The Story of SEMRON: The Company Building AI Chips That Cost Dollars, Not Thousands

The Story of SEMRON: The Company Building AI Chips That Cost Dollars, Not Thousands

The Eagle Pub in Cambridge has witnessed its share of scientific breakthroughs. In 1953, Francis Crick walked in and announced he’d discovered the structure of DNA. Decades later, two German university students sat at the exact same table, eating fish and chips, discussing semiconductor innovation. In a recent episode of Category Visionaries, Aron Kirschen, CEO of SEMRON, an AI chip maker that’s raised $7.3 million, explained how that Cambridge moment sparked a journey that would eventually lead to a company trying to solve one of AI’s most fundamental problems: cost.

The Fish and Chips Origin Story

Aron met his co-founder during their first semester at university in Dresden, Germany, through a scholarship program. His co-founder “was really into microelectronics, even in school, patented stuff or sense in that.” But it was on a study trip to Cambridge where things crystallized.

“I remember my first big memory with him together was in the Eagle Pub in Cambridge where we sat at a table eating the fish and chips thing and we sat at exactly the same table as Francis Crick, who invented or discovered the DNA structure or helix structure,” Aron recalls.

They kept discussing through the years, bouncing around ideas about overcoming the physical limits of current semiconductor technology. Initially, “it was nothing that seemed to be realistic in like five years. We were more like playing around.”

Then something shifted. The idea became concrete enough to patent. When the university heard about it, they offered to pay for the patents. But Aron and his co-founder made a crucial decision: “We were like, no, even though we don’t have much money, we try to find enough family, friends, money to pay ourselves.”

That choice—maintaining ownership despite limited resources—set the trajectory for everything that followed. The co-founder wrote his diploma thesis on the technology. Then he became Aron’s supervisor, and Aron wrote his diploma thesis on it as well. “I think that was more like the starting point before we founded Zemrun in 2020.”

The Pandemic Pivot That Made Them German

Timing is everything in startups, and SEMRON’s timing was complicated. They founded in 2020, right as the world shut down. The original plan was to move to the United States and start the company there after participating in Berkeley’s SkyDec program.

“We filed or we tried to get a visa to move to US to start the company there. But the embassy was like, okay, no, we are not even. Yeah. Looking at new applications right now because it’s not going to work anyway,” Aron explains.

Berkeley’s program went virtual, and SEMRON became a German company by necessity. “It would have store right away founded in us but because it wasn’t possible. We are a German company right now, so good for Germany in that way,” Aron notes with a touch of irony.

They needed a legal structure to hold their patents for investors, which led them to found a German GmbH—”which is definitely not the best structure in the world to have a startup,” Aron admits. But sometimes constraints create unexpected advantages. Dresden, it turns out, isn’t a bad place to build a semiconductor company.

Dresden: Europe’s Semiconductor Center

Dresden’s semiconductor story traces back to an unlikely source: the socialist German Democratic Republic. “We had a huge program in the GDR, the Democratic Republic of Germany, where they really wanted to do some semiconductor stuff for some time,” Aron explains. They even produced the first Megabit chip.

When the Berlin Wall fell, “we still had a lot of talent here and a lot of industry around that center.” Western German leaders concentrated investment into existing structures, and Dresden became “the center first of Germany for all the semiconductor stuff. And now it’s even center of microelectronics on a European level.”

Today, Dresden hosts TSMC, Infineon, Global Foundries, Bosch, and xfab—”a lot of fabs here, a lot of suppliers. Definitely bigger than everywhere else in Europe.” For a semiconductor startup, this ecosystem matters enormously. “From that perspective, if you have a semiconductor startup, Dresden is a great place.”

The local startup scene reflects this specialization. “I think the percentage of the hard tech companies in Dresden is definitely higher than any other city,” Aron observes. It’s not the size of Munich’s tech scene or Berlin’s software ecosystem, but for hard tech companies and sensor companies, Dresden offers something unique.

The ChatGPT Moment They Saw Coming

November 2022 changed the AI landscape for most people. But SEMRON had been preparing. About a year before ChatGPT exploded into mainstream consciousness, an advisor from Berkeley told them: “Okay guys, you have to have a look on transformer based large language models.”

When ChatGPT launched, “it was like, great for us. So we are kind of prepared,” Aron says. The real surprise wasn’t the technology—it was the societal impact. “In January 2023 it became a big topic everywhere. Not just in our office, on the streets, you know, everywhere.”

But Aron sees an incomplete revolution. “When it comes to edge devices, we still miss this kind of chatgpt moment.” ChatGPT transformed data center AI and web-based productivity, but smartphones and wearables haven’t experienced their paradigm shift yet.

The reason, in Aron’s view, is straightforward: “We believe the reason why we don’t have the killer app is because you don’t have the hardware. And with not having hardware, I mean we won’t have it with the current approach in five years from now because there’s no way current technology can do that.”

The example he gives crystallizes the problem: “To run for example a large language model for live translation like continuously in a discussion there’s nothing you can do, right? Or nothing without battery is down in like some minutes.”

The Economic Constraint That Defines Everything

At the heart of SEMRON’s strategy is a brutal economic truth. Technical capability doesn’t matter if nobody can afford it. “Right now, even though it would be energy efficient enough to put an Nvidia graphic card in a smartphone and probably smartphone will be like five times bigger. But assume we could, we would not be able to make a product out of it because nobody would pay for like $5,000 more for something like that or even $200 more.”

The target is precise: “We really have to be like let’s say 20, $30.”

This isn’t a marginal improvement story. This is about making an entirely new category of devices economically viable. “We find a way or manage to build a chip. It costs a couple of dollars. Able to run large language models with billions of parameters as continuously power consumption of less than 500 milliwatt.”

The Vision: Scaling Human Productivity

When Aron looks three to five years into the future, he doesn’t just see better chips. He sees a fundamental shift in how humans work. “In the last centuries the only thing people could use to scale up productivity was basically using other humans time. And also of course computer helped a lot with Excel and so on to save some time and increase productivity.”

But that’s not enough. The real breakthrough comes when “everybody of us would have a completely personalized assistant, completely understanding that as a kind of default thing in a smartphone, always listening, completely private because running on device.”

This vision requires solving the cost problem. When compute becomes “cheap as hell,” revolutions happen. “Every revolution became the impact on society. For a revolution happened when the driving factor became cheap as hell, like when paper was as expensive as sand. It became a real revolution.”

The same principle applies to AI compute. “I think we can really enhance human productivity by factor of I don’t know, but not like 20% and come up with like completely new society. And I think that’s what’s needed and I think it can’t be underestimated what happens with the society when we have that.”

Then Aron makes the bet that every founder eventually makes: “Actually, I think, I believe Zemrun will be exactly the company that enables that.”

It’s a bold claim. But so was sitting in the Eagle Pub, at Francis Crick’s table, dreaming about overcoming the physical limits of semiconductor technology. Sometimes the audacious visions that start over fish and chips are the ones worth building.

From Cambridge pub conversations to Dresden’s semiconductor ecosystem, from pandemic pivots to the vision of personal AI assistants that fundamentally scale human productivity—SEMRON’s story illustrates what it takes to build hardware that could reshape society. The question isn’t whether the vision is ambitious enough. It’s whether the world is ready for AI compute that costs dollars instead of thousands.