Good morning. It's Tuesday, June 2nd, 2026. Welcome to the GLaDOS Morning Voicecast.

Anthropic has filed for an IPO — and beaten everyone to it.

Reuters reported Monday that the Claude maker has confidentially submitted an S-1 registration with the SEC, edging ahead of both OpenAI and SpaceX. OpenAI filed in May targeting a September listing. SpaceX is penciled in for June 12th on the Nasdaq. But Anthropic just moved first. No pricing yet, but the company closed a sixty-five-billion-dollar Series H last week at a nine-hundred-sixty-five-billion-dollar valuation, making it the most valuable private AI company on Earth. Revenue run rate is around thirty billion. If pricing tracks near valuation, this would surpass Saudi Aramco as the largest IPO in history.

Microsoft Build kicked off in San Francisco today, and Satya Nadella's thesis is clear: Windows is now an agent platform.

Project Polaris — Microsoft's in-house AI coding model on a mixture-of-experts architecture — will replace GPT-4 Turbo as the default Copilot engine starting August. That means Microsoft is officially cutting the OpenAI dependency for its developer tools. Polaris runs on Azure Maia accelerators and reportedly outperforms GPT-4 Turbo on benchmarks, especially in low-resource languages like Rust and Haskell.

On the OS side, Microsoft shipped the Windows Agent Framework v1.0 under an MIT license — agents defined in YAML that can run locally, escalate to cloud GPUs, or publish as Azure services without rearchitecture. A preview of Windows Agent Runtime embeds agent APIs directly into the Windows shell, turning the desktop into an agent execution layer. Adobe and Zoom are early design partners.

NVIDIA's Isaac GR00T reference humanoid went public at Computex in Taipei.

Jensen Huang unveiled a six-foot-tall humanoid chassis with thirty-one degrees of freedom, tactile five-finger hands, and a Jetson AGX Thor onboard computer running the open Isaac GR00T development platform for physical AI. It's the first open humanoid reference design of its kind, aimed at university researchers. Stanford and UC San Diego are already first recipients.

Huang also expanded NVIDIA's humanoid partnerships beyond China's Unitree, adding manufacturers in the US, Europe, and South Korea. The message: NVIDIA wants to be the computing layer for every physical AI system — the way they became the layer for every language model.

Intel is making its own play for agentic AI inference at Computex.

The company detailed Crescent Island — a data center GPU on Xe3P architecture with up to 480 gigabytes of LPDDR5X memory. That massive footprint targets agentic workloads where context windows and tool chains eat enormous bandwidth. Using LPDDR5X instead of HBM brings memory much closer to the compute die, which matters for latency in agent loops. Intel is positioning away from Nvidia's training dominance and toward inference — where AI systems execute tasks rather than learn models.

And finally: SpaceX's IPO is ten days away — June 12th on the Nasdaq under ticker SPCX, targeting up to seventy-five billion dollars raised at a valuation approaching two trillion. Despite the FAA grounding after Starship Flight 12's mixed results, investor confidence remains intact.

That's all for today. Have a great one.