AI Trends — 18 May 2026

Your daily briefing on the most important developments in artificial intelligence.

🔥 Top Story: Google I/O 2026 Is Tomorrow — Here's What to Expect

The AI world's biggest event of the spring lands Monday, May 19 at 10am PT — Google I/O 2026. After the Android Show on May 12 front-loaded platform announcements (Googlebooks, Gemini Intelligence for Android), the main keynote is expected to be model-heavy.

What's confirmed:

• Android XR glasses preview, with hardware partners Samsung, Warby Parker, Gentle Monster, and XREAL
• "Latest Gemini model updates" and "agentic coding" are on the published agenda
• Aluminium OS — Google's Android-based replacement for ChromeOS — is expected to launch with full consumer laptop hardware partnerships

What's widely expected:

• A flagship Gemini 4.0 release (or named variant), with improvements to multimodal reasoning, longer context windows for Workspace integrations, and stronger agentic reliability
• Expanded Google Cloud agent-building toolkit with Workspace integrations and enterprise pricing
• Shutdown of internal "Mariner" agent project confirmed — Google consolidating around Gemini-native agentic capabilities

US Treasury Secretary Bessent stated last week he expects a "big step-function jump" from upcoming Gemini releases — an unusual government-level signal of confidence in a specific product announcement.

Watch live: io.google and Google's YouTube channel.

💰 Anthropic Closing In on $900 Billion Valuation Round

Bloomberg confirmed that Anthropic's $30 billion fundraise — which would value the company at over $900 billion — is on track to close as soon as the end of May. No term sheet has been signed as of May 16, but the round is progressing.

If closed at this valuation, Anthropic would surpass OpenAI's $852 billion March valuation for the first time. The funding is earmarked for compute infrastructure to meet surging demand, primarily fulfilling Amazon and Google Cloud commitments coming online through 2027.

This follows a string of major business wins for Anthropic:

• Q1 2026 revenue grew 80x year-over-year, with ARR now above $44B
• PwC signed an all-in Claude partnership across its global advisory and audit operations
• Gates Foundation signed a strategic AI deal with Anthropic in mid-May
• Anthropic rented SpaceX's entire Colossus 1 supercomputer — 220,000+ NVIDIA GPUs, 300MW — for Claude training and inference

🤖 Meta Avocado: May Window Closing With No Launch

Reuters sources said in April that Meta's next flagship model — codenamed Avocado — would land "May or June." May is now two-thirds over with no announcement.

Internal test results reportedly show Avocado performing between Gemini 2.5 and Gemini 3.0, below the threshold needed to compete credibly with GPT-5.5 or Claude Opus 4.7 on developer benchmarks. With Google I/O tomorrow set to deliver a major Gemini release, Meta faces an awkward positioning problem:

• Announce before I/O → get buried in the pre-show noise
• Announce same week → invite direct head-to-head comparison
• Slip to June → lose May momentum entirely

A June launch is now the most likely outcome.

📱 OpenAI AI-First Device Rumors Return

Reports circulating this week point to growing internal signals that OpenAI is exploring an "AI-first device" that would eliminate traditional app interfaces entirely in favor of an always-on AI layer. OpenAI's confirmed hardware partnerships with MediaTek and Qualcomm — the same chip partners named in earlier smartphone rumors from May 1 — add plausibility.

No announcement has been made. OpenAI has not publicly confirmed any device project, but analyst attention is sharpening as the Google I/O hardware reveal approaches and the broader AI-native device category heats up.

⚖️ Regulation Roundup

The US Commerce Department's CAISI (Center for AI Safety and Innovation) now has pre-deployment AI evaluation agreements with all five major frontier labs: OpenAI, Anthropic, Google DeepMind, Microsoft, and xAI. The EU is in discussions with Anthropic about access to its Mythos restricted model but has not reached the same stage as its OpenAI agreement.

The UK AI Safety Institute published updated red-teaming guidance for frontier models this week, signaling coordinated international attention on capability evaluation ahead of the Google I/O release cycle.

📊 By the Numbers

🗓️ What to Watch This Week

• May 19 — Google I/O 2026 keynote, 10am PT. Gemini 4.0 and Aluminium OS expected.
• End of May — Anthropic $900B round expected to close
• June (likely) — Meta Avocado model launch
• Ongoing — OpenAI AI-first device speculation; EU-Anthropic Mythos negotiations

Sources: AIToolsRecap, Bloomberg, Reuters, Fortune, MIT Technology Review