One-Line Summary

Google shifts from chatbot demos to always-on Gemini agents embedded in its core apps, while lowering paywalls and expanding TPU supply to turn AI into everyday workflow.

Big Tech

Google puts Gemini agents across Search, YouTube, and Workspace

Google is turning Gemini from a chatbot into a hands-on helper inside products you already use, like Search, YouTube, and Docs. Sundar Pichai highlighted Ask YouTube for conversational answers that jump to the right video moment, voice-powered Docs Live to capture ideas and draft by speaking, AI Overviews with over 2.5 billion monthly users, AI Mode in Search above 1 billion, and the Gemini app at 900 million monthly active users. ¹

Behind the scenes, Google says the scale of AI usage has exploded: 3.2 quadrillion tokens processed per month across its surfaces, about 19 billion tokens per minute via its model APIs, with 8.5 million developers building monthly and 375 Google Cloud customers each processing over 1 trillion tokens in the past year. ¹

Google also introduces two model lines. Gemini Omni (starting with Omni Flash) generates video from text, images, video, and audio, and rolls out in the Gemini app, Google Flow, and YouTube Shorts beginning May 19, 2026. Gemini 3.5 Flash emphasizes speed and agentic coding; starting May 19, it becomes the default model for the Gemini app and AI Mode in Search. ²

To sustain this push, Google points to infrastructure and trust. It expects 2026 capital expenditures of roughly $180–$190 billion, and details TPU 8t/8i with higher efficiency and the ability to scale training across more than 1 million TPUs. For content provenance, SynthID watermarks expand to Chrome and Search, with new partners including OpenAI, Kakao, and Eleven Labs. ¹

Industry & Biz

Blackstone commits $5B to a TPU-focused venture with Google

Blackstone is investing $5 billion in equity to form a U.S.-based AI infrastructure company with Google that offers data center capacity and Google’s Tensor Processing Units (TPUs) as a service, targeting the first 500MW online by 2027. CNBC reports the venture will be led by Benjamin Treynor Sloss, formerly Google’s chief programs officer. ³

Business Insider notes the move gives enterprises another route to access compute beyond Nvidia-centric “neocloud” providers, with Google Cloud CEO Thomas Kurian saying the venture provides more options to reach specialized AI hardware. ⁴

Anthropic acquires SDK-generator Stainless used by rivals

Anthropic is buying Stainless, a developer-tools startup whose software auto-generates SDKs from API specs and is used by OpenAI, Google, and Cloudflare. Terms were not disclosed; The Information previously reported talks above $300 million, per Yahoo Finance. ⁵

Anthropic says it will wind down hosted Stainless products while customers retain rights to their generated SDKs. Analysis from ResultSense frames the deal as vertical integration into the agent-connectivity layer, complementing Anthropic’s Model Context Protocol to make Claude-based agents easier to wire into enterprise systems. ⁶

Intel and Qualcomm show interest in Tenstorrent, Bloomberg reports

Bloomberg reports Intel and Qualcomm have held early talks about acquiring Tenstorrent, an AI chip startup that designs accelerators it says are efficient for certain workloads. The interest comes amid renewed momentum for alternatives to Nvidia and AMD. ⁷

Bloomberg characterizes the discussions as early; any deal would signal further consolidation pressure across AI compute suppliers as demand remains intense. ⁷

New Tools

Google cuts AI Ultra pricing and adds a $100 tier

Google introduces a $100/month AI Ultra plan and reduces the top-tier price to $200/month, adding 5× higher usage than Pro for the $100 tier, 20TB storage, and access to Gemini 3.5 Flash and agent-first tools like Antigravity; Gemini Spark access is included on Ultra (U.S.-only) with testing beginning in late May. ⁸

All paid AI plans get access to the latest models, including Gemini Omni for video creation and editing and Gemini 3.5 Flash for fast agents and coding. Engadget notes the Ultra tier’s storage is now 20TB, down from 30TB previously. ⁹

Google is moving from daily prompt caps to a compute-based usage model in the Gemini app, refreshing every five hours until a weekly cap, with optional top-up AI credits for Pro and Ultra across products like Antigravity and Google Flow. ⁸

Community Pulse

Hacker News (176↑) — Excited by slick demos and integrations, but uneasy that conferences feel all-AI-all-the-time. ¹⁰

"There was genuinely a moment in this presentation when I get kept saying how it was incredible that "Gemini put together a todo list for me"" — Hacker News ¹⁰

Hacker News (341↑) — Mixed reactions to Google’s AI subscriptions: concerns about reliability and citations, with some users eyeing alternatives. ¹¹

"It somehow seems to interpret whatever sources it's grepping as the exact opposite of what those sources say fairly often. I've lost track of how many times I've clicked on the sources it cites, and every single one is in agreement, but the AI claims the opposite." — Hacker News ¹¹

"I started switching to DDG on some devices, this will motivate me to finish the transition ! Thanks" — Hacker News ¹¹

Hacker News (1,136↑) — Optimism around Anthropic’s team and culture, with high expectations for recent leadership additions. ¹²

"Come on, Anthropic ARE the good guys if there are any. Certainly the incentives of trillions will do what money does, but they have assembled an incredibly altruistic and philosophically-minded crew. I’m rooting for them and trying real hard not to get cynical." — Hacker News ¹²

"Karpathy embedded within an organization is way more impressive than him out on his own with hot takes and little projects. I hope he does great things for Anthropic." — Hacker News ¹²

What This Means for You

Google’s shift to embedded, conversational AI means fewer clicks to get answers or draft content: Ask YouTube jumps to the right clip, while Gmail Live and Docs Live aim to pull data and draft by voice. If you research, summarize, or plan in your job, these features compress time-to-answer and reduce context switching. ²

For heavier users, the new $100 AI Ultra tier offers more headroom and access to faster models; the move to compute-based limits in the Gemini app means your cap depends on task complexity, not just prompt count. Budget how you use image/video inputs vs text-only tasks to stretch your allowance. ⁸

Trust and attribution matter more as AI output proliferates. Google is expanding SynthID/C2PA checks to Search and Chrome, so you can verify if an image was AI-generated or edited before you reuse it in decks, websites, or ads. Make “check provenance” a standard step in your content workflow. ²

If your team needs compute for pilots, Blackstone’s $5B TPU venture signals more pathways to specialized hardware. Without committing, you can scope inquiries with vendors on TPU access for inference-heavy workloads and compare cost/performance against your current GPU stack. ³

Action Items

1. Test Gemini 3.5 Flash on a real task: Use the Gemini app or AI Mode in Search to draft an email, summarize a document, or outline a campaign brief, and compare speed vs your current assistant.
2. Compare Google AI plans: If you routinely hit limits, try the $100 AI Ultra tier for a week to see if the higher usage and model access improve your workflow.
3. Verify image provenance before reuse: Use Google Search’s image tools (e.g., Lens) to check Content Credentials/SynthID on any image you plan to publish.
4. Prompt-build a tiny Android app in AI Studio: Generate a simple native app (e.g., a habit tracker), preview it in the emulator, and export to GitHub to assess quality.
5. If you rely on Stainless-generated SDKs, back them up: Export and store your generated SDKs and plan for maintenance now that Anthropic is winding down hosted Stainless products.