← Explore more
Pro-AIPerson

Sepp Hochreiter

AI Researcher

Inventor of the LSTM — the architecture that powered a decade of speech and translation breakthroughs — now building xLSTM and the startup NXAI to give Europe its own competitive large language models.

In his words

xLSTM gives us a unique opportunity to revolutionize AI technology right from the heart of Europe and challenge the dominance of transformer models.
JKU Linz, 2024

Biography

Sepp Hochreiter (born 1967 in Bavaria) is a German computer scientist who heads the Institute for Machine Learning at Johannes Kepler University Linz. As a diploma student in 1991 he identified the vanishing-gradient problem, and with Jürgen Schmidhuber he invented Long Short-Term Memory (LSTM) — the recurrent architecture that, two decades later, ran the speech recognition and translation systems inside billions of phones. Colleagues in Austria like to say that without Sepp there would be no Siri.

A European answer to the transformer

Rather than treating the transformer era as the end of his story, Hochreiter went back to his own architecture. In May 2024 his lab unveiled xLSTM, an extended LSTM with exponential gating and matrix memory that he argues scales linearly where transformers scale quadratically — making it far faster at inference and better suited to industrial and edge applications. To commercialize it he co-founded NXAI in Linz (announced February 2024, with Netural X and PIERER Digital Holding), an explicit bid to build a European counterweight to OpenAI.

Hochreiter is blunt that Europe cannot outspend Silicon Valley on compute, so it must out-engineer it with more efficient models — NXAI has since released xLSTM-based models and applied the architecture to robotics and industrial time-series work. He is a frequent, sometimes prickly, critic of transformer orthodoxy and of Europe’s dependence on American AI infrastructure, but his complaint is never that AI is moving too fast — it is that Europe is building too little of it.

Where they stand in the war

Sources & further reading

Canonical record: https://battlelines.ai/topic/sepp-hochreiter