At the Sifted Summit London 2024, leading AI experts gathered to discuss the evolving landscape of artificial intelligence during the panel "The Age of AI: What's Next for Europe?" The panel included Jonas Andrulis, Founder & CEO of Aleph Alpha; Connor Leahy, CEO of Conjecture; Carina Namih, Partner at Plural; and Rashid Ivaev, Founding Member and Global Sales Director at Nebius. The discussion was moderated by Tim Smith from Sifted.
Tribume gained seven key insights from the panel that startup founders need to know.
"Training is still happening, but there's more and more push towards inference — that's where the money is coming back." — Rashid Ivaev
Rashid Ivaev highlighted an industry shift from AI model training to inference. As a provider of compute resources, he noted that while training remains essential, it is a cost center for companies. The real revenue is generated during the inference phase when models are deployed and start delivering value.
"Training is your cost center. The only way you make money is when you have that model already being served. And then inference is the part where you can get the money back," Ivaev explained. He also mentioned that companies are focusing on training specialized, use-case-specific models to better commercialize their AI solutions.
"The European ecosystem is getting to that point of maturity where you have successful or indeed failed founders and operators trying again or coming back in as investors with all of that scar tissue to really help the next generation." — Carina Namih
Carina Namih discussed the maturation of Europe's startup ecosystem. Reflecting on her own experience of having to move to Silicon Valley to secure visionary investors, she now sees Europe as a fertile ground for ambitious founders.
"We are very pragmatic about playing the talent and capital arbitrage. You've got to be smart about using different jurisdictions to build your business," Namih said. She emphasized that with experienced founders and operators reinvesting in Europe, startups no longer need to relocate to Silicon Valley to succeed.
Connor Leahy echoed this sentiment, noting that London offers abundant talent and more accessible visas compared to the U.S. "I've been told horror stories about creating a startup in London that have not materialized. So far, I've found capital and talent to be very accessible here," he added.
“Once you want to start earning money, you might want to come to Groq or SambaNova or Cerebras from NVIDIA, because that's where these guys would bring you the most cost-optimized structure for you to make the better profit margin." — Rashid Ivaev
Addressing concerns over NVIDIA's dominance in AI hardware, Ivaev pointed out that while NVIDIA is a key player, companies like Groq, SambaNova, and Cerebras are challenging its monopoly. These companies offer hardware solutions that optimize costs, especially in the inference phase where profitability is crucial.
"NVIDIA is really pushing hard on the training component — that's their cash cow. But when we talk about inference, that's where you make the money, and you have to optimize the unit economy," he stated. Ivaev believes the market will evolve, reducing NVIDIA's dominance as other players offer more cost-effective solutions.
"What we're seeing is just this amazing advance in the simulation-to-real and back again loop." — Carina Namih
Namih expressed excitement about the advancements in embodied AI, where AI integrates into the physical world through robotics. She highlighted investments in companies like Monumental, which uses AI-powered robots to address labor shortages in industries like construction.
"You're able to take a video of a physical environment, train your robot to understand and self-drive in that environment, and then basically stick it in the space physically without requiring a huge amount of non-recurring engineering," Namih explained. This technology not only fills labor gaps but also frees people to engage in higher-value work.
However, she also emphasized the importance of safety and biosecurity as AI technologies evolve. "We think very carefully about where we're allocating resources to manage threats," she added.
"This is really the direction we're taking, asking what are the things that LLaMA cannot do? What is the thing Claude cannot do, but has great value potential in, for example, manufacturing?" — Jonas Andrulis
Jonas Andrulis discussed the potential for AI in manufacturing, particularly in Germany. He emphasized focusing on areas where existing models like LLaMA or Claude fall short.
Jonas Andrulis shared that Aleph Alpha recently collaborated with Bosch to capture the knowledge embedded in their manufacturing documents, which are like their own unique language. These specialized documents cannot be effectively fine-tuned into models like LLaMA because they don't fit well into the existing vocabulary, but Aleph Alpha found a way around this limitation. By addressing such specialized challenges, companies can create significant value in industries like manufacturing that can't be solved with widely used models.
"Is software good or bad? Doesn't make sense, because what will win is what is most addictive." — Connor Leahy
Connor Leahy raised concerns about the unregulated advancement of AI technologies. He compared the current AI race to historical instances where technology had unintended negative consequences.
"What happens if you build a system that's smarter than humans that we don't know how to control? We get out-competed," Leahy warned. He stressed that without proper oversight, the most addictive and profitable technologies could overshadow those that are beneficial or true.
"I believe in a few years, we'll look back at today's office work with the same horror we look at the hard backbreaking manual labor before the industrial revolution." — Jonas Andrulis
Looking optimistically toward the future, Andrulis envisioned a world where AI eliminates mundane office tasks. "People had to read hundreds of stupid emails, they had to do their stupid taxes—all this soul-crushing work that we do day in and day out," he said.
By automating these tasks, AI could free humans to engage in more meaningful activities. "I'm a huge fan of liberal democracy. It's amazing that we lifted billions of people out of poverty. Technology is always also responsibility and leverage and can go wrong, but I believe in its potential to improve our lives," Andrulis concluded.