Google moving up market and join premium tier club
This week’s signal: Google launch capability parity with frontier labs. Adding AI Ultra at $100/month brings all three labs now at the same premium price point. Anthropic projected its first operating profit on $10.9B in quarterly revenue. AI value is real, and labs can no longer afford the real compute cost with discount. Every major lab will now pricing as the value is proven. With capability gaps closes, access to compute will separate the winers.
AI Feature Releases
Google I/O
Gemini 3.5 Flash — new flagship; claims 4× faster than competing frontier models at half the cost; live on Vertex AI, API, and AI Studio. Google
Gemini Omni — any-to-any generative model launching with video output; conversational multi-turn editing with character consistency and physics; live in Gemini app and YouTube Shorts.
Gemini Spark — persistent 24/7 cloud-based autonomous agent; proactive background execution across Google’s ecosystem; rolling out to AI Ultra subscribers.
Antigravity 2.0 — agent-first developer platform refresh; demoed building a full OS on stage in 12 hours.
AI Ultra at $100/month. All Gemini models, Spark, 30TB storage, and Workspace; same price as Claude Max and ChatGPT Pro. Google
Microsoft Copilot — Outlook email grounding + GPT-5.5 Thinking. Email threads now pull directly into M365 Copilot Chat context; GPT-5.5 Instant and GPT-5.5 Thinking available across all devices. Microsoft Learn
OpenAI — Guaranteed Capacity. Enterprise compute reservation program, 1–3 year terms with tiered discounts; Altman: “the world will be capacity-constrained for some time.” OpenAI
Cursor — Three releases this week Cursor:
Composer 2.5 — new long-running task model; better instruction-following on complex multi-step work; $0.50/M input, $2.50/M output; faster tier at $3/$15; double usage first week.
Cursor in Jira — assign tickets to Cursor or @mention in comments; cloud agent scopes task from ticket context, opens a PR on completion.
Cursor 3.5 — Automations in Agents Window — automations now run inside the Agents Window with multi-repo support; agents reason across codebases end-to-end; 5 new no-repo automation templates in Marketplace.
Venture & Market
Anthropic — $10.9B Q2 revenue, first operating profit $559M. Up 130% from $4.8B in Q1; profitability may not sustain full year due to $1.25B/month SpaceX compute contract. WSJ, CNBC
SpaceX S-1 filed publicly. Discloses $1.25B/month Anthropic compute deal through May 2029; $18.67B 2025 revenue post-xAI acquisition; targeting $1.75–2T valuation. CNBC, Reuters
Andrej Karpathy joins Anthropic. OpenAI co-founder joins pretraining team; will lead a new team using Claude to accelerate pretraining research. TechCrunch, Axios
Anthropic — acquires Stainless ~$300M. SDK and MCP server tooling startup that previously served OpenAI, Google DeepMind, and Cloudflare; Anthropic now controls the MCP connector layer. Anthropic, TechCrunch
Anthropic × KPMG — global alliance. Claude Cowork + Managed Agents embedded in Digital Gateway across 138 countries; 276,000 employees get Claude; exclusive Claude-powered tax offering; KPMG named preferred Anthropic PE partner. Anthropic, KPMG
Cloudflare — Project Glasswing findings. Cloudflare tested Anthropic’s unreleased Mythos Preview across 50+ internal repos; model chains low-severity bugs into working exploit proofs-of-concept autonomously — something prior frontier models couldn’t do. Anthropic has no plans for public release. Cloudflare Blog, The Decoder
Extended Reading
Serval’s CEO Jake Stauch- Rebuilding IT From the Ground Up for the AI Age: Sequoia Training Data podcast (May 2026). Pat Grady interviews Stauch on why the right answer looks like boring enterprise software with unlimited intelligence, why new model releases sometimes have to be rolled back when prompt tuning breaks, and why “fewer, better” people is the only durable moat when your product may need to be rebuilt every six months.