Vincent Favrat
Inference AI Factories: Switzerland's Path to European Technological Sovereignty
At Panoramai's Real World AI panel, Vincent Favrat, Founder and Manager of Scale-Up-Factory, delivered the session's most visionary and strategically ambitious presentation. Drawing on philosophical concepts of planetary intelligence and leveraging Switzerland's precision innovation heritage, Favrat outlined a compelling blueprint for European AI competitiveness through revolutionary inference infrastructure that challenges conventional scaling assumptions.
The Philosophical Framework: Noosphere and Planetary Intelligence
The moderator positioned Favrat's vision within the noosphere —« a concept of both Pierre Teilhard de Chardin (Jesuit priest and a paleontologist) and Vladimir Warnowski (Russian geochemist) that thematized that earth as three layers:
the 1st is the Geosphere, made out of Inert matter
the 2nd the Biosphere, made out of Biological life
and finally the 3rd is the Noosphere, made out of the interconnections of thought and consciousness of humans.
This third layer would be all encompassing and connecting billions of people and devices in a new type of intelligence, a symbiotic type of intelligence."
This philosophical foundation informed Favrat's practical approach to AI infrastructure, positioning inference capabilities not merely as technical solutions but as foundational elements for humanity's next evolutionary layer of deep awareness and extended connectivity.
The Strategic Question: Europe's AI Infrastructure Reality
Favrat opened his analysis with fundamental questions about European positioning in global AI competition. « When I was told to speak about real world AI, I thought I don't want to speak about humanoid robots! I want to bring a theme that worries me and that gives me hope at the same time, » he explained, framing his presentation around both concerns and opportunities.
His core inquiry focused on enablers: « What is the enabler of real-world AI? And the answer to that for me is inference AI factories. » The follow-up question revealed his strategic ambition: « Can Switzerland play a role in this particular infrastructural shift and can it lead in Europe on this niche yet large market of the AI inference factories? »
The Investment Misalignment: Training vs. Inference
Favrat's most compelling analysis revealed a critical misalignment between AI investment patterns and real-world value creation. While acknowledging data quality and AI-Model training importance, he identified Inference as the overlooked critical component. « If you look at the life cycle of AI and the level of investment made in AI, actually only 10% is on the AI-model training and 90% is on adapting these models and applying them to the real life, to real use cases, to real business. 90%, not less than that, » he emphasized.
This insight reframes current European strategy. While the United States invests heavily in Model training infrastructure like Stargate with its 500 billion dollars announcement and Europe commits 200 billion euros to five gigafactories, Favrat questioned the strategic focus. « The question that we can ask ourselves is what type of factories are they? Are they training factories or are they inference factories? And the honest answer is that they are mostly training factories. »
The strategic vulnerability becomes apparent when considering that earlier speakers had declared the race for LLMs over. Favrat's response: « Houston, we have a problem, right? Because the future of AI and data sovereignty might be largely based on this overlooked infrastructural segment. »
The Swiss Solution: Precision Over Scale
Rather than competing with UAE, China’s and the US’ trillion-dollar investments through scale, Favrat proposed leveraging Switzerland's traditional strength in precision innovation. « What I think we can do, that's a question mark, is revolutionize the hardware setup of inference. And the way I see it is that we can take 0.005% of what is spent in the U.S. on a large scale project like Stargate and invest it in a very cutting-edge type of new factories, leading with a new generation of inference factories. »
The scale comparison illustrates Swiss competitive advantages: « If you need acres of land to deploy Stargate, we would need only around 25 square meters to build such a first blue-print factory. I see Pascal from Unlimistrust in the audience: would you have 25 square meters in the basement? I guess it's manageable, right? »
The Technology Trinity: Canton Vaud Innovation
Favrat identified three complementary technologies from Canton Vaud that could be packaged into revolutionary inference infrastructure. Cerebras, founded by EPFL's Jean-Philippe Fricker, produces wafer-scale processors representing « the best performing HPC solution today for this type of applications for inference factories. ». The company raised 750 million to date and achieves breakthrough performance. This Vaud invented technology has now become a US company—« We are used to the story. »
Advanced cooling technology from EPFL addresses energy consumption challenges in massive data centers (JJ Cooling for instance), while Exergo's waste heat recycling systems convert data center thermal output to heat neighborhoods and cities. Combined, these three technologies could create « the most powerful, greenest and most sovereign inference factory in the world. »
Performance Differential: Quantified Superiority
Favrat presented compelling performance data comparing current Azure capabilities to Cerebras systems for Llama4 processing. « 38, so we're speaking about on tokens per second, right? For Llama4 specifically, to 2,749. So it's not a close race! We speak about 20 times, 50 times, even 100 times faster. »
This dramatic performance advantage positions Swiss technology not as incremental improvement but as transformational capability that « comes from here and we could be implemented and scale from here. »
Strategic Applications: Swiss Excellence Sectors
Favrat identified three sectors where Switzerland's expertise demands high-performance inference capabilities. Pharmaceutical research could achieve « up to 80 times faster drug discovery. » Financial services, particularly Geneva's high-frequency trading sector, could boost performance « with a factor 15X. » Scientific research supporting climate modeling, genomics, and computational research at institutions represents the third critical application area.
« Why? Because we have the CERN, we have EPFL, ETH Zurich, Biopole, the SCDC, and many other institutions that are needing this type of powers, right here in Canton de Vaud and Switzerland » Favrat explained, positioning academic excellence as both market demand and competitive advantage.
The Capital Allocation Challenge
Favrat's most sophisticated insight addressed European capital allocation rather than scarcity. « In Switzerland and Europe generally, we don't have a problem of money because we are extremely wealthy and there is a lot of capital. The problem is of another nature, it is the misallocation of capital. »
The misallocation manifests in pension funds « building up piles of money and basically powering the US economy, the big techs of California. That means the money of all of you goes to power Meta and SpaceX and so on. » This creates a strategic vulnerability where European wealth finances American technological dominance.
The Scaling Vision: Network of Modular Micro Factories
Favrat's ultimate vision transcends single facility deployment. « Can we do a blueprint of this technology here and then scale it in Europe to make a network of micro factories that are mighty powerful? » he challenged, positioning Switzerland's traditional approach of being « little and innovative » as a competitive advantage for distributed inference infrastructure.
Sovereignty and Independence: The UBS Example
The European scaling potential addresses sovereignty concerns while leveraging Swiss precision. When addressing infrastructure dependence concerns, Favrat posed a provocative question about financial sector sovereignty. « Do you want like UBS to use the GPT models for your clients, I mean open AI for the banking system? Is that the Swiss sovereignty we want in terms of data protection or do we need something else? »
This example illustrates broader implications for European technological independence, positioning inference infrastructure as critical national security and economic sovereignty infrastructure rather than merely commercial advantage.
The Agentic Workflow Reality
Favrat grounded his vision in immediate market transformation. « We have to be consistent with the facts. The fact is that we are entering the agentic workflow era, and the power need for compute is going to explode. This is not a question. This is a fact. »
The choice becomes binary: « Either you build some European sovereign tech stack and factories or you outsource that to the US and its Far West or to China in the Far East, which is the case already very largely for companies of big size. Could there be a middle ground, in Europe, led by Switzerland. »
Innovation Heritage: Switzerland as an AI Sandbox and powerhouse for Europe
Favrat concluded by connecting his vision to Switzerland's innovation heritage. « So it's better to pump up good energy and to solve real life problems in real world AI, » he declared, positioning inference factories as continuing Switzerland's tradition of breakthrough innovations that solve global challenges. We have the brain power and we our regulation light, well situated outside of the AI act. These are key advantages when it comes to attracting talents and to innovate without being slowed down by overregulation. Can Switzerland become a little but highly efficient sandbox and powerhouse for AI in Europe?
The discussion was further enriched by practical considerations about infrastructure requirements. The moderator highlighted the explosive growth in token requirements for reasoning models, noting that companies with 200 developers using AI coding assistance would require 16 H100 GPUs, while current Swiss infrastructure provides only 4-8 GPUs, underscoring the infrastructure gap Favrat's vision aims to address.
Key Achievement: Favrat demonstrated how Switzerland can leverage its precision innovation heritage, strategic geographic positioning, and existing technological capabilities to create competitive advantages in AI inference infrastructure that address both immediate business needs and long-term European technological sovereignty while requiring minimal capital compared to competing approaches, positioning small-scale precision as superior to large-scale centralization for the distributed intelligence future.