Japan’s Sakana Builds a Fable Alternative, SpaceX Rents Colossus to Everyone, Google Backs A24

Today we’re covering the biggest AI stories of June 23rd, 2026. Japanese lab Sakana AI launched Fugu — a model that routes each request through a pool of AI models behind a single API, claiming frontier-level performance without relying on any one provider, and positioning itself as the direct answer to the export control risk that pulled Anthropic’s models offline — though early user reviews are mixed, with some reporting the experience doesn’t match the benchmark claims. SpaceX is quietly turning its Colossus supercomputer into one of the most valuable compute rental businesses in AI — with Anthropic paying $1.25 billion a month, Google at $920 million, and now stealth startup Reflection AI signed on for $6.3 billion in Nvidia compute, all from infrastructure that was originally built just to train Grok. Plus, Google put $75 million behind indie film powerhouse A24 and paired them with DeepMind researchers to build filmmaker-shaped AI tools — not generic video generation — and today’s community workflow comes from Hasnain in Toronto, who fed Claude his company’s outdated ERP API documentation and built automated twice-daily sales reports, Zapier integrations to ShipStation, and real-time inventory tracking — work he says would have cost thousands to hire a developer to do.