Ora Computing Raises €3.5M Seed to Shrink AI Models by 80% — and Slash Compute Costs

Ora Computing was founded by quantum-computing researchers Stefan Sack and Raimel A. Medina, who spun out from the Institute of Science and Technology Austria (ISTA) . The company emerged from stealth in late 2025 and has already validated its solution with players in the automotive and edge silicon sectors .

How the Compression Software Works

Ora's software uses an information-theory-based approach to shrink large AI foundation models for inference — meaning it mathematically restructures the model to remove redundancy while preserving output quality . Key claims from the company and its investors include:

• Up to 80% size reduction: Models can be shrunk dramatically in memory footprint .
• Up to 4× faster inference: Compressed models run significantly faster on the same hardware .
• Accuracy loss of only 0–5%: The trade-off between compression and accuracy is minimal .
• Hardware-agnostic: The software works across different chip architectures (CPU, GPU, edge silicon) without custom infrastructure or retraining .
• No retraining needed: Drops directly into standard inference frameworks .
• Continuous compression curve: Unlike binary "choose a compression level" tools, Ora's algorithm maps the full size-accuracy trade-off continuously, letting customers optimise for their hardware and cost constraints .
• Proven on a 70B-parameter model: The company says it compressed a 70-billion-parameter model in hours for under $1,000 in compute, versus typical costs of hundreds of thousands of dollars .

Cost, Energy, and Carbon Impact

Because compressed models require less compute, Ora states that its technology translates directly into lower energy consumption and reduced carbon emissions . The company estimates that at 1% market penetration, its technology could eliminate more than 50,000 tonnes of CO₂ annually .

This aligns with the broader industry consensus that inference accounts for over 90% of the lifetime energy use of deployed AI models, making compression a powerful lever for sustainability . A 2025 study in PMC found that applying model compression techniques like pruning and distillation to BERT-size models resulted in energy reductions of 32–48% while maintaining accuracy above 95% . Ora's specific CO₂ savings estimate, however, is labelled as an "estimate" at a benchmark penetration rate, not a measured result .

Funding Use and Commercial Plans

Ora has stated three primary uses for the seed capital :

1. Team expansion — growing the engineering and commercial team.
2. Extending compression to larger frontier models — pushing beyond current capabilities to compress the next generation of state-of-the-art models.
3. Launching a commercial product — targeting cloud inference providers and companies deploying AI at the edge (e.g., automotive, industrial equipment) .

The company's own website now advertises an automated LLM compression service that can fit models on edge devices, on-prem servers, or cloud infrastructure "in hours, not months," with claims of reducing memory footprint by up to 70% and cutting GPU bills by over 50% .

Source Reliability Note

The primary source is a detailed announcement published on Greencode Ventures' official website . Corroborating coverage comes from multiple outlets including The Recursive , IT Boltwise , Vestbee , tech.eu , and Fundup.ai — all reporting the same round details, compression figures, and use-of-funds plan. No independent third party has yet validated Ora's specific CO₂ savings estimate, which the company labels as an estimate rather than a measured result.