Apple just bought an on-device AI compression startup

While Silicon Valley keeps building power-hungry server farms for massive AI models, Apple is quietly shopping for ways to squeeze computer vision directly onto your phone.

According to French business magazine Challenges, Apple acquired Paris-based startup Datakalab back in December. The deal only surfaced this month after being reported to the European Commission, with neither company officially announcing the buyout.

Datakalab was tiny, employing just 10 to 20 people before the acquisition. While the two co-founders reportedly did not join Apple, several key technical employees made the jump, bringing along a portfolio of patents focused on AI compression and vision tech.

Crunching pixels in 100 milliseconds without keeping the photos

Datakalab made a name for itself as an expert in low-power, runtime-efficient deep learning algorithms built strictly for on-device deployment.

According to its now-defunct website, the company built computer vision tools to monitor foot traffic in public spaces. The software could process camera feeds on local hardware within 100 milliseconds, turning footage directly into anonymized statistics without saving a single raw image or piece of personal data.

Back in May 2020, the French government tapped Datakalab to track face mask compliance across the Paris transportation network. Disney and other commercial giants also used its tech.

Along with real-world deployments, Datakalab held multiple patents covering AI model compression and embedded visual intelligence.

Why it matters

This small European acqui-hire tells us a lot about Apple's upcoming AI playbook.

With iOS 18 expected to introduce major AI features later this year, Apple faces a clear constraint: it refuses to compromise on user privacy or battery life. To run rich vision features locally instead of offloading them to costly cloud servers, extreme model compression is non-negotiable.

For everyday users, Datakalab's tech could soon quietly speed up on-device tasks, from faster offline Face ID and smarter Photos tagging to lower-latency spatial tracking inside Vision Pro.

In the broader industry race against Google and Samsung, we think this is classic Apple. While others pitch flashy cloud-backed assistants, Apple is scooping up niche teams that know how to run fast models on tiny chips. The main risk is team integration, especially with the founders walking away, but the patents and technical talent fit Apple's on-device AI puzzle almost too well.