OpenAI released GPT-5.5 (codename "Spud") to paid ChatGPT and Codex subscribers today, calling it "a new class of intelligence for real work." The model excels at multi-step agentic tasks — planning, tool use, and self-checking — and narrowly beats Anthropic's Claude Mythos Preview on Terminal-Bench 2.0. It carries OpenAI's internal "High" cyber-risk classification but does not cross the "Critical" threshold; API access is delayed pending additional safeguards.
Google unveiled its eighth-gen TPU as a split architecture: TPU 8t for training (9,600-chip superpods, 3x Ironwood performance) and TPU 8i for inference (1,152-chip pods, 80% better perf/dollar), both targeting TSMC 2nm for late 2027. Sundar Pichai disclosed that 75% of all new code at Google is now AI-generated and engineer-approved, up from 50% last fall. Gemini Enterprise, a full agent platform with Agent Designer and long-running process agents, saw 40% paid-user growth in Q1, and Google's models now process 16 billion tokens per minute via API.
A memo from White House OSTP Director Michael Kratsios alleges that Chinese entities are running systematic campaigns to distill U.S. frontier AI models, using tens of thousands of proxy accounts and jailbreaking techniques to exfiltrate proprietary capabilities. The memo, distributed across government agencies, directs them to share threat intelligence with AI labs. Beijing rejected the accusations as "baseless," but the timing could complicate a planned US-China summit next month.
Meta informed employees today that it will lay off roughly 8,000 workers starting May 20, with additional cuts planned for H2 2026. Another 6,000 open roles will go unfilled. The cuts hit Reality Labs, Facebook, recruiting, sales, and global operations. The restructuring is explicitly tied to Meta's $135B AI investment push, continuing a pattern of trimming non-AI headcount to fund frontier model development and infrastructure.
SpaceX preempted Cursor's in-progress $2B fundraise at a $50B+ valuation with a deal that gives SpaceX the option to acquire the AI coding startup for $60B later this year — or pay a $10B collaboration fee if it walks away. The acquisition is being deferred until after SpaceX's summer IPO to avoid complicating financial filings. It would be the largest AI acquisition in history and signals Elon Musk's intent to build an AI empire spanning xAI, Grok, and now developer tools.
Anthropic and Amazon finalized a 10-year, $100B+ AWS infrastructure deal securing up to 5 gigawatts of Trainium capacity, with Amazon investing an additional $5B immediately and up to $25B more tied to milestones. Separately, at Google Cloud Next, Anthropic expanded its TPU commitment to approximately 3.5 GW starting in 2027, making it the anchor customer for Google's next-gen chips. Anthropic is now running a three-lane compute strategy across AWS Trainium, Google TPUs, and Nvidia GPUs.
Thinking Machines Lab, the 14-month-old startup founded by former OpenAI CTO Mira Murati, signed a deal valued in the single-digit billions for Google Cloud AI infrastructure including Nvidia's latest GB300 chips. The startup's product, Tinker, automates creation of custom frontier models using reinforcement learning — a compute-intensive approach that justified the massive infrastructure commitment. This makes Thinking Machines the third frontier AI developer to ink major Google Cloud capacity deals this month, behind Anthropic and Meta.
A PwC survey of 1,217 senior executives across 25 sectors finds that the top 20% of AI-adopting companies generate 7.2x more AI-driven revenue and efficiency gains than the average competitor. The differentiator isn't just deployment — leaders use AI for business reinvention and new revenue streams, not just productivity. The study warns the gap between AI leaders and laggards will continue to widen as leading firms learn faster and scale proven use cases.
The Connecticut Senate voted 32-4 to pass Senate Bill 5, one of the most comprehensive U.S. state-level AI bills to date. It regulates developers of frontier AI models, creates a state AI sandbox for testing, requires AI chatbot operators to detect and respond to self-harm indicators, and mandates employee notification when AI is used in hiring decisions. The bill now heads to the House.
xAI's Grok chatbot has been throwing "high demand" errors for more than 48 hours, locking out both free users and paid SuperGrok subscribers. The outage comes at an awkward time for xAI, which is simultaneously trying to close the Cursor acquisition and demonstrate enterprise reliability. No timeline for resolution has been provided.
