Vienna's Ora Computing raises €3.5M Seed to make AI models smaller, faster and cheaper to run

Vienna-based Ora Computing has raised a €3.5 million Seed round to compress and optimise AI models. The round was co-led by Constructor Capital and Greencode Ventures, with foundational investor XISTA Science Ventures also backing the company.

The cost is in running the model, not building it

Most of the attention in AI goes to training ever-larger models. The quieter, mounting cost is inference — actually running a model to answer a query, over and over, at scale. That bill is what increasingly determines whether deploying AI makes economic sense, and it rises with model size.

Ora Computing attacks that directly. Its software and algorithm stack compresses and optimises AI foundation models, the company says, cutting their memory footprint by up to 80% and running them up to four times faster. The payoff is twofold: models small enough to run at the edge — on local or constrained hardware rather than a distant data centre — and meaningfully cheaper cloud inference for everyone else. Techniques in this space typically trim a model's precision and prune redundant parts so it does more with less.

A contrarian thesis from quantum researchers

Founded in 2024 by Stefan Sack and Raimel Medina — both quantum-computing researchers from ISTA, the Institute of Science and Technology Austria — the company is built on a deliberately contrarian premise. "We founded Ora Computing to challenge the assumption that massive scale is needed to reach useful intelligence," Sack says. "The next wave of AI adoption will be driven by more compact models."

That framing matters because it cuts against the prevailing bigger-is-better orthodoxy. If the useful frontier can be reached by compact models rather than ever-larger ones, the economics — and the hardware requirements — of deploying AI shift for everyone downstream.

What the Seed buys

The capital goes to three things: growing the team, extending Ora's compression techniques to the largest frontier models, and launching a commercial product for cloud inference providers — the customers whose entire margin structure turns on the cost of running models efficiently. At €3.5 million this is an early, focused bet, but it is aimed at one of the most universal pain points in applied AI: not whether a model is capable, but whether anyone can afford to run it.