Willow picks CoreWeave for AI training, skips DIY infrastructure

Willow Care, the company behind the AI dictation app Willow Voice, has turned to CoreWeave for its model training needs, opting not to build or maintain its own infrastructure. The decision, reported by SiliconANGLE, highlights how some AI startups are prioritizing providers that offer more than just compute power.

The app blends speech recognition with language processing, which likely requires ongoing training. CoreWeave’s appeal, per Willow co-founder and CTO Lawrence Liu, includes its “forward-deployed support”—a service layer that may ease deployment and scaling. This aligns with CoreWeave’s recent moves, including its push toward a full-stack AI cloud for agent workflows, as we noted last week.

The situation reflects a common challenge in enterprise AI: companies developing complex models sometimes struggle to manage the infrastructure behind them. While open-weight models are narrowing the gap with proprietary systems, the shift toward private, storage-backed models—another trend we covered earlier this month—could make backend operations even more demanding.

CoreWeave’s collaboration with Nvidia on the Vera CPU, aimed at agentic AI workloads, suggests the company is exploring infrastructure that might handle both model reasoning and task execution. For Willow, working with CoreWeave could free up resources; for CoreWeave, the partnership offers a chance to see if startups are willing to pay for more than just GPUs.

Sources: siliconangle.com