Large AI models, the massive troves of language, vision, and audio data that power generative artificial intelligence services, are becoming as important to AI development as operating systems were to smartphone development. They are almost like platforms. Jua, a Swiss firm, is employing that approach to explore novel AI applications in the physical world. It received $16 million to create a big natural “physics” model.
The firm is young. It will first model and predict weather and climate patterns for participants in the energy industry. The business indicated it would launch in a few weeks. Its approach will also target agriculture, insurance, transportation, and government.
Promus Ventures, Kadmos Capital, Flix Mobility founders, Session.vc, Virtus Resources Partners, Notion.vc, and InnoSuisse are joining 468 Capital and the Green Generation Fund in this seed round for the Zurich business. 
Andreas Brenner, Jua’s CEO and co-founder with CTO Marvin Gabler, says that the increasing “volatility” of climate change and geopolitics has necessitated more accurate modeling and forecasting for physical world organizations like energy and agriculture. According to the U.S. National Centers for Environmental Information, 2023 was a high-watermark year for climate disasters, causing tens of billions of dollars in damage. This is driving organizations to have better planning tools and predictive tools for market analysts and others who use that data.
This is not a new problem, and engineers have already been using AI to solve it.
DeepMind produced GraphCast, Nvidia FourCastNet, and Huawei Pangu, which released a weather component last year to much interest. As reported last week, a team is using meteorological data to construct AI models to study other natural events, such as bird migration patterns.
Jua replies twice. Firstly, it claims that their model is 20x better than GraphCast since it consumes more data and is bigger. Second, weather just scratches the surface of physical issues, solutions, and problems.
He stated, “Businesses must improve their capabilities to respond to all this [climate] volatility.” So we’re fixing that problem in the short term. As we consider the future, we are creating the first basic model for nature. Building a computer model that learns physics is crucial for establishing artificial general intelligence, as language understanding alone is insufficient.
Although the firm has yet to produce its first goods, investors are taking a risk beyond AI hype.
Gabler led research at a weather forecasting company. Q. met and worked on deep learning technologies for the German government before joining Jua. Brenner developed a fleet management software firm and worked in energy. Together, these experiences provide technical awareness of the issues and possible solutions and direct comprehension of the industry’s perspective.
As it develops the product, it shows investors and potential clients early work for data input.
One goal appears to be to rethink predictive model components. In creating a weather prediction model, Brenner said, “Using weather stations is pretty obvious.” To create its models, it also consumes “much more noisy data,” such as current satellite photos, topography, and other “more novel, recent data.” “The key difference is that we are building this end-to-end system where all of the data used in different value chain steps is now in the same pool,” he said. The business claims 5 petabytes of training data, compared to 45 for GPT3 and 1 for GPT4. However, linguistic data may require less data than a physical world model.
Another major goal is to design something more efficient to lower operating expenses for the firm and its consumers. Brenner added, “Our system uses 10,000 times less compute than the legacy systems.”
Jua’s rise and funding are noteworthy.
Foundational models will underpin the next generation of AI applications, so organizations that construct and manage them are valuable and powerful.
OpenAI, Google, Microsoft, Anthropic, Amazon, and Meta—all U.S. companies—are the top players in this space. That has led some in Europe to find and finance domestic champions as alternatives. 468 Capital also backs Germany’s Aleph Alpha, which is constructing huge language models like the U.S. pioneers but in closer partnership with potential consumers. Its slogan is “Sovereignty in the AI era.”.
Ludwig Ensthaler, a general partner at 468 Capital, said, “Andreas, Marvin, and the team are building the world’s first foundation AI for physics and the natural world, which will be capable of providing powerful insights for a wide range of industries dependent on true understanding of nature, from insurance companies and chemical and energy providers to disaster planning teams, agriculture organizations, airlines, and aid charities.
An AI startup that seeks to comprehend climate change, improve catastrophe preparation, and maybe minimize environmental damage has a “good guy” vibe. The greater picture for a firm building an AI that can understand the physical world is that it may be used in more material science, healthcare, chemistry, and other difficulties. The idea raises several problems, similar to those facing other AI models, concerning safety, dependability, and more, which Jua is already considering, albeit informally.
“Models work and are accepted when you enforce consistency,” Gabler remarked. “You must ensure the models learn physics from scratch to solve problems correctly.”
