AI News Roundup: GTC Preview, Perplexity’s Personal Computer, States Rush to Pass AI Laws

NVIDIA Promises a “World-Surprising” Chip at GTC 2026

NVIDIA’s GPU Technology Conference kicks off Monday in San Jose, and CEO Jensen Huang is promising a chip that will “surprise the world.” Analysts expect the keynote to formalize the transition to the Vera Rubin platform — which entered full production earlier this year and delivers up to a 10x reduction in inference token cost over Blackwell — while offering the first architectural details on Feynman, next-generation silicon purpose-built for agentic AI reasoning and long-term memory.

Microsoft, Meta, AWS, Google Cloud, and Oracle Cloud will be among the first to deploy Vera Rubin–based instances in the second half of 2026. Microsoft has confirmed that its Azure “Fairwater” AI superfactories were engineered years in advance to accommodate Rubin’s power, thermal, and networking requirements. With more than 1,000 sessions across March 16–19, GTC remains the industry’s marquee hardware event.

Perplexity Launches “Personal Computer” — a 24/7 AI Agent on Mac mini

At its inaugural “Ask 2026” developer conference, Perplexity unveiled Personal Computer, an always-on AI agent that runs locally on a dedicated Mac mini with full access to files and apps. Instead of typing precise commands, users describe broad goals — like “create a podcast about whales” — and the system figures out which local apps to open, which files to use, and how to get it done.

Every sensitive action requires user approval, sessions generate full audit trails, and there’s a kill switch for emergencies. Access is limited to Perplexity Max subscribers at $200 per month, with a waitlist now open. The product positions Perplexity squarely against OpenAI’s Operator and Anthropic’s Claude Computer Use in the race to build persistent, agentic desktop assistants.

States Race to Pass AI Laws Before Session Deadlines

With legislative sessions wrapping up across the country, several states pushed through significant AI bills this week. Washington gave final passage to HB 2225, a chatbot safety bill targeting AI companion apps, and sent HB 1170 — an AI transparency measure requiring consumer notification — to the governor’s desk on March 11.

Virginia passed three AI bills ahead of its Saturday adjournment, including HB 580 on AI fraud and abuse and HB 797 establishing a framework for independent AI verification. Utah closed its session after passing bills limiting AI use in schools, protecting people from deepfakes, and requiring that medical decisions remain in human hands. The flurry of state-level action underscores how far ahead of Congress the states are moving on AI governance.

Anthropic–Pentagon Fight Continues to Reshape the Industry

The fallout from Anthropic’s lawsuit against the Department of Defense — now in its fifth day — continued to send shockwaves through the AI industry. On Wednesday, Anthropic filed for an emergency stay with the D.C. Circuit Court of Appeals, warning that the Pentagon’s “supply-chain risk” label could cost it billions in lost revenue. Pentagon CTO Emil Michael fired back, claiming Claude would “pollute” the defense supply chain because it has “a different policy preference” baked in.

The dispute has become a litmus test for the AI industry’s relationship with government. More than 30 OpenAI and Google DeepMind employees filed a statement supporting Anthropic in their personal capacities. OpenAI robotics lead Caitlin Kalinowski resigned last week over OpenAI’s own Pentagon contract, saying “surveillance of Americans without judicial oversight and lethal autonomy without human authorization are lines that deserved more deliberation.” Meanwhile, Google has been quietly expanding its Pentagon AI contracts, including a deal to provide AI agents to the DOD’s 3-million-person workforce.

Meta Delays “Avocado” AI Model, Explores Licensing Gemini

Meta’s latest flagship AI model, code-named Avocado, has been pushed from March to May after internal benchmarks showed it lagging behind Google’s Gemini 3.0 and the latest models from OpenAI and Anthropic. In a surprising twist, Meta leadership has reportedly discussed temporarily licensing Gemini from Google to bridge the gap until Avocado is ready.

The delay comes just days after Meta closed its acquisition of Moltbook, the Reddit-like social network for AI bots that went viral in January. Moltbook CEO Matt Schlicht and COO Ben Parr will join Meta’s Superintelligence Labs on March 16, as the company pivots its AI agent strategy from model performance toward social infrastructure for autonomous agents.

Google AI Mode Hits 75 Million Daily Users

Google’s AI Mode in Search now reaches more than 75 million daily active users globally following a burst of March expansions. On March 4, the company rolled out Canvas — a workspace inside Search for drafting documents, writing code, and building structured projects — to all U.S. users without requiring a Google Labs opt-in. Google also expanded AI Mode to 53 new languages, including 13 African languages.

The moves cement Google’s strategy of embedding AI deeply into its dominant search product rather than building a standalone chatbot competitor. With Canvas, Google Search is evolving from an answer engine into a full productivity platform.