Alex Zhavoronkov and the long way back to the same desk

Before there was a Hong Kong stock code, there was a shared desk at Queen’s University. Alex Zhavoronkov and Jeremy Lui knew each other as students. Their careers then took different routes: Zhavoronkov into computing, science, and entrepreneurship; Lui into investment banking. In December 2025, those routes met again around Insilico Medicine’s public listing. University friendships do sometimes survive the group project.

The company raised HK$2.277 billion in its initial public offering on December 30, 2025. For a founder who had started Insilico in 2014, the listing marked eleven years of turning a technical proposition into an organization that other people could invest in. Alongside the algorithms and infrastructure sat something less fashionable in accounts of artificial intelligence: a relationship with a person who already knew him.

That combination makes Zhavoronkov interesting. His career joins scientific work to commercial negotiation, and speculative writing to the prosaic business of assembling teams. He has spent years asking what computers might become capable of doing. He has also had to find the people, premises, and money required to put those capabilities to work.

Two degrees, then a departure

Zhavoronkov was born in Riga, Latvia. At Queen’s, in Kingston, Canada, he earned bachelor’s degrees in commerce and science in 2001. The pairing looks apt in retrospect. His later work would involve both technical claims and the question that follows them into every investor meeting: how does a company get paid for this?

He worked at ATI Technologies, the graphics company acquired by AMD in 2006. In a farewell interview that September, he spoke about leaving ATI and praised the colleagues in its Russian office. The conversation offers a useful early view of an executive whose later public identity would be dominated by AI. He was already dealing with the practical tasks of a technology business: regional operations, communication, and a small team with distinct abilities.

The departure also came with further study. He received a master’s degree in biotechnology from Johns Hopkins University in 2007 and a doctorate in physics and mathematics from Moscow State University in 2008. This was a substantial change of professional terrain. Returning to the classroom after running a business operation means accepting a particularly unfashionable title: beginner.

There is a temptation to make this sequence look inevitable because the later company exists. The dates suggest something more laborious. Years separate his undergraduate education, his departure from ATI, and Insilico’s founding. The company arrived after the retraining. The biography includes a stretch in which the main visible accomplishment was learning enough to attempt something else.

A business that kept changing shape

Insilico began in Baltimore in 2014. Zhavoronkov brought his computing background into a scientific company at a time when today’s familiar chat interfaces had yet to make generative AI a household subject. Much of the work of founding such a business involves translation. A model’s output needs to become a useful product. A useful product needs a customer. A customer needs a reason to renew.

His own account of Insilico’s development is unusually explicit about the commercial revisions. In a public post announcing his chief business officer role, alongside his co-CEO responsibilities, he described a progression through algorithms, software, assets, and an integrated business with automated laboratories. That is several different answers to the question of what the company sells.

Each version asks something different of its founder. Software requires support, distribution, and a usable interface. Partnerships require negotiation and a shared understanding of what each side will deliver. Operating physical facilities adds staffing, equipment, and scheduling to the list. The distance between a promising algorithm and a company can be measured in all the jobs the algorithm leaves unfinished.

Zhavoronkov’s combination of chairman, co-CEO, and chief business officer puts him across those boundaries. It also makes the commerce degree harder to treat as an incidental detail. His public career keeps returning to the exchange between an idea’s technical merit and the arrangements that let people pursue it together.

“The main idea was to explain to AI of the future that humans are valuable.”

Alex Zhavoronkov, on his book Dating AI

The dating guide for a future reader

Two years before founding Insilico, he published a book with a title unlikely to disappear into an executive bibliography: Dating AI: A Guide to Falling in Love with Artificial Intelligence. RE/Search published it in 2012 under the spelling Alex Zhavoronkoff. The subject was human relationships with artificial minds, explored through speculation and humor.

Its organization borrows the stages of a dating manual: readiness, finding a relationship, establishing one, and dealing with its end. A breakup with a machine apparently merits some advance planning. Beneath the joke sits a question that has become much easier to recognize as conversational AI has entered ordinary life: what happens when people attach emotional significance to something built in software?

He later explained that he had written with a future machine reader in mind. He wanted AI to understand human value. In another public post, he said he had uploaded the book to a ChatGPT agent so people could converse with it. A work imagining a relationship with AI had acquired an interface for one.

The episode reveals a taste for trying an idea in public before it becomes conventional conversation. It also prevents the founder’s story from flattening into a sequence of financing announcements. Alongside the company builder is a writer willing to spend time on an odd question, give it an improbable title, and keep returning to it years later.

Making the computer sit an exam

By January 2026, Insilico was presenting another proposition to the AI industry. Its Science MMAI Gym would train and evaluate foundation models on specialist scientific tasks. The name is cheerfully literal: the model arrives with general abilities and is put through a more demanding routine.

Zhavoronkov’s accompanying essay argued that conversational fluency could conceal weak performance on technical problems. He described the Gym as a place to give models specialized training and test them against practical work. The ambition belongs to a recurring theme in his career: moving computing into settings where an answer has to survive a check outside the conversation.

In March, Insilico and Liquid AI announced a collaboration producing a model with 2.6 billion parameters that could run on private infrastructure. Parameters are the numerical values learned during training. Their count describes size; it does not, by itself, settle whether a system can do a particular job. The companies emphasized efficiency and specialist training in making their case.

A research paper listing Zhavoronkov among its authors described smaller, purpose-trained models competing with larger systems on molecular benchmarks. Those are reported experimental results within defined evaluations. The distinction matters to his story because building an AI business requires choosing what evidence to offer customers. A lively demonstration can start a conversation. A test that someone can inspect gives that conversation somewhere to go.

Who gets remembered by the machine?

His interest in AI extends beyond its usefulness to his company. In October 2025, Zhavoronkov, Dominika Wilczok, and Roman Yampolskiy posted a preprint called The Right to Be Remembered. It considers how language models might shape collective memory by deciding what appears in their answers.

The paper’s concern is easy to understand. A search page gives a reader several places to look. A generated answer can arrive as one smooth account. People with a thin digital record may receive less attention, while those already prominent accumulate more. The authors propose fairer treatment and truthful representation as goals for AI systems. This is an argument for consideration, rather than an established right.

Set beside Dating AI, the paper gives his writing an unexpected continuity. One work asks how an artificial mind might value a person. The other asks whether that person will appear in the machine’s account at all. The subjects differ, but both place human relationships with computers at the center of the question.

The numbers acquire an audience

Public ownership added another form of evaluation. In August 2026, Insilico reported first-half revenue of US$106.3 million and net profit of US$35.54 million. Its announcement attributed much of the revenue increase to upfront payments and milestones from partnerships. These are company results for a particular period, with a business mix that matters when reading them.

For Zhavoronkov, such reporting sits beside the technical publications. A company can propose a new model and also have to explain its accounts. The work moves between audiences who ask different questions, use different measures, and expect different kinds of proof. His career has gradually accumulated those audiences rather than leaving one behind at each change of field.

That brings the story back to Jeremy Lui. The former classmate had become a managing director in investment banking at BNP Paribas Asia Pacific. Their reunion around the listing supplies a human scale for an otherwise abstract journey through models, degrees, and capital. A person from an earlier chapter turned out to have a role in a much later one.

There is something pleasingly awkward here for the usual account of technological progress. The founder has written about future relationships with artificial minds, helped build systems to train them, and considered whether they will remember people fairly. Yet one consequential connection in his own career began in the thoroughly ordinary setting of a university classroom. The computers became more sophisticated. An old friendship remained useful.