A Weekly Journal of Turkish Founders

The Ledger

Dossier No 014 · 20 April 2026 · Istanbul

Founder & CEO · Theon

Deniz Karahan

On leaving Google Brain, building AI infrastructure in Istanbul, and why the next great technology company might not come from Silicon Valley.


It starts, as these things often do, with a WhatsApp message. "Come to Maslak," Deniz Karahan wrote on a Tuesday in March, "I want to show you something." The something turned out to be a server room — modest by Silicon Valley standards but improbable by Istanbul's — humming beneath the glass-and-steel offices of Theon, the AI infrastructure company Karahan founded in 2023.

Three years in, Theon has become the quiet engine behind enterprise AI adoption across Turkey and the broader MENA region. The company builds the middleware layer — the connective tissue between foundation models and enterprise systems — that allows organizations to deploy, monitor, and govern AI at scale.

What makes Theon interesting is not simply what it builds, but who builds it. Karahan, thirty-four, grew up in Ankara, studied computer engineering at Bogazici, spent four years at Google Brain in Mountain View, and returned to Istanbul in 2022. "They think I moved backwards," he told me, unprompted, within the first ten minutes of our conversation. "I think they're not looking at the right map." We spoke for nearly three hours. What follows is a condensed version of that conversation.

Let me start with the obvious question. You had a position at Google Brain. You were working on some of the most interesting problems in machine learning. Why leave?

Because I was working on problems, not on outcomes. At Google Brain, the intellectual stimulation was extraordinary — I was surrounded by people whose papers I had studied in university. But there is a difference between advancing a field and building something that changes how people work. I started to feel like I was contributing footnotes to other people's theses. I wanted to write my own.

The more honest answer is that I missed consequence. When you work at a company of Google's size, your contribution is diffused across so many layers that you lose the direct line between what you build and what it does in the world. I wanted that line back.

Why Istanbul? You could have started Theon in San Francisco, Berlin, London — anywhere with a deeper tech ecosystem.

There is an assumption embedded in that question, which is that a deeper ecosystem is necessarily a better one. I am not sure that is true. A deeper ecosystem gives you more resources, yes, but it also gives you more noise, more competition for attention, more pressure to conform to whatever narrative the market is currently rewarding.

Istanbul gave me something else: proximity to a problem that the deeper ecosystems were ignoring. Enterprise AI adoption in Turkey and the wider region is not a technology problem — the models exist, the cloud infrastructure exists. It is an integration problem, a governance problem, a trust problem.

And there was a personal dimension. I grew up in Ankara, I studied in Istanbul. My parents are here. I had spent four years living in a country where I was always, in some small way, a guest. I wanted to build something at home.

Describe Theon's product for someone who has never thought about AI infrastructure.

Think of it this way. A foundation model — GPT, Claude, Gemini, whatever — is like an engine. A very powerful engine. But an engine alone does not get you anywhere. You need a chassis, a steering wheel, brakes, a fuel system. You need someone to build the car.

Theon builds the car. We provide the orchestration layer that sits between the AI models and the enterprise systems — the CRMs, the ERPs, the internal databases, the compliance frameworks. We handle model routing, observability, and governance.

It is not glamorous work. Nobody writes breathless blog posts about middleware. But it is the work that determines whether AI actually gets used in production or remains a series of impressive demos.


How would you describe the Turkish tech ecosystem in 2026?

Improving, but still adolescent. There is genuine talent here — the engineering programs at Bogazici, ODTU, Bilkent produce world-class graduates. There is more capital available than five years ago, though still far less than in comparable economies.

What is still missing is institutional memory. Silicon Valley has seventy years of accumulated knowledge about how startups work. We are building that knowledge base in Turkey from scratch, one company at a time.

The other thing that is missing is patience. The Turkish market rewards quick exits and visible growth. Building deep infrastructure requires a longer time horizon.

Tell me about the fundraising experience. What is it like to raise capital as a Turkish founder?

It is an exercise in translation. Not linguistic translation — my English is fine — but contextual translation. When you pitch to a European or American VC, you are not just explaining your product. You are explaining your market, your regulatory environment, and — whether anyone admits it or not — you are arguing for the legitimacy of your geography.

I had a meeting with a well-known fund in London where the partner spent the first twenty minutes asking me about political stability. Not about our technology, not about our customers. About politics.

We ended up raising our seed from a combination of a Turkish family office, a Gulf-based institutional fund, and a European VC that had previously invested in the region. The key was finding investors who already had the context.

How do you build a team in Istanbul? You are competing with Google, Microsoft, Amazon for the same engineers.

You compete on mission, not compensation. I will never match Google's salary for a senior ML engineer. But I can offer something Google cannot: ownership of the problem.

We also benefit from a talent pool that is often overlooked. There are extraordinary engineers in Turkey who, for family reasons or personal preference or visa complications, are not going to relocate to San Francisco. They are not less talented. They simply made a different choice. Theon is built for them.

The culture matters too. We are a company of thirty-two people. We have lunch together every day. There is a warmth and a directness to how we work that I think is distinctly Turkish.


Who are your competitors? How do you think about competing with global infrastructure companies?

Our competitors fall into two categories. The first is the global players — LangChain, the various orchestration frameworks, the cloud providers' own AI platforms. They are formidable but they are building for the global median.

The second category is other regional players, and honestly, there are very few.

Our advantage is that we are the only company that can deploy an orchestration layer that complies with KVKK — Turkey's data protection law — while also handling Arabic-language document processing for a Gulf client. That specificity is our moat.

Tell me about a moment when you thought Theon might not make it.

August 2024. We had been operating for about a year. We had built the core product, signed two pilot customers, and were burning cash at a rate that gave me about five months of runway. Then both pilots stalled simultaneously.

I remember sitting in this office, looking at the financial model, and realizing that if neither pilot converted to a paid contract within three months, I would have to let go of half the team. The weight of that responsibility was unlike anything I had experienced at Google.

What saved us was a third client who accelerated their timeline because a competitor had just deployed an AI system that was producing compliance violations. We signed a twelve-month contract in September 2024, and that single deal gave us the runway to survive.

What drives you? When the fundraising is hard and the clients stall, what keeps you going?

Stubbornness, mostly. There is a particular kind of stubbornness that I think is required to build a company in a market that the global tech industry has largely written off.

Beyond stubbornness — I am motivated by a very specific vision of what Turkey could be. This country has eighty-five million people, a young and educated population, and a deep entrepreneurial culture. The idea that Turkey should be a consumer of technology built elsewhere, rather than a producer of its own, is not a law of nature. It is a failure of ambition and infrastructure.

And I should say — I enjoy the work. Building systems, solving hard problems, working with smart people.


What is next for Theon?

We are expanding into two adjacent markets: the Gulf states and North Africa. The enterprise AI adoption curve in those regions is about eighteen months behind Turkey.

On the product side, we are building what we call "governance-native AI" — AI systems where compliance, auditability, and control are built into the foundational architecture.

If we do this well, Theon becomes the default infrastructure layer for enterprise AI across the Eastern Mediterranean, Turkey, and the Gulf.

But I am also trying to resist the temptation to move too fast. I would rather be indispensable to fifty clients than superficially useful to five hundred.


Closing Note

Karahan walked me to the elevator after we finished talking, and we stood for a moment looking out the floor-to-ceiling windows at the Bosphorus, just visible between the towers of Maslak. "Everyone who visits this office comments on the view," he said. "But the view I care about is the one facing the other direction." He gestured toward the server room we had visited hours earlier. "That is where the future is being built."

It is an earnest thing to say, and Karahan is an earnest person — a quality that would be a liability in the ironic precincts of San Francisco but feels entirely appropriate in a city that has been reinventing itself for two thousand years. Whether Theon achieves the regional dominance its founder envisions remains to be seen. But there is something persuasive about a person who left the most prestigious AI research lab in the world to build middleware in Maslak, and who, three years later, seems more certain than ever that he made the right choice.