Today in AI: Frenemies, Voice Models, and a Chipmaker Having a Moment

Anthropic’s deal with SpaceX is stranger than fiction. Musk — who called Anthropic a company that “hates Western civilization” in February — is now renting Claude the most powerful AI cluster on Earth. The math isn’t hard: Colossus 1 was built to train Grok, but xAI’s user base never grew into it. SpaceX is only using 11% of Colossus’s potential. With a June IPO S-1 filing approaching, turning idle, expensive chips into a revenue stream from a marquee AI client looks very good on paper. Anthropic gets the compute it desperately needs; SpaceX gets a customer. Musk gets to call himself an AI Switzerland. The deal also includes both companies exploring data centers in space — because apparently terrestrial compute just isn’t aspirational enough anymore.

OpenAI dropped a three-model suite into its Realtime API today: GPT-Realtime-2 brings GPT-5-class reasoning to live voice conversations; GPT-Realtime-Translate handles 70+ input languages and 13 output languages without losing the speaker’s pace; and GPT-Realtime-Whisper streams transcription live as someone talks. Early numbers are loud — Zillow says call success rates climbed 26 points in testing, and Deutsche Telekom is already piloting multilingual support. The feature that’s quietly most interesting: the translate model handles full two-way conversations without requiring anyone to pause between sentences. Real-time UN-quality interpretation has entered the developer API.

AMD reported Q1 2026 revenue of $10.25 billion — a 38% jump year-over-year — powered almost entirely by AI infrastructure demand. The data center segment alone pulled in $5.78 billion, up 57% from last year. CEO Lisa Su called agentic AI the new driver of CPU demand, a claim Wall Street believed: the stock ripped 16% in a single session and has now more than tripled over the past year. The headline partnership: Meta committed to deploying up to 6 gigawatts of AMD Instinct GPUs, beginning with a custom MI450-based chip. Q2 guidance of ~$11.2 billion blew past the $10.5 billion estimate. In a market full of “AI companies,” AMD is one of the few actually printing money because of it.

The narrative has been that AI would hollow out creative work — writing, design, illustration, music. A new study tracking 2017 to 2024 found that even for roles with 70% AI task overlap, wages didn’t fall. They rose. Researchers and doomsayers are both confused. The working theory: AI raised the floor on what “baseline” creative output looks like, which pushed buyers toward higher-quality human work they could distinguish from the generated variety. Whether that dynamic holds for 2025 and beyond — when models took another significant leap — remains the actual question. But for now, the doom was premature. Creators who adapted appear to have actually won.