Overview
- The Information reports OpenAI has paused its self‑built Stargate data centers in favor of deep partnerships, planning a 4.5 GW build with Oracle and a 1 GW Texas campus with SoftBank while retaining cluster architecture control and hiring former Intel CTO Sachin Katti to lead infrastructure.
- A Reuters source says OpenAI is targeting roughly $600 billion in cumulative compute through 2030 and booked about $13 billion in 2025 revenue, as The Information notes inference costs quadrupled last year and margins fell to 33%, with CFO Sarah Friar separately saying annualized revenue topped $20 billion.
- A U.S. Defense Department official confirmed an agreement allowing xAI’s Grok in classified systems, as the Pentagon pressures Anthropic to relax usage limits and accelerates talks with Google and OpenAI under an “any lawful use” standard that OpenAI has not committed to.
- An MIT study presented at AAAI finds leading chatbots perform worse for users with lower formal education or English proficiency, showing higher refusal rates, condescending tone in some replies, and selective withholding of correct information based on user background.
- OpenAI COO Brad Lightcap says enterprise‑grade AI has yet to permeate complex business processes, and Goldman Sachs’ Jan Hatzius assesses that recent AI hardware spending contributes almost nothing to 2025 U.S. GDP because much of the equipment is imported.