Step aside, Nvidia, this global supercomputing stock is up 95% this year — and it's betting big on AI

Published Sun, Mar 17 20248:53 PM EDTUpdated Mon, Mar 18 20241:17 PM EDT

While chip darling Nvidia has grabbed headlines for its impressive stock performance, driven by excitement around artificial intelligence, an under-the-radar German company has quietly seen its shares nearly double in 2024. ParTec , a developer of supercomputing systems, has surged over 95% year-to-date as investors have started to take notice of its potential. While the company has been listed for less than a year on the Frankfurt Stock Exchange, its roots date back to 1999, when it was founded as a spin-off from the Karlsruhe Institute of Technology in Germany. The small-cap firm, with sales totaling 52 million euros ($56 million) in the first half of last year, bagged a 300 million euro ($327 million) contract to build the first ultra-powerful supercomputer on the continent The system, Jupiter, is expected to be capable of at least one quintillion — or billion billion — calculations per second. JY0-DE 1Y line How are investors taking it? Hendrik Leber, fund manager at Acatis Datini Valueflex Fonds, owns ParTec shares and says he sees the company as a "very clear" investment opportunity. "Europe wants to spend a couple of billion on supercomputing in the next couple of years, and they want to spend it locally in the EU. ParTec has the technology to do it and will be well-positioned for these public tenders," Leber added. What exactly does ParTec do? ParTec, with between 60 and 70 employees, provides the software systems needed to set up supercomputers, primarily for research. "Their main product is software 'middleware', and possibly quantum computers in the future," senior equity analyst Miguel Lago Mascato of equity research firm Montega told CNBC Pro. Lago Mascato expects the company's shares to rise by another 21% to 130 euros over the next 12 months. While competitors like Hewlett Packard can also bid on supercomputing projects, Lago Mascato said ParTec's middleware gives them a "unique selling point." Patents and royalties Earlier this year, independent auditors valued ParTec's 150 patents for building and designing supercomputers at 767 million euros, which the company expects will "significantly strengthen" its equity base. However, some are skeptical that the company will be able to monetize its patent portfolio. "At this point in time, our visibility for ParTec realizing licensing income from their patent portfolio in the near future is very low," Lago Mascato said bluntly. However, fund manager Leber dismissed concerns around the lack of royalties and pointed toward the 300 million euro Jupiter contract as an example of real cash flow. "Orders are very real and economic profits are also very real," Leber said. "I don't care about synthetic valuations of patent portfolios. I see the very realistic computing centers, and that's where the revenues and profits come from." Liquidity risks Another risk factor investors should be aware of with ParTec is the limited liquidity in its shares. Less than 13% of the company's shares are freely floated. Due to the thin trading volumes, Leber noted that even a 2 million euro stock placement took about a month to complete. This lack of liquidity means the stock price may not always efficiently reflect the company's underlying value. Future opportunities with AI ParTec sees significant opportunities beyond traditional supercomputing by providing systems tailored to train large AI models. Bernhard Frohwitter, chief executive of ParTec, told CNBC Pro that there was significant demand for systems with contextual understanding beyond just language models. Large language models, such as those behind ChatGPT, generate text, images, and videos by predicting the next word or pixel in a sequence. However, Frohwitter said companies are demanding AI models that require "interior logic" in physics, mathematics, and mechanics instead. ParTec is already "under enormous pressure from institutions, governments even, to build such machines" capable of training these foundation AI models, Frohwitter added. "We are in the process. In a few months, we will offer artificial intelligence machines that can do the training, data management, inference, and all that together," he said.