A Caltech spinout called PrismML claims to have compressed a 27 billion-parameter AI model to run entirely on an iPhone, potentially challenging the dominance of cloud-based AI. The company emerged from stealth on March 31, 2026, backed by $16.25 million in seed funding led by Khosla Ventures, and says it reduced Alibaba's Qwen 3.6 model from roughly 54 GB to under 4 GB while keeping all parameters active.
The technology behind the claim
PrismML's proprietary compression techniques, branded as the Bonsai model family, achieve up to 14x smaller memory footprints and 8x faster inference compared to full-precision models. CEO Babak Hassibi emphasized that the approach retains every parameter rather than pruning capabilities, potentially reducing the need for massive data center investments that have driven AI infrastructure costs into the hundreds of billions.
Why Apple is reportedly interested
Apple has reportedly held discussions with PrismML about integrating its compression technology into future AI offerings. Apple's current on-device model, the AFM 3 Core Advanced, maxes out at 20 billion parameters with a sparse architecture. PrismML's technology could allow Apple to surpass that limit without redesigning its chip architecture, enabling full-capability LLMs that process data locally without sending it to the cloud.
Implications for crypto and decentralized compute
Running a complete LLM on-device aligns with crypto's core principle of user data sovereignty. However, it also threatens decentralized compute networks like Akash, Render, and io.net, which have attracted investment based on the assumption that AI inference demand will overwhelm centralized providers. On-device inference at this scale could compress the addressable market for cloud-based AI compute, particularly for consumer applications.
What to watch next
PrismML's claims remain largely unverified by independent benchmarks, and its $16.25 million seed round is modest compared to nine-figure AI startup raises. For crypto-native investors, the key signal will be whether Apple formalizes its relationship with PrismML and how that shapes the next generation of iPhone AI features.