Story Spacex Control
Orbital AI, Chip Wars, and a Talent Drain Threatening Google DeepMind
SpaceX has confirmed the name 'Starmind' for its planned orbital AI constellation — a system of up to one million satellites designed to run AI inference directly in orbit rather than routing data through ground-based data centers. With an FCC filing real and a trademark filed, the scale is staggering: one million satellites would be roughly thirty times the current Starlink constellation. The concept envisions distributed computing infrastructure in low Earth orbit that eliminates ground-based latency pathways. The announcement arrives as SpaceX shares are reportedly sliding amid what journalists are describing as a 'governance backlash,' with former SEC Chairman Gary Gensler predicting a 'great rebalancing' as early investors' lockup periods approach expiration in August — a concern that a large volume of exits concentrated in a short window could pressure SpaceX's private market valuation.
Nvidia CEO Jensen Huang described data centers built on smuggled chips as a 'dead end' this week — a carefully chosen phrase that doubles as both a technical argument about supply chain integrity and a message to customers in restricted markets that long-term risk from contraband hardware is prohibitive. Export controls on advanced AI chips remain one of the primary mechanisms the U.S. is using to maintain computational advantage over China, and Huang's comment amounts to an endorsement of that policy framework from a company whose revenues are directly shaped by it.
The talent story most worth watching may be the exodus from Google DeepMind. Bloomberg reports that two more senior researchers — Jonas Adler and Alexander Pritzel — are set to leave for Anthropic, following earlier departures that reportedly included a Nobel laureate and the co-lead of the Gemini project. DeepMind has historically been Google's crown jewel for fundamental AI research, the organization behind AlphaFold and AlphaGo. Losing multiple senior researchers in rapid succession is a signal about how researchers assess relative intellectual freedom, resources, and mission clarity across organizations — and right now Anthropic appears to be winning that assessment.
A report that the White House has sidelined Anthropic CEO Dario Amodei from high-stakes export control negotiations — working instead with co-founder Tom Brown — suggests trust or relationship dynamics at the government interface that go beyond technical substance. On the product side, Google added a screen-control capability to Gemini 3.5 Flash, its fast and cost-efficient model tier. Screen control — the ability to observe and interact with a computer display rather than just respond to text — is a key requirement for genuine autonomous agent tasks. That this capability is now available in the lightweight, cheap tier means agentic AI is moving down-market quickly.
A Nature peer-review critique questioning Microsoft's Majorana quantum chip claims rounds out the technology picture. Microsoft staked a significant portion of its quantum computing roadmap on the Majorana approach — a type of topological qubit theoretically more stable than competing designs. A challenge to the underlying experimental evidence in one of the world's most rigorous scientific journals is not merely a PR problem; it represents a potential years-long setback for a multi-billion-dollar research program.