Your browser doesn’t support HTML5 audio
We’re on a mission to put high-performance, low-cost, general-purpose computing platforms in the hands of enthusiasts and engineers all over the world.
That’s the mission statement for Raspberry Pi, best known for is single-board computers originally created back in 2008 to help teach IT in schools. Since then its use case footprint has expanded into robotics, semiconductors, industrial automation, the Internet of Things and other areas, but that basic mission remains the same.
But it doesn’t apply to everyone, as current events have reminded us. This year has seen a higher profile for the topic of ethical use of AI in warfare and combat situations. The battle between Anthropic and the US Department of War in regard to what the latter may and may not use the former’s tech for has highlighted one end of the arguments. At the other end are the likes of Palantir, whose CEO happily reminds shareholders that his firm’s tech will kill people and he hopes they don’t have a problem with that!
What’s all that got to do with Raspberry Pi, given its origins in education and accessibility? The answer lies in reports that the firm’s boards are finding their way into weapons systems via the so-called gray market. This isn’t just paranoia as CEO Eben Upton confirms:
We are aware of the use of our products in particularly drone platforms, particularly Russian drone platforms in the context of the invasion of Ukraine. This news has prompted us to take another look, a very detailed look at the gray market...We’ve put a lot of work in, over the last six months in particular, particularly for unit selling into China, to strengthen the information that we require that our resellers provide us in terms of the end use of our products.
Getting that extra level of detail helps to clamp down on unethical uses of the technology, he adds:
We do believe it's possible. I don't think this is not a council of despair. We do believe that it is possible to meaningfully limit, hopefully to very, very small numbers, potentially to zero, to limit the flow of our products into these sort of applications. Nobody is happy when they wake up in the morning and see to these stories, and we are doing our very best to get in the way of this diversion of our products
Chips with everything
On a cheerier note, Upton reckons that the past 12 months have been “the crossover year” for the firm’s expanding semiconductor business, with a a 47% year-on-year increase in chip sales. That brings the total sold to 8.4 million units, meaning, says UptonL
We sold more chips than boards for the first time...we really are just getting started. We believe that semiconductors let us make better boards and modules. And we believe that over time, they will become a valuable second franchise in their own right.
Raspberry Pi is at heart a product company, and we love to launch products more than pretty much anything else. 2025 was a somewhat slower year for us than 2024 in terms of product launches, but still one of our strongest historically. We launched many new hardware products, including new microcontroller variants.
But the current ‘star of the show’, he declares, is Raspberry Pi Connect, the firm’s first software product, a secure remote access solution for Raspberry Pi OS, allowing users to connect to their Raspberry Pi desktops and command line directly from any browser:
The Raspberry Pi Connect platform continues to go from strength to strength. We have 370,000 registered devices on the Connect platform [as of] the end of 2025 and over half-a-million registered devices today. In 2025, we launched many new features aimed at OEM customer pain points, including over-the-air firmware updates and deeper integration with our imagery utility to simplify the onboarding of devices and of users.
There are plans to bring Raspberry Pi Connect to the microcontroller family. I think we are refining at the moment if you think that Raspberry Pi Connect has its roots as a remote access solution. I think we are refining what the proposition there might look like for remote access. I think the proposition for over-the-air updates, which we added to Raspberry Pi Connect at the end of last year, microcontrollers is much more straightforward.
The AI story
And of course, it’s 2026, so Raspberry Pi has its own AI story to tell, with Upton stating that the concept of running AI workloads on Raspberry Pi is very important to the future of the platform:
We have to remember that we come to this opportunity from a position of enormous strength. We're already the most popular platform for edge computing and many of our customers' workloads already have an AI flavor to them generally because there is a lag in terms of how new technology makes its way into the OEM space. When we say an AI flavor, what we tend to mean is visual AI, vision AI, generally the recognition and classification of objects in still and video images coming into the Raspberry Pi.
Some of those workloads will run on the CPU, some of them may be GPU-accelerated. Some of them may be accelerated by a dedicated accelerator platform for one of our partners, notably Hailo and Sony. We have fantastic relationships with these accelerator vendors and with the model vendors, And we leverage these relationships over time to expand the platform performance envelope to expand the range of AI workloads, which will run performantly on Raspberry Pi products.
Upton identifies a number a trends underpinning the general edge AI opportunity that he envisions for the company:
Frontier models in the cloud are hugely capable, but when we build IoT applications on those cloud models, we build in latency, we build in ongoing cost, and we build in a dependence on the reliability and availability of the network, which translates into inherent brittleness in the resulting product. In contrast, if we are able to run AI workloads at the edge of the network, that has the potential to deliver IoT applications, which are more robust, more cost effective and which deliver lower latency and address pervasive concerns around privacy and data sovereignty.
Over time, AI capabilities trickle down the performance ladder, he adds:
Models become smaller, they become sparser, they are more heavily quantized, all without sacrificing performance. We've seen this already in vision AI...Our largest OEM customers today are AI customers. They are running vision AI applications on Raspberry Pi, and the workloads that they're running today on Raspberry Pi would have required a large NVIDIA GPU a decade ago. So we've seen this happen already.
We've seen the confluence of the increased performance of edge devices like Raspberry Pi and the reduction in compute intensity of models at constant quality output. We've seen those converge to enable vision AI at scale on Raspberry Pi platforms. And we're seeing this same trend replay itself today in Large Language Models (LLMs) and other generative workloads. Taken together, this confluence of expanding performance and shrinking performance requirements will gradually bring more workloads into play for Edge AI.
That all translates into opportunity for Raspberry Pi, he says:
Quite simply, the opportunity is to leverage our existing position in edge computing to become the platform of choice for AI at the edge. We have the opportunity to serve the existing demand for edge AI compute across markets from industrial automation to defense. We have the opportunity to become the default embedded host for Agentic AI. We're already seeing this in the popularity of agentic platforms like OpenClaw running on Raspberry Pi. Together, these give us the opportunity to define the future direction of AI as inference migrates to the edge over the next decade.
The rise of agentic AI is a sign of things to come, he predicts:
Often what we find is that our enthusiast customer base is ahead of the game. The things our enthusiasts are doing now, our industrial customers, our OEM customers we'll be doing in three, four, five years time. Substantial use of Raspberry Pi in this space is an indicator that agentic AI, particularly in this kind of future agentic AI in which the agent has migrated into the local client device is a powerful abstraction to put around the underlying AI technology.
My take
I’ve always had a fondness for Raspberry Pi, a nice UK success story in the tech field. I confess I hadn’t realized just how ‘industrial’ its use cases had become over the years. One to keep a closer eye on in future.
And good luck with stamping out the unethical abuses, people! You’re on the right side of history once again.