DeepSeek warns of a ‘significant’ price rise, reversing its cheap-AI pitch
THE SO WHAT
DeepSeek walking back its cheap-API promise is a reminder that undercutting on price is rarely sustainable once scale and capex hit. If your roadmap leans on ultra-low-cost inference, model-agnostic abstractions and multi-vendor routing just became more important.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIGoogle may spend $1.5bn on a 35-person startup to catch up on AI coding
Paying over $1.5B for a ~35-person AI coding startup is a statement that agentic software engineering is now a strategic control point, not a side feature. If your product depends on developer mindshare, assume the majors will spend aggressively to own the coding workflow and plan your integration or defensibility accordingly.
Applied AISuno is now letting users press their AI music slop to vinyl, thus alienating streaming services, artists and audiophiles
Turning AI-generated tracks into physical vinyl pushes generative content from the feed into collectible objects — and into a thicket of rights, royalties, and platform politics. Labels, distributors, and merch platforms need a stance on AI-origin content now, before the long tail shows up in their catalogs and storefronts.
Googlers react to Jeff Dean's shock departure: 'Google is finally open-sourcing him for humanity'
One of the few remaining 'single points of technical gravity' in big tech research just moved from inside a fortress to the open market. For operators, expect more cross-pollination of frontier research talent into startups, labs, and consortia that were previously gated by big-company loyalty.
Applied AIReddit is ramping up its AI moderation tools. Moderators arent happy.
Reddit rolling out AI-powered Rules Hub to all new subreddits while mods fight to keep Automod is a governance clash, not just a tooling upgrade. If your product relies on volunteer or power-user moderation, expect similar resistance when you swap trusted primitives for opaque AI systems—co-design beats mandate.