It was a busy week for AI — and that’s an understatement. Google I/O dominated the headlines with a wave of product launches, Anthropic locked in a landmark compute deal, and a federal jury handed Elon Musk a decisive courtroom loss. Here’s everything that matters.
—
## Google I/O 2026: Gemini 3.5 Flash, Gemini Omni, and Gemini Spark
Google’s developer conference delivered the biggest AI product day of the year. Gemini 3.5 Flash is available now across all Google products and APIs, claiming to be 4x faster than rival frontier models and optimized for agentic coding and long-horizon tasks. Gemini Omni adds true multimodal generation from any input type including video, rolling out to paid subscribers now. The headline announcement was Gemini Spark — Google’s answer to OpenAI Operator — a 24/7 personal AI agent that runs on Google Cloud VMs and can autonomously work across Workspace, third-party apps, and the open web.
—
## Google Launches $100/Month AI Ultra Tier
Alongside the product launches, Google announced AI Ultra — a $100/month subscription tier (down from a previous $250 price point) that includes 5x higher Gemini usage limits, 20 TB of cloud storage, YouTube Premium, and early access to Gemini Spark. It’s a clear move to compete with OpenAI’s Pro tier and signals that the subscription AI market is settling toward the $100/month price point for power users.
—
## Elon Musk Loses Lawsuit Against OpenAI
A federal jury in Oakland took less than two hours to unanimously dismiss all of Elon Musk’s claims against OpenAI, Sam Altman, and Greg Brockman. The jury found the case time-barred under the three-year statute of limitations — Musk’s allegations that OpenAI “stole a charity” by transitioning to a for-profit structure never made it to the merits. Musk’s legal team announced plans to appeal.
—
## Anthropic Signs $15B/Year Compute Deal with SpaceX
Anthropic secured what may be the largest compute deal in AI history: exclusive access to SpaceX’s entire Colossus 1 supercomputer in Memphis — 220,000+ NVIDIA GPUs across 300 megawatts of capacity — at a reported cost of approximately $15 billion per year. The deal is designed to directly expand capacity for Claude Pro and Max subscribers. Anthropic is already scoping Colossus 2 and has expressed interest in space-based compute infrastructure with SpaceX.
—
## Cursor Ships Composer 2.5 at a Fraction of Frontier Costs
Cursor released Composer 2.5, built on Kimi K2.5 with significant post-training work. The model benchmarks near Claude Opus 4.7 on SWE-Bench — but prices at $0.50 per million input tokens, roughly 10x cheaper than comparable frontier models. For developers who run AI coding at scale, this is a meaningful cost story.
—
## Pope Leo XIV to Publish First Papal Encyclical on AI
On May 25, Pope Leo XIV will publish Magnifica humanitas, the first papal encyclical ever dedicated to artificial intelligence, framing AI as the defining social and moral challenge of the era. The Vatican has invited a notable lineup of speakers to present alongside the Pope — including Anthropic co-founder and interpretability research head Christopher Olah. The document is expected to carry significant weight in global policy and ethics conversations.
—
## EU AI Act Omnibus: Compliance Deadlines Extended
The EU Council and Parliament reached a provisional agreement on the AI Act Omnibus, extending compliance timelines for high-risk AI systems — Annex III systems now have until December 2027, Annex I until August 2028. The agreement also reduces compliance burdens for SMEs and expands enforcement powers for the AI Office. For companies building in Europe, this buys meaningful runway.
—
## OpenAI: Content Provenance Research, Personal Finance, and Codex Expansion
OpenAI had a quieter but productive week. The company published new research on content provenance for AI-generated media. ChatGPT Pro users in the US can now connect financial accounts for AI-powered money analysis. And Codex expanded to ChatGPT mobile while shipping a new Windows-native sandbox with tighter security controls — a nod to enterprise deployment requirements.
—
## Grok 4.3 Expands with Video Input and Workspace Integrations
xAI’s Grok 4.3 continued its broader rollout this week, adding native video input and a new Skills system that connects Grok to SharePoint, Outlook, Google Workspace, Notion, GitHub, Linear, and custom MCP servers. Older Grok 4 variants were officially retired on May 15. The workspace integrations put Grok in more direct competition with Microsoft Copilot for enterprise users.
—
A lot happened this week. The clearest signal: the compute arms race is accelerating, the AI subscription market is maturing, and the broader cultural and regulatory conversation around AI is only getting louder.
