By clicking submit, you agree to share your email address with TFN to receive marketing, updates, and other emails from the site owner. Use the unsubscribe link in the emails to opt out at any time.
By clicking submit, you agree to share your email address with TFN to receive marketing, updates, and other emails from the site owner. Use the unsubscribe link in the emails to opt out at any time.
Running large AI models is expensive. For companies operating at scale, compute costs can reach tens of millions of euros per month and continue to rise as models get larger.
Two quantum computing researchers at a top European science institute think the answer is to use smaller models rather than build more infrastructure. Their startup, Ora Computing, raised €3.5M in seed funding from Constructor Capital and Greencode Ventures to prove this idea.
The company compressed a 70-billion-parameter model in just a few hours for under $1,000, while industry standards put the cost at hundreds of thousands of dollars.
“We founded Ora Computing to challenge the assumption that massive scale is needed to reach useful intelligence. The next wave of AI adoption will be driven by more compact models that are highly efficient and optimised for specific use cases rather than large general-purpose cloud models,” says Stefan Sack, co-founder and CEO.
Ora focuses on the challenge of inference, which means running an AI model to get results. At a large scale, inference is one of the fastest-growing costs for businesses. The global AI inference market could reach $254.98 billion by 2030, growing at 19.2% each year. Models are now so large that running them in the cloud is expensive, and using them directly on devices such as cars or factory machines is often not feasible.
Ora Computing, based in Vienna, was founded in 2025 by Sack and Raimel Medina, who previously worked with the Serbyn group at the Institute of Science and Technology Austria, an independent research centre in Europe. They moved from quantum computing research to focus on what they see as a more urgent problem in AI: the high cost of running models once built.
The startup shrinks large AI models, cutting their memory use by up to 80% and making them up to four times faster, while keeping accuracy loss between 0 and 5%. Most compression tools only work with certain hardware or force users to pick between just a few compression levels.
Ora’s algorithm lets companies choose exactly how much to shrink their models and how much accuracy to keep. The solution works with standard inference frameworks and does not require custom software or changes to existing infrastructure.
The startup has also tested its technology with customers in the automotive and edge-silicon industries, where chips are built to run AI directly on devices rather than in the cloud.
The market for compression tools is crowded. Neural Magic, which raised $50 million from Andreessen Horowitz before Red Hat bought it in November 2024, worked on making AI models run faster on regular processors. Qualcomm’s open-source AI Model Efficiency Toolkit helps reduce model size while preserving accuracy, but it primarily works with Qualcomm’s Snapdragon chips. Intel’s Neural Compressor is a free Python library for Intel CPUs and GPUs, offering methods like quantisation, pruning, and knowledge distillation.
Both tools are free and popular, but they work best with their own makers’ hardware. Ora says its algorithm works across many types of hardware and lets customers fine-tune the balance between size and accuracy, rather than locking them into one vendor’s system.
Constructor Capital, an early-stage investor from Switzerland, and Greencode Ventures, a Helsinki-based firm focused on applied AI for energy, industry, and infrastructure, led the €3.5M seed round. XISTA Science Ventures, part of the ISTA innovation network and an early supporter, also joined as a returning backer. This is Ora’s first round of outside funding, bringing its total raised to €3.5M.
“AI’s energy appetite is growing faster than the world can build the infrastructure to feed it. Compressing models radically without sacrificing accuracy makes a tremendous difference to customers,” says Terhi Vapola, founder and managing partner of Greencode Ventures.
The new funding will help Ora grow its team, expand its compression technology to the biggest AI models, and launch a commercial product for cloud inference providers. Smaller models use less energy, and Ora estimates that if it reaches 1% of the market, its technology could cut more than 50,000 tonnes of CO₂ each year.
The AI efficiency field is attracting significant investment, but most of it goes to hardware such as custom chips, improved cooling, and larger data centres. Ora is taking a different path by focusing on software to make models run more efficiently. Whether this works will depend on how quickly the industry recognises that cutting costs may mean using smarter compression rather than just adding more infrastructure.
By clicking submit, you agree to share your email address with TFN to receive marketing, updates, and other emails from the site owner. Use the unsubscribe link in the emails to opt out at any time.
© 2026 Tech Funding News – All rights reserved.
Leave a Reply