*Today's digest draws from a backlog of AI and technology news accumulated over the past several days, as no new newsletters arrived on May 25.*
---
Google I/O 2026: Gemini Goes Everywhere
Google's annual developer conference this year amounted to an all-in declaration on the agentic AI era. The company announced what it's calling the biggest change to search in 25 years — an expanded AI-powered interface powered by Gemini, which now functions as the common layer across all Google services rather than a standalone product.
Three new Gemini-branded offerings stole the show. Gemini Omni Flash generates 10-second video clips from conversational prompts, supports personal avatar creation, and embeds SynthID watermarks for provenance tracking. Gemini Spark is positioned as a personal AI agent, integrating with Gmail, Calendar, Google Docs, and more than 30 third-party tools to proactively draft updates and surface relevant context. And Antigravity 2.0 is Google's infrastructure layer for orchestrating multi-agent workflows — the company demonstrated building a rudimentary operating system using 96 parallel agents working in concert.
Google also released Agent Executor, an open-source runtime standard designed to make long-running agent workflows more reliable and portable across systems. Google is increasingly positioning itself not just as an AI model provider but as the infrastructure backbone for the emerging agentic web. The Chrome team backed that framing with 15 browser updates oriented around "agentic web" principles, including WebMCP support and new DevTools built for agents.
---
Anthropic: A Profitable Quarter, a Famous Hire, and an Unusual Compute Deal
It's been a landmark stretch for Anthropic. The company projects its Q2 revenue will hit $10.9 billion, putting it on track for its first operating profit — a significant milestone for a company that has spent heavily on safety research and infrastructure.
On the talent side, Andrej Karpathy, the OpenAI co-founder who had departed to build his own education startup, has joined Anthropic to lead a new pre-training research team. Karpathy is one of the most respected figures in the field, and his move is widely seen as a signal of Anthropic's deepening ambitions in foundational model research.
The company also made a striking infrastructure move: it's now spending $1.25 billion per month to rent 300 megawatts of compute from xAI's Colossus facility — a competitor's infrastructure — to address its own capacity shortages. The scale of that arrangement surprised observers, though early disclosures in SpaceX's S-1 suggest Anthropic's contract effectively recoups capital expenditure for xAI in under a month, illuminating just how lucrative compute leasing has become.
Separately, Anthropic acquired Stainless, the SDK infrastructure company, for more than $300 million. Stainless had previously served competitors, giving Anthropic meaningful leverage over a layer of developer tooling that other AI companies depend on.
---
OpenAI: IPOs, Math, and Your Money
A federal jury ruled against Elon Musk in his lawsuit against OpenAI, clearing a significant legal obstacle as the company prepares for what it expects to be a trillion-dollar IPO. Reports indicate OpenAI is preparing a confidential S-1 for a September 2026 listing, while SpaceX — which has its own compute relationship with Anthropic — filed its own S-1 seeking a valuation above $1.75 trillion, attributing 93% of its total addressable market to AI infrastructure.
OpenAI made headlines on the research front as well: one of its models disproved the 80-year-old Erdős unit distance conjecture in geometry, a result that, if verified, would represent one of the clearest examples yet of AI generating genuinely novel mathematics rather than pattern-matching from existing proofs.
On the product side, OpenAI shipped Codex integration into the ChatGPT mobile app, allowing users to trigger desktop-based coding tasks remotely. ChatGPT also gained a personal finance feature that connects to more than 12,000 financial institutions. Meanwhile, ChatGPT integration landed natively in Microsoft PowerPoint, enabling users to create and reorganize presentations through natural language.
---
Chips, Hardware, and the Race for Compute
NVIDIA reported a record quarter — $81.6 billion in revenue, driven overwhelmingly by its Data Center segment. CEO Jensen Huang also announced the Vera CPU, a new chip purpose-built for agentic workloads, representing a $200 billion market claim. NVIDIA is also deepening its footprint in Singapore, opening a robotics and AI infrastructure research hub.
Not to be outdone, Alibaba unveiled the Zhenwu M890 AI chip, which it claims delivers triple the performance of its predecessor, while AMD committed $10 billion to develop 2-nanometer high-performance computing chips in Taiwan — a direct challenge to NVIDIA's dominance in the AI accelerator market.
---
Robotics: From Warehouses to Living Rooms to Space
Figure AI's F.03 humanoid robot completed 200 hours of continuous autonomous operation while sorting packages without a single failure — a benchmark that signals a meaningful maturation in robotic reliability. A San Francisco startup called Gatsby is now piloting an on-demand home cleaning service using a Unitree G1 humanoid robot. ETH Zürich unveiled HELIOS, a four-armed humanoid designed for zero-gravity assembly work in space. And Chinese EV maker XPENG has begun mass-producing L4 autonomous robotaxis equipped with proprietary Turing AI chips.
---
AI and Labor: The Reckoning Continues
The human side of the AI transition is growing harder to ignore. Meta announced 8,000 job cuts to redirect workers toward AI-focused teams; Intuit is cutting 3,000 jobs for similar reasons. California Governor Gavin Newsom signed an order directing state agencies to study subsidies and retraining programs in anticipation of broader AI-driven displacement. Meanwhile, student protests at multiple commencements — including one where former Google CEO Eric Schmidt was heckled — have drawn attention to youth unemployment rates of 30–40% among recent graduates who attribute the trend to AI. Nearly half of Gen Z workers surveyed now say they believe AI's risks outweigh its benefits.
---
Quick Takes
Security: A Google report found LLMs are actively enabling industrial-scale cyberattacks by identifying unknown vulnerabilities and generating "morphing malware." Separately, attackers accessed 3,800 GitHub repositories via a trojanized VS Code extension. 1Password partnered with OpenAI to provide just-in-time credential access for AI coding agents.
Creative AI: Spotify partnered with Universal Music Group to let Premium users create paid AI-generated covers and remixes. A 95-minute feature film titled "Hell Grind," generated entirely by AI, premiered at Cannes. Stable Audio 3.0 released an open-weight model capable of generating six-minute music tracks.
China: Alibaba's Qwen3.7-Max model features a 1-million-token context window and reportedly outperforms Claude Opus-4.6 Max on math benchmarks. DeepSeek is building a free, open-source rival to Claude Code and OpenAI's Codex. Chinese open-weight models now account for 61% of top usage on OpenRouter, at roughly 9x lower cost than premium US models for bulk agent work.
Policy: The Trump administration delayed an executive order that would have allowed federal agencies to evaluate advanced AI models before release. The White House approved $9 billion in classified funding for AI infrastructure at spy agencies, earmarked for Nvidia Blackwell chips.
Miscellaneous: Fitbit's 5.0 includes a Gemini-powered AI coach for Premium users. Figma added an AI agent that can generate and bulk-edit designs on the canvas. Jeff Bezos described his $38 billion startup Prometheus as focused on building an "artificial general engineer" and next-generation CAD tools. Apple is preparing a significant Siri overhaul ahead of WWDC.