Google Data Infrastructure
The Administration Wants to Ban 'AI' — While Google Shows Off What It Can Do
President Trump posted on Truth Social this week reiterating his September executive order directing that the term 'artificial intelligence' be replaced with 'super intelligence' in all relevant government documents. The practical administrative burden is manageable — contractors and agencies will adopt new vocabulary — but the symbolic and interoperability consequences are more significant. 'Artificial intelligence' carries decades of history in international standards bodies, treaty frameworks under active negotiation, and regulatory regimes in the EU, UK, and Asia. A unilateral U.S. renaming creates immediate friction in any intergovernmental coordination context. The chosen replacement term also carries a specific meaning in AI safety research — referring to a hypothetical system that exceeds human intelligence across all domains — and using it as a generic substitute collapses that distinction in ways that some safety researchers find troubling and some technology accelerationists find welcome.
Meanwhile, Google gave a substantial public showcase of Gemini 4 Argon, its latest AI model, now available to paying app users, Cloud enterprise customers, and developers through the API. The demonstration featured a code-rendered interactive 3D library and an automated analysis of NASA Mars data that produced presentation-ready slides — capabilities that advance Google's position in what analysts describe as applied professional workflow tasks, not just text generation.
OpenAI, for its part, has reportedly been in talks to purchase Google's tensor processing unit chips outright rather than renting access to them. OpenAI's compute stack is currently heavily dependent on Nvidia hardware through its Microsoft relationship; owning Google TPUs would represent meaningful supply chain diversification and create an unusual commercial arrangement in which two of the most direct AI competitors would also have a hardware supply relationship.
Underlying all of it is a physical infrastructure problem that Ross Perot Jr. flagged publicly this week: the United States is running out of large-format warehouse space suitable for AI data center conversion. Optimized AI training clusters require not just square footage but specific power density, cooling infrastructure, fiber connectivity, and physical security — none of which standard commercial warehouses provide without major capital investment. The International Data Corporation estimated earlier this year that AI-optimized data center capacity needs to expand roughly 40% by 2028 to meet projected demand, and that kind of expansion scales on construction and utility permitting timelines measured in years, not software release cycles. Separately, Elon Musk pushed back on Verizon's public statements about SpaceX's terrestrial wireless network, saying a full consumer deployment is two to three years away — a more specific timeline than Musk typically offers, though one contingent on FCC spectrum allocation decisions still pending.