Ora Computing, a startup developing software
to optimise and compress AI foundation models, has closed a €3.5 million seed
funding round led by Constructor Capital and Greencode Ventures, with continued
support from founding investor XISTA Science Ventures.
As AI adoption accelerates, the cost of AI
inference has become one of the industry’s most significant challenges.
Organisations deploying AI at scale increasingly face compute costs reaching
tens of millions of euros per month, while the growing size of foundation
models creates additional barriers for applications that require local
deployment on devices such as vehicles, industrial equipment, and edge
hardware.
Ora Computing addresses this challenge through
software that compresses AI models by up to 80 per cent, enabling them to run
up to four times faster while maintaining high performance, with accuracy
reductions typically ranging between 0 and 5 per cent. By reducing the
computational resources required for inference, the technology also lowers
energy consumption and associated carbon emissions. The company estimates that
achieving just 1 per cent market penetration could result in annual CO₂ savings exceeding 50,000 tonnes.
Stefan Sack, CEO and co-founder of Ora
Computing, said the company was created to rethink the conventional view that
larger models are the only path to achieving practical intelligence:
We believe the next wave of AI adoption will
be driven by compact, highly efficient models optimised for specific
applications rather than increasingly large, general-purpose cloud models. Ora
is building the software and algorithmic foundation that enables this
transition.
Unlike many existing model compression
approaches, Ora’s technology operates across different hardware platforms and
integrates directly with standard inference frameworks, eliminating the need
for custom software layers, infrastructure changes, or capital-intensive
retraining.
The company’s algorithms continuously map the
trade-off between model size and accuracy, allowing customers to optimise
deployments based on their specific hardware, performance, and cost
requirements. Ora has demonstrated this capability by compressing a
70-billion-parameter model within hours at a compute cost of less than $1,000,
compared with industry benchmarks that can reach hundreds of thousands of
dollars for similar tasks.
The newly secured funding will support team
expansion, further development of the company’s compression capabilities for
the largest frontier models, and the launch of a commercial product targeting
cloud inference providers and organisations deploying AI at the edge.