A Spanish startup is quietly doing what most Western companies only talk about
Everyone assumes the AI chip race is already over — that NVIDIA won, that the US locked it up, and that everyone else is just buying what they’re told to buy. That assumption is wrong, and Barcelona-based Openchip is one of the clearest reasons why.
I build bots for a living. I spend a lot of time thinking about what’s running underneath them — the silicon, the memory bandwidth, the power draw per inference call. Most bot builders don’t think about this stuff until their cloud bill arrives. But the hardware layer is where the real constraints live, and right now those constraints are almost entirely controlled by a handful of American and Taiwanese companies. That’s a fragile position for anyone building serious AI infrastructure in Europe.
Openchip wants to change that equation. The company is a full-stack system-on-chip and software builder focused on AI and high-performance computing. Their approach centers on in-house designed, high-performance RISC-V architecture — an open instruction set that gives them real independence from proprietary chip ecosystems. They’re targeting a 2028 product launch, and if they hit that window, they’ll be entering a market that looks very different from today’s.
Why 2028 is actually a smart target, not a delay
When I first saw the 2028 date, my instinct was skepticism. That’s four years away. In AI time, four years feels like a geological era. But think about what the agentic AI space will look like by then.
Right now, most AI deployments are still prompt-response systems. You send a query, you get an answer. Agentic AI is a different animal entirely — systems that plan, execute multi-step tasks, call tools, manage memory, and run autonomously for extended periods. The compute profile for that kind of workload is not the same as running a chatbot. Agentic systems need efficient, sustained inference at scale, not just raw peak throughput. That’s a different hardware problem, and it’s one that current GPU-centric architectures aren’t perfectly suited for.
Openchip is positioning itself for that specific moment. A chip designed from scratch with agentic workloads in mind — lower power draw, tighter software integration, purpose-built for the inference patterns that autonomous agents actually produce — could be genuinely competitive against hardware that was designed for a different era of AI.
What MWC 2026 told us about their trajectory
In early 2026, Openchip showed up at MWC Barcelona, 4YFN, and Talent Arena. For a startup, that’s a deliberate signal. MWC is where you go when you want to be taken seriously by telcos, infrastructure buyers, and enterprise decision-makers. 4YFN is where you go when you want to attract investment and talent. Showing up at both in the same week means they’re building on multiple fronts simultaneously — product, funding, and team.
That kind of presence matters. European deep tech has historically struggled with the gap between research quality and commercial execution. Openchip seems aware of that gap and is actively working to close it.
What this means for bot builders and AI developers
Here’s why I think people in my corner of the industry should pay attention to this. The cost and availability of inference compute is the single biggest constraint on what we can build. Right now, we’re almost entirely dependent on a supply chain that runs through a small number of chokepoints. Any disruption — geopolitical, logistical, or competitive — hits everyone building on top of that infrastructure.
A solid European alternative in the AI chip space doesn’t just benefit European companies. It creates pricing pressure, supply diversity, and new architectural options for everyone. If Openchip ships a chip in 2028 that’s genuinely optimized for agentic inference workloads at lower power costs, that changes the economics of running autonomous agents at scale. That matters to me when I’m designing bot architectures. It should matter to you too.
The contrarian read nobody wants to say out loud
The mainstream narrative says Europe is too slow, too fragmented, and too risk-averse to compete in AI hardware. Openchip is a direct challenge to that story. They’re not trying to out-NVIDIA NVIDIA on raw GPU performance. They’re picking a specific future workload — agentic AI — and building toward it with a clean-sheet design and an open architecture.
That’s not a long shot. That’s a focused bet on where the compute needs are actually heading. And from where I sit, building bots that are getting more autonomous by the month, it looks like a pretty well-aimed one.
Watch Barcelona. 2028 is closer than it sounds.
🕒 Published: