Remember When Europe Was “Too Late” to the AI Hardware Race?
Remember when the conversation around AI chips was basically a two-horse race between Nvidia and whoever Google was quietly building in a data center somewhere? European hardware ambitions were mostly met with polite skepticism — good ideas, not enough scale, too far behind. That narrative is starting to look a little shaky. Barcelona-based Openchip is one of the reasons why.
I review AI toolkits for a living. I spend most of my time stress-testing agents, benchmarking inference pipelines, and writing honest takes on what actually works versus what just looks good in a press release. So when a chip company shows up on my radar, I pay attention — because the hardware underneath your stack is the part that quietly determines whether your agentic workflows run fast and cheap, or slow and expensive.
What Openchip Actually Is
Openchip is a full-stack system-on-chip and software company focused on AI and high-performance computing. They design their own RISC-V-based chips in-house, which already puts them in a pretty specific category — this isn’t a fabless startup slapping a brand on someone else’s silicon. They’re building the whole thing, hardware and software together, as a European systems company.
Their target launch window is 2028. That’s the date they’re working toward for bringing their agentic AI-focused hardware to market. Earlier this year, at MWC 2026 in Barcelona — alongside 4YFN and Talent Arena — they showed up and reinforced their position in Europe’s tech ecosystem. That’s not nothing. MWC is where real conversations happen between buyers, builders, and investors. Showing up there with a credible story matters.
Why 2028 Is Actually an Interesting Target
From a toolkit reviewer’s perspective, 2028 feels far away until you think about what the AI deployment space is going to look like by then. Right now, agentic AI is still in its awkward teenage phase. Agents are capable but expensive to run, often power-hungry, and heavily dependent on cloud infrastructure that adds latency and cost at every step.
The companies building hardware specifically for agentic workloads — not just LLM inference, but the multi-step reasoning, tool use, and memory retrieval that agents actually do — are positioning themselves for a market that doesn’t fully exist yet. That’s a calculated risk, but it’s not a blind one. The trajectory is clear enough that betting on 2028 as a meaningful inflection point isn’t unreasonable.
Openchip’s reported focus on lower-power designs is where things get genuinely interesting for anyone thinking about deployment costs. If you’re running agents at scale — and more teams are every month — power consumption and compute cost are the two variables that will make or break your unit economics. A chip designed from the ground up with those constraints in mind, rather than retrofitted from a general-purpose architecture, could offer real advantages.
What I’m Watching For
I want to be straight with you: the verified facts here are limited. We know Openchip exists, we know they’re targeting 2028, we know they use RISC-V and showed up at MWC26 with a credible presence. What we don’t have yet is benchmark data, pricing, or a clear picture of which part of the agentic stack they’re optimizing for.
That matters a lot. “AI and HPC” is a wide target. Are they going after edge deployment? Data center inference? On-device agents? The answer shapes everything — who their real competitors are, what the integration story looks like for developers, and whether their software layer is something toolkit builders will actually want to work with.
From where I sit, the questions I’d want answered before getting excited are:
- What does their software stack look like, and how does it interface with existing agent frameworks?
- Which workloads are they benchmarking against, and who’s doing the benchmarking?
- What’s the supply chain story — who’s fabricating the chips, and how does that affect availability at launch?
Europe Needs This to Work
There’s a broader point worth making here. Europe has been trying to build a credible AI hardware story for years, and most of the attempts have stalled at the “promising research” stage. A company that gets from Barcelona to a real product launch by 2028 — with solid agentic AI performance and competitive power efficiency — would do more for European tech credibility than a dozen policy papers.
Openchip isn’t there yet. But the direction is right, the timing is plausible, and the problem they’re solving is real. I’ll be watching the 2027 updates closely. If the benchmarks show up and the software story holds together, this one could be worth a serious look for anyone building agent infrastructure at scale.
🕒 Published: