CEO Peter Sarlin on Silo AI's US$665m acquisition by AMD - the largest AI acquisition in Europe

Peter Sarlin is the CEO and Co-founder of Silo AI, a private Finnish AI lab based that was recently acquired by US chipmaker AMD for US$665 million – the biggest deal of the year in Europe in the field.

He recently talked to GOLD magazine about the state of European competitiveness, the rise of artificial intelligence and whether existential concerns about AI are premature.

The conference hall is now nearly empty, save for the few staff cleaning leftover water bottles and crumpled event programmes.

Beyond the hall’s doors, though, the Limassol Economic Forum still hums with post-event energy, punctuated by the occasional sound of laughter. In a quiet corner, Peter Sarlin, who a few hours ago, was on a panel discussing the sweeping impact of AI on business, recalls how, in 2017, when building the Helsinki-based Silo AI, he and his co-founders saw that Europe had a rich pool of top AI researchers primed to take on some of the toughest, most complex challenges and set out to create a space where these top minds could translate bold ideas into real-world solutions.

The pressing need for an EU single market

Today, the private AI lab has over 300 people on a roster brimming with PhD holders and senior academics. “I think,” he says, “it is extremely important that we have universities doing cutting-edge research and educating. But for that to create value, it usually doesn’t happen through an algorithmic innovation.” To create value, he explains, those algorithms need to be applied to specific problems in a specific industrial area. “And that requires you to be part of a product,” he adds.

In September 2024, in a report on the future of European competitiveness, former ECB chief Mario Draghi highlighted the stubborn snag that, for all this deep well of talent, examples like Silo AI are sporadic, as innovation tends to trip over commercialisation – a challenge keenly felt by the Cypriot ecosystem.

Despite sizeable funding in public grants, with the two state universities absorbing almost a third of the Research and Innovation Foundation’s €180 million budget (between 2016 and the first half of 2024), academic research rarely found its way to the market. Peter Sarlin says, “It’s a significant issue if we don’t have a strong and sizable tech sector in Europe – or Finland and Cyprus – because it’s very difficult to get the productivity gains we want to keep value in the region.” Those who control the end-users and the products embedded with AI, he says, will be the ones capturing most of the value generated.

So, what's the solution? Sarlin echoes another key point from the Draghi report: the EU must create a single market for businesses across Cyprus, Finland and beyond to participate in. In practice, many startups, particularly those in smaller markets like Cyprus, find themselves boxed in by local boundaries and this fragmentation stunts local scenes, resulting in only a small trickle of success stories.

In the AI arena, the gap between Europe and the US is massive. French Mistral AI recently hit a major €5.8 billion valuation after raising €600 million, a remarkable feat that still pales in comparison with Open AI, the creator of the GPT models, which had an estimated valuation of US$157 billion as of October 2024. Earlier this month, Sarlin became a signatory to the EU Inc. petition, one of the first initiatives to come out of the Draghi report, joining other big-time European founders to focus on how to make it easier for a European company to serve a single European market. “But eventually,” he says, “all this should really aim at fostering an environment where Europe builds what I’ve been calling ‘moonshots.’ In Sarlin’s vocabulary, moonshots refer to ambitious projects – much like what Silo AI is doing – and that’s the only way to attract talented people since they want to challenge themselves on the biggest stage. And capital will usually follow. “I’ve always believed that capital markets are global and if you have a good enough initiative, you’ll find private funding,” he says.

Pushing the boundaries of technology to solve real-world problems

Peter Sarlin's extensive research career revolved around a singular task: to solve real-world problems by pushing the boundaries of technology. He began by probing human behaviour through the prism of mathematical economics and econometrics, a path inspired by the specific, urgent questions raised by the 2008 global financial meltdown. But it didn’t take long for him to realise that modelling behaviour in a Hilbert space wasn’t going to fix the system. So, he pivoted to statistics and eventually machine learning, hunting for a way to build the guardrails of the financial system. This led him to work with the ECB (under Draghi’s stewardship), where he helped create the first models for financial stability surveillance – some of which are still in use today.

He also worked with the IMF and a series of central banks before moving to unsupervised neural networks, tools that underpin Generative AI. “I’ve been known to say that we embrace the problems, not the solutions – and that’s something we do at Silo. I think that’s what’s always been driving me: that need or desire and also the utility I personally get from managing to solve a specific problem and moving to the next one,” he says. Tackling one challenge after another is not only a source of motivation but also underscores for him an essential link between academia and entrepreneurship: to truly succeed in both, one needs an entrepreneurial mindset geared toward problem-solving.

How Silo AI’s powerhouse co-founding team first formed

In 2017, when the Finnish Government launched the Artificial Intelligence Programme to draw up a plan for an AI strategy, the then Prime Minster Juha Sipilä held an event to commemorate the initiative, which naturally drew quite an illustrious crowd, Sarlin included. It was there that he crossed paths with people that would form Silo AI’s powerhouse co-founding team: Juha Hulkko, Elektrobit’s founder and chairman; Johan Kronberg, PwC Finland’s former CEO and chairman; Tero Ojanperä, Nokia’s former CTO, Chief Strategist, and Head of Research; Kaj-Mikael Björk, a machine learning specialist; and Ville Hulkko, an entrepreneur specialising in AI with deep Silicon Valley experience. While for such a team, going bootstrapped wouldn’t be too hard, Sarlin reveals that there was a deliberate strategy behind their decision to launch Silo without external funding. “We believed it was crucial to take a sustainable approach, growing alongside our clients,” he says. By 2019, Silo AI was turning a profit and maintained a 100% average growth rate through 2022. That same year, it brought in its first outside investment for an undisclosed sum from Swedish private equity firm Altor.

The dangers posed by the new wave of artificial intelligence

As the deliberations of the Limassol Economic Forum fade into a murmur behind the doors, with an occasional housekeeping announcement blasting through the speakers – apparently, someone forgot to turn them off – the conversation with Sarlin shifts to a prominent theme of the panel discussion in which he participated earlier: the dangers posed by the new wave of artificial intelligence.

The potential of Generative AI to automate jobs, with Goldman Sachs projecting that it will affect some 300 million full-time jobs in the US and Europe, could exacerbate income inequality, disrupting societal structures the wrong way. The UN’s International Labour Organization, though, presented a global analysis indicating that Generative AI won’t be replacing entire job roles but will instead transform them. The debate is ongoing, with no clear verdict but the bigger concern for many is the issue of alignment: what if advanced AI systems develop goals that are not aligned with human goals?

For Peter Sarlin, any discussions about such existential threats are premature. “These large language models are doing token-by-token predictions. We feed them with trillions of words in huge documents and texts, of course, but the only thing they are doing is learning to predict the next token,” he explains. “Then you need to abstract up from that to understand the implications. There will be implications but these models are not going to be the ones that become sort of superhuman robots that take over the world. These robots will serve us and create a lot of value.” Sarlin does acknowledge some risks, particularly as Generative AI has passed the Turing test, blurring the line between human and robot. Indeed, its potential to be used as propaganda tool has brought together both sides of today’s political divide in a rare agreement.

Why Silo AI focuses on open-source models for specific industrial use

Silo provides AI infrastructure and tooling, a platform for model development and deployment and open-source foundational models. Much of its work nowadays centres on LLMs and Gen AI, customised and optimised for targeted industrial applications. It works in three segments: smart machines (which include autonomous vehicles), smart devices and smart society. It has done over 200 projects so far, collaborating with industry giants like Rolls Royce – where they leverage sensor data to support steering, with insurer Allianz – tackling NLP and text-processing innovations, and Unilever – working on the complex supply chain and logistics challenges. “But really, where we focus is on companies that have put software at the core of their business and AI at the core of their software,” he emphasises.

One of Silo AI’s standout innovations is the LLM Viking, trained on LUMI, the most powerful supercomputer in Europe, powered by chips from US giant AMD. Viking addresses a critical need: high-performing language models that are both open-source and effective in low-resource languages – languages with scarce digital content, like Greek. “We’ve built these open-source LLMs to strengthen European digital sovereignty and democratize access to LLM technology,” Sarlin explains. “It enables regional companies to create value on top of these models so that this value ultimately stays within Europe.” Currently trained in the Nordic languages, its scope will expand as development progresses. Viking is also built with translation pairs, equipping it with top-tier multilingual capabilities, with Sarlin highlighting that the development of base models built on data and information accurately representing the diverse European languages, citizens, organisations and the cultural landscape not only aligns with European values but also promotes digital sovereignty by allowing Europe to maintain control over how downstream applications and value creation unfold. Choosing to focus on open-source models for specific industrial use cases was also a calculated move, as competing head-on with the likes of Open AI on general purpose models would have been unwise.

AMD’s acquisition of Silo AI for a sizeable US$665 million

When news broke of AMD’s acquisition of Silo AI for a sizeable US$665 million, it sparked conversations about what this move means for Europe’s AI landscape. Sarlin’s response is measured, as is typical of such deals. He says, “It's a very exciting future.” The company won’t deviate from its roadmap with European language models but it will, of course, expand beyond that. And, he adds, this is a great opportunity for Silo’s team, reinforcing his earlier point that talented people want to tackle grand-scale problems. He revisits an idea promulgated by famous investor Marc Andreessen, who in 2011 proclaimed that “software is eating the world,” referring to how companies like Spotify and Netflix built the platform economy, fuelled by network effects. “At the core of that, where every user adds value and creates a better product, which then attracts more users, today sits AI.” And what of this virtuous cycle, where every industry is gradually transformed by software? AI is accelerating that transformation. “We will continue to be headquartered in Europe,” he says, “and we get to do that at AMD scale and create value to many of the companies we've been working with.”

Thanks to Parklane, a Luxury Collection Resort & Spa, where the photo was taken.

(Photo by Giorgos Charal.)

This interview first appeared in the November edition of GOLD magazine. Click here to view it.