Sunday, May 17, 2026

Top story

The Agent RaceGoogle DeepMind Blog

Introducing Gemini Omni

Google's Gemini Omni represents a significant capability milestone in multimodal AI, directly competing with OpenAI and Anthropic's reasoning models. This launch will force enterprise AI strategy reassessment across teams currently locked into competitor ecosystems.

Gemini Omni model release

The briefs

Eclipse Ventures' $2.5B Cerebras investment validates a contrarian thesis about specialized AI chips over general-purpose compute, reshaping how the industry thinks about infrastructure capex and competitive advantage in the AI race.

$2.5B investment by Eclipse in Cerebras

Chinese AI labs have achieved breakthrough capabilities in video generation that now exceed US competitors, reshaping the competitive landscape and forcing Western AI companies to recalibrate their model development strategies.

ByteDance and Kuaishou outperforming Western rivals in video generation

Google is expanding Gemini beyond consumer/enterprise chat into specialized scientific research tools. This signals a major shift toward vertical AI products and opens a new revenue stream in high-stakes discovery workflows.

Gemini for Science launched as dedicated product line

Google is expanding Project Genie (a generative world model) to mass market via Street View integration, enabling AI Ultra subscribers to simulate real-world environments. This is a significant product capability ship that combines two existing Google assets into a new use case—moving generative simulation from research into production.

Project Genie expanded to Google AI Ultra subscribers globally

Anthropic's Claude faces critical security vulnerabilities affecting enterprise deployment. As AI agents gain system access, architectural flaws in Claude Code and Chrome extensions create attack vectors for token theft and command execution—raising questions about production readiness of AI tools handling sensitive data.

Claude Code RCE flaw via malicious deeplinks

Samsung is developing high-bandwidth memory (HBM) packaging technology to enable on-device AI inference at scale. This represents a shift in smartphone architecture toward local AI processing rather than cloud dependency, with implications for latency, privacy, and device capability benchmarks.

Samsung developing new HBM packaging technology

This research exposes a critical gap between how confident AI systems claim to be and how confident users perceive them to be—a finding with direct implications for enterprise AI adoption, user safety, and risk assessment in production systems.

Study reveals user overestimation of AI confidence in responses

GPU pricing power is accelerating as AI infrastructure demand outpaces supply. When the fastest compute hits five-figure prices without demand destruction, it signals either a genuine bottleneck or a margin-expansion window that won't last.

NVIDIA RTX 6000 Blackwell pricing now exceeds $10,000 USD

Google is expanding Gemini's reasoning capabilities and ecosystem reach with 'Extended' thinking mode and new app integrations, positioning it as a more capable competitor to Claude and ChatGPT in enterprise and productivity workflows.

Gemini app rolling out selectable 'Extended' thinking level (in addition to standard mode)