edge 2.0,” Shapiro told Cybernews.

To better understand what edge computing is, I sat down with Andreas Hollander, CEO of Scaleout, who is working with NATO and BAE Systems on securing AI systems deployed across edge environments, from drones and satellites to connected vehicles.

The interview below is edited for length and clarity.

Why is edge computing or edge AI so important? What is it in comparison to cloud computing?

That's a very good question because it's also something that is quite ill-defined sometimes. Edge can mean many different things, but essentially, edge computing means that you have your compute hardware — CPUs, GPUs, and storage — deployed very close to the sensor data. Edge computing could be your mobile phone. It has a processor. It is deployed near the end user. It can also be a vehicle computer sitting in the vehicle, working with an ADAS (advanced driver-assistance system) system.

And the reason you need to do this is that some algorithms need to operate with low latency. In these cases, you can't tolerate the delay of sending data back to the cloud, and again back to the front line. But it can also be because you have security-sensitive data that you cannot centralize or move to a remote storage location. So everything needs to stay on-premise.

Hospitals are a good example. At least in the Swedish jurisdiction, sensitive patient data often can't move outside of the physical hospital perimeter. So that also mandates edge computing to some extent.

What are other examples of industries where edge computing has become useful?

A very good example is vehicles. Autonomous or adaptive safety systems relying on, for example, cameras for obstacle tracking or obstacle detection — those machine learning algorithms need to run on the onboard computer in the vehicle, because the latency of moving video data to the cloud is just too great. So that's a good example of edge computing driven largely by performance latency, and that's for safety-critical systems.

Another good example would be in a defense context, where we have started working a lot. There, you have sensors like drones — surveillance drones, for example — where you also have sensitive data that can't be centralized. And you also have severe bandwidth constraints or networking constraints in moving the data long distances. So to create an autonomous agent — a drone that can operate and fly without a controller — you need an algorithm on the actual drone that can track objects and steer even without any connectivity to the ground. So that's another example. And these are increasingly deployed, of course, in Ukraine and other places.

Are edge devices more expensive, more advanced than your typical device?

If you want high-performance, AI-capable edge devices, they tend to be more expensive than commodity devices. For example, gaming workstations are definitely more expensive than server-side, cluster-based hardware, because you need to squeeze more performance into a smaller form factor. And especially if you're thinking about industrial settings, where you also need industrial-grade hardware that can operate under wider temperature ranges and perhaps under IP67 (waterproof) conditions. So yes, they become more expensive.

But on the other hand, it's also an area where things are evolving very rapidly, and many hardware manufacturers are leaning into this. That, as I said, is essentially edge AI for embedded systems.

But you also have a lot of edge use cases where it could be sufficient to take a commercial server — a Linux server — and put it in a smart building, for example, or in a hospital. And it doesn't have to be much more expensive than what you could have in your home. The challenge now is, of course, that in general, memory and GPUs are very expensive. But that's sort of part of the game now.

But is it perhaps not necessary for every organization to pursue on-device computing? When do you actually need some edge services, and when is cloud computing enough? How do you decide between the two?

If you can use server-based or cloud-based compute, that's usually much easier. When you consider edge computing, you really can't move raw data to the cloud. There are a couple of reasons for when this occurs. One of them would be that the latency is too great. So you need to make real-time decisions at the edge, and you can't afford the round trip of sending data, getting an AI prediction, and then getting that back to the device.

If you are operating under network conditions that are less than ideal, it can take too long, especially if you need to make decisions on a millisecond time scale. It is often the case in both defense and industrial applications that performance is not good enough when relying on the cloud. You have to go to the edge.

You need to do the computing locally because you are dealing with private, confidential, or secure data that you can't centralize or send to a cloud provider. There can be many reasons why you don't want to do that. It can be jurisdictional. It could be that you're in a regulated industry. It could also be that this is like company secrets. You want to keep them under your own physical control.

We're talking much more about the cloud-edge strategy. This is tied to the sovereignty discussion, where we increasingly want control of our data and processing.

Edge computing doesn't have to happen on a device, does it? You can have a small server or something like that built on your premises, right?

When we talk about edge computing, we often think of the device edge. That could be all the way to the phone or the embedded machine in the car. But we can think of edge computing as a modern IT architecture that is a hierarchical system where you start with a hyperscaler on top, and you can have some services there, and then you have a layer called the near edge that could be regional data centers or private cloud data centers.

And then we often talk about the far edge. There we're back to what you said. It could even be a desktop computer or a gateway node sitting, for example, in a factory or in a building, such as a hospital.

At the bottom of this hierarchy, we think about the devices. That could be a mobile phone or a popular Nvidia platform called Jetson – a hardware platform designed to be used in vehicles, embedded systems, or drones. Raspberry Pi is another common example of those architectures.

Many companies are introducing AI applications. There's a lot of talking to the cloud and, uh, probably sending sensitive corporate data there. What are some common attack vectors?

We use chatbots or LLMs to automate things and build agents. The first and foremost problem or threat, I would say, is the human factor. We inject data that should not go into these algorithms. That's a corporate data governance problem. There's also a lot of shadow IT happening around these models. So essentially, using them safely is a challenge.

We do a lot of work on machine learning-specific attacks. We are building a platform that lets you deploy machine learning algorithms to the edge. We can also then retrain these algorithms continuously with new data at the edge. [Scaleout is keeping raw data on-device and sharing only model updates, limiting what an attacker can intercept, corrupt, or replay across a distributed network.]

If we have deployed models across multiple different cars or trucks in a vehicle fleet, a good example is our long-term collaboration with Scania Connected Vehicles. The project aims to enable this type of fleet learning, where fleet trucks and fleet vehicles can collaborate and learn together, but without sharing data or sending it to the cloud. That's the type of system we build.

We are, of course, very interested in what the risks are with these things. What can you do if you find a machine learning model? Can you learn something about the input data that came into that model, the raw images, the training data, or is that safely obfuscated by the model?

Data poisoning is very interesting here as well. What if someone tries to put bad data into the training data and build backdoors into the model?

Many are very interested in reverse engineering attacks. How can you reverse engineer data from algorithms? And the short answer to that is that today, not much is possible. But it's a rapidly evolving field.