Company@X

Signal of the Day

OpenAI has officially designated its upcoming model, Astra, as its first “critical” model under its Preparedness Framework, implementing strict development controls and pausing its general availability to ensure its advanced cyber capabilities can be deployed safely. While CEO Sam Altman maintained that keeping powerful models restricted to a select few is a bad strategy, the lab is delaying Astra’s release to build defensive safeguards.

Key Announcements

Whatnot · Source Whatnot has raised $545 million in Series G funding at a $20 billion valuation, solidifying its position as the dominant live-commerce platform across North America, the UK, and Europe. Only six months into 2026, the company’s Gross Merchandise Value (GMV) has already surpassed its entire 2025 total of over $8 billion, driven by weekly additions of over 650,000 users and a buyer base that has more than doubled over the past year. This massive round underscores the commercial dominance and explosive growth of live-streamed retail outside of Asian markets, where the format originally matured.

Google · Source Google showcased the capabilities of its newly introduced video-generation model family, Gemini Omni Flash, demonstrating how developers and creators are using it to switch camera perspectives, alter lighting settings, and animate hand-drawn sketches. The tech giant is rapidly rolling out the model across the Gemini App, Flow, Google AI Studio, and the Gemini Enterprise Agent Platform. This aggressive multi-surface product deployment signals Google’s intent to build an immediate developer ecosystem around agent-native, multimodal video editing before competitors can establish a foothold.

Amazon Web Services · Source AWS has consolidated its technical debt remediation, code translation, and business logic extraction capabilities into a unified modernization suite called AWS Transform. This announcement coincides with Gartner naming AWS as a Leader in the 2026 Magic Quadrant for AI Augmented Code Modernization Tools, ranking them highest in the Ability to Execute category. By demonstrating that the platform can analyze over 7.9 billion lines of code and accelerate full-stack Windows modernization by up to five times, AWS is addressing the substantial corporate IT budget sink of maintaining legacy software.

Google Cloud · Source To address the soaring compute costs of running autonomous agents, Google Cloud launched GKE Agent Sandbox, a solution that increases agent container density by up to 3.5x and slashes compute costs per agent by up to 75%. To ensure enterprise-grade security accompanies these cost-saving measures, Google also introduced Model Armor, allowing companies to centralize and govern AI security policies in a single location rather than dispersing them across individual code blocks. These infrastructure tools highlight Google’s strategic shift toward solving the practical, unit-economic bottlenecks that enterprises face when scaling up multi-agent workloads.

OpenRelay · Source OpenRelay has officially launched its unified inference platform, aggregating global AI compute across 22 locations and four continents to expose them as a single developer endpoint. Already routing 100 billion tokens per week across eight accelerator SKUs—including NVIDIA, TPUs, Trainium, and AMD—the company offers inference up to 20% cheaper than direct hardware rentals. By abstracting hardware management and cloud switching, OpenRelay directly addresses the severe chip access and fragmentation issues currently plaguing the AI ecosystem.

Google · Source Google is shipping major feature updates to Ask Maps, expanding the conversational AI tool to over 150 English-speaking markets. The update introduces conversational food ordering with initial partners Square and Toast (and Uber Eats coming soon), Gmail integration for personalized intelligence, real-time transit information, and conversational tips and edits. This launch demonstrates Google’s commitment to transforming search from traditional directory queries into an agentic, actionable transaction portal embedded inside maps.

Also Noted

• Oklo (Source): The advanced fission company achieved first criticality at its Groves Reactor in less than a year following its groundbreaking, drawing public commendations from backer Sam Altman.
• Magnitude (Source): The team launched Magnitude, an open-source, terminal-based local AI agent designed to run models 100% privately and offline right on user hardware, completely eliminating token costs and external API dependencies.
• SK Telecom (Source): The South Korean telecom giant released A.X K2, a 688-billion parameter open-source Mixture-of-Experts model with a 256K context window that is optimized for math and long-context reasoning under the Apache 2.0 license.
• Browser Use (Source): Gregor Zunic announced the release of Browser Use Cloud v4, claiming a major technical breakthrough that solves the speed and cost bottlenecks for large-scale web-browsing AI agents.
• Standard Machines (Source): Jacob Peake announced the launch of Standard Machines, which builds reinforcement-learning environments for chip design to enable small teams to quickly tape-out advanced chips.

🔍 There is a strong undercurrent of AI-driven cybersecurity and physical operations in today’s posts—would you like me to compile a comprehensive brief on how these startups are targeting industrial automation and hardware infrastructure?