AI Austria monthly: March'25 Update

Our demands to kickstart AI development in Austria, Lower Austria opens an AI lab, AI incubator program for students, events, AI jobs & more

Matthias Grabner

Ting Wasner-Lian

Clemens Wasner

, and 2 others

Mar 04, 2025

Spring is just around the corner, bringing fresh momentum to AI and strategic policies. Explore the latest edition of our newsletter for insights on the newest developments in Austria and beyond.

🚀 Latest Developments

🏛️ Austria’s AI Future: Our Demands for the Next Government 🇦🇹

Austria has world-class talent, innovative startups, and leading researchers. However, without bold investment in AI, the country risks falling behind. While neighboring countries are significantly increasing their AI funding, Austria still lacks a clear strategy. Brain drain and the loss of competitiveness are growing concerns that require immediate action.

To provide suggestions our to the incoming administration we teamed up with AI Empowered politics, a non-profit promoting social understanding of AI in politic

Tax incentives and investment: Introduce tax benefits for AI startups and establish a national AI innovation fund.
AI education at scale: Double the number of AI professorships by 2027 and implement a nationwide AI education program.
Austria among the top five digital nations by 2030: Strengthen AI infrastructure and talent development to build a competitive advantage.
A dedicated AI ministry: Establish an independent AI ministry with a structured budget and a clear action plan.

We are looking forward to our exchange with the incoming administration and make your voice heard » full statement

👨‍💻 New AI internship incubator program

Bitmovin, a Carinthian video-streaming scale-up, has launched the “AI & Beyond Internship Incubator” to engage students from HTLs and technical universities in practical AI projects. Participants will collaborate with Bitmovin’s product development team and receive mentorship from experienced professionals. Notably, previous AI features developed by interns have been successfully integrated into Bitmovin’s products. » apply here

🤝 New Initiative: Young AI Leaders Linz Hub
Introducing the Young AI Leaders Linz Hub—the newest branch of the global Young AI Leaders Community by AI for Good. Organized by ITU in collaboration with more than 40 UN agencies and co-convened with the Government of Switzerland, this network spans 80 hubs across 46 countries. The Linz Hub offers mentorship, resources, and a collaborative platform for emerging innovators to craft ethical AI solutions that address real-world challenges 🌍🧑‍💻. AI Austria supports this initiative as it aligns with our mission to foster a digital landscape where innovation benefits every community. » Read more

🇦🇹 Lower Austria Launches Its Own AI Lab

A major step for AI innovation: Lower Austria is establishing its first AI Lab to drive research, business applications, and talent development in artificial intelligence. The lab aims to strengthen the region’s AI ecosystem by supporting companies, fostering collaboration, and accelerating AI adoption across industries. » Read more

📺 Interview Hohes Haus: Why is Austrian politics ignoring AI?

At the beginning for February Hohes Haus on ORF, ft. AI Austria and selected experts, took a closer look why AI remains absent from political debates—despite its transformative impact on our economy and society. » link to the segment

📊 KI Startup Barometer
We support the updated KI Startup Barometer—a study designed to capture the current sentiment in Austria on Artificial Intelligence (KI) and startups. In the survey, you’ll answer a few questions aimed at identifying challenges and opportunities, with the goal of developing actionable measures to improve the framework conditions for these fields. If you have 5 minutes, please share your insights to help guide our future initiatives. » Take the Survey 📝⏱️

✨ Generative AI News ✨

👉 OpenAI has introduced GPT-4.5, an upgraded AI model with improved accuracy, creativity, and emotional intelligence. It enhances pattern recognition, generates creative insights without reasoning, and reduces hallucinations from 61.8% in GPT-4o to 37.1%, although not blowing anyone away with its performance increase. It’s currently available as a research preview for Pro users and developers.
👉 Large Diffusion Models are a thing – yup, creating content out of noise, like it has been done with images very long is now coming to text as well. Large Diffusion Models (LDMs) leverage bidirectional context for better reasoning but suffer from higher latency due to multi-step generation, still showing immense potential. Find a comparison here
👉 Anthropic’s new Claude 3.7 Sonnet combines general-purpose language modeling with advanced reasoning, allowing users to toggle “extended thinking mode” for deeper problem-solving. Unlike competitors like OpenAI’s o3 and Grok 3, Claude 3.7 openly reveals its thought process in raw form, though Anthropic may hide this feature in future updates. The model outperforms rivals in software engineering and user interaction benchmarks while maintaining near state-of-the-art performance in fields like math and medical reasoning.
👉 Funfact of the month: DeepSeek shared its inference and revenue figures at the conclusion of its open-source week. While the company provides many services for free, full monetization of every token could generate around $200 million in annual revenue with strong profit margins. Source

🌏 Asia spotlight

Alibaba has announced that it is partnering with Apple to provide artificial intelligence features to iPhones in China.

Kunlun Tech unveiled China’s first AI short film creation video generation model, SkyReels-V1, and China’s first SOTA-level emotion-action controllable algorithm based on a video pedestal model, SkyReels-A1.

Chinese automakers are accelerating DeepSeek AI model adoption to power smart cars.

Universities across China have launched artificial intelligence (AI) courses in February based on Chinese AI startup DeepSeek.

Tencent releases new AI model, says replies faster than DeepSeek-R1.