Google I/O breakdown: impressive technology, unfocused products, and VO3's viral potential

Summary

Google I/O demonstrated impressive technological capability across generative AI models, but the company failed to articulate a coherent product strategy. The keynote was so densely packed—The Verge's recap ran 30 minutes compared to 5 minutes for Apple events—that individual breakthroughs got lost in the noise.

Technology advances

Google's underlying progress is real. Gemini 2.5, the Image-to-4-image model, and Video 3 represent significant leaps. Video 3 shows algorithmic superiority over OpenAI's Sora in controlled tests. A yellow Ferrari stays yellow throughout the video, and text overlays remain properly tracked. Google appears to have solved something important around maintaining state and coherence in generated video.

Google also holds structural advantages competitors cannot match. The company invented the transformer, trains and runs inference on its own TPUs (increasingly co-developed alongside models), and owns YouTube—a constant stream of video training data that no amount of dealmaking with Hollywood studios can replicate for rivals.

The product gap

None of this translates into compelling generative AI products. Ben Thompson identified the core issue: Google dominates the pillars of generative AI—algorithms, compute, and data—but cannot make compelling products. The keynote felt unfocused and random, with demos that looked impressive but lacked a connecting narrative.

Project Mariner, an AI browser extension that automates keystrokes and mouse clicks, exemplifies the gap. It exists and works, but Google didn't explain why users should care or how it fits into strategy. Deep Think, a reasoning-focused version of Gemini, is technically interesting but arrived without context. Notebook LM and other announced tools read like experiments, not products.

Google Flow proved the point experimentally. On iOS, the product rejected users and directed them to use a desktop browser instead. For a company of Google's scale, this is unacceptable. The forced funneling toward pre-made stock videos instead of custom generation, combined with opaque rate-limiting on Video 3, suggests Google is treating these as expensive research previews rather than products meant to be used.

Search as coherent strategy

Google's search strategy shows actual coherence. AI Overviews have already reached 1.5 billion monthly active users across 200+ countries, driving higher satisfaction and increased search frequency. Google is now introducing AI Mode, a chat-like search interface powered by Gemini 2.5. The product logic is clear: Google tests ideas at scale in AI Mode, and features that prove useful graduate to the main search experience.

Liz Reed, head of search, stated that the goal is to refine AI Mode features until they can graduate to core search. The funnel runs from models and infrastructure at the foundation, through experimental ideas in AI Mode, to proven features in search. It is the only place at Google I/O where the company articulated a clear progression.

The real competitive tension for Google is not whether it launches a ChatGPT killer. Google has repeatedly failed at direct competitive products. The tension is whether OpenAI's ChatGPT becomes good enough at web search that users stop going to Google for factual queries altogether. ChatGPT can now search the web, but it is slower than Google search for simple lookups. Speed and habit still favor Google's search box, but if ChatGPT closes that gap while offering a unified interface for both search and reasoning tasks, the calculus shifts.

Video 3 monetization

Google announced a $250-per-month tier (ramping to $500) called Google AI Ultra for higher compute limits and early access to experimental tools like Video 3 and Deep Think. Yet the company rate-limits Video 3 for subscribers and advertises it as a research preview despite charging for it. This is contradictory. Google knows the cost to generate an 8-second video, yet offers no transparent pricing model. An auction-based system where users pay variable rates based on current GPU utilization, like Google Ads, would better match demand to supply.

The viral potential of Video 3 also remains unrealized. Early content mimicked Studio Ghibli animation, but that became a meme rather than a tool. Breakthrough moments—where AI-generated content goes viral because it is genuinely good—have not emerged yet. A man-on-the-street interview created with Video 3 that circulated without attribution could create that moment, but Google has not positioned the product or community to discover that use case.

Takeaway

Google's technology is world-class. Its product strategy is scattered. Search gives the company a path forward. It is the only product with a clear funnel, user scale, and monetization logic. Everything else risks becoming another forgotten launch in Google's history of sunsetted products. The real question for investors is whether Google can execute the search strategy faster than OpenAI can build a search box that also reasons.