Company@X

Signal of the Day

Tesla and SpaceX have announced a massive, vertically integrated hardware push to construct “Terafab” in Grimes County, Texas. This facility aims to become the largest chip manufacturing plant ever built, designed to produce over 1 terawatt of compute per year to satisfy internal chip demands that current and future global supply chains cannot support.

Key Announcements

OpenAI · Source OpenAI has launched its upgraded GPT-5.6 model family, shaking up both its paid and free tiers with significant reasoning and UI updates. Paid Plus and Pro tiers are upgraded to GPT-5.6 Sol, which powers everyday chats and boasts a 68% reduction in factual errors within highly sensitive domains like finance, law, and medicine compared to GPT-5.5 Instant. To address developer and user feedback on compute budgets, OpenAI introduced a reasoning slider that allows paid users to manually adjust the reasoning effort the model puts into responses. Meanwhile, Free and Go tier users get unlimited chats with GPT-5.6 Luna starting tomorrow, featuring a new “Think” button that allows them to opt-in to deep reasoning on difficult queries. This bifurcated strategy highlights OpenAI’s focus on retaining high-value power users while scaling its mass-market tier with controlled, on-demand compute options.

Amazon Web Services · Source AWS has launched Amazon Web Search natively on Amazon Bedrock, resolving a major enterprise hurdle in LLM deployment. Generally available today, this feature allows developers to enable real-time, cited web grounding across all Bedrock models using a single API parameter. Crucially, the native implementation completely eliminates the need for third-party search APIs and external vendor reviews, ensuring that enterprise queries never leave AWS secure data boundaries. By resolving these compliance and data-leakage bottlenecks, AWS is aiming to capture highly regulated enterprise AI workloads that have previously hesitated to use web-grounded agents.

Meta · Source Meta has demonstrated major strides in AI reasoning by entering its internally trained Muse Spark models into five prestigious international STEM Olympiad competitions (including the APhO, IPhO, and IMO). Operating under live competition conditions and graded by official judges, the models achieved gold medals and perfect scores on physics theory exams. Notably, the Muse Spark models operated with zero external tool use—eschewing calculators, search, and code interpreters—relying instead entirely on multi-agent orchestration and parallel reasoning to showcase raw cognitive depth. In tandem with these academic triumphs, Meta’s Muse Spark 1.2 matched GPT-5.6 Sol in third place on game development leaderboards while running at $0.69 per test, which is 10x cheaper than models like Fable, Opus, and Sol.

Vercel · Source Vercel, in partnership with industry heavyweights AWS, Cursor, GitHub, and OpenAI, has introduced “Agent Plugins”. This new open standard is designed to extend agentic ecosystems by bundling specialized skills and Model Context Protocol (MCP) servers across different agent runtimes. This collaborative standard represents a strategic move to prevent ecosystem fragmentation, allowing developers to build toolkits and integrations that run seamlessly regardless of the underlying developer interface or agent.

Google DeepMind · Source Google DeepMind and Google Research have open-sourced the code and weights for WeatherNext 2, a state-of-the-art AI model published in Nature that predicts tropical cyclone tracks and intensities. WeatherNext delivers roughly a decade’s worth of weather forecasting progress in a single jump, providing an average of 24 hours of extra lead time to prepare for severe storms. Because the model can generate a 15-day probabilistic forecast scenario in under a minute on a single TPU, open-sourcing its weights lowers the operational barriers for localized forecasting agencies and academic researchers globally.

Hadrian · Source Precision manufacturing startup Hadrian has announced a massive $1.37 billion Series D funding round at a $7.87 billion valuation, highlighting strong investor confidence in physical AI and national defense logistics. The round was co-led by WCM, Washing Harbour, Valor, 137 Ventures, and Baillie Gifford, with JPMorgan’s Strategic Investment Group acting as an anchor co-lead. Hadrian plans to use the capital to accelerate its mission to reindustrialize American manufacturing, betting heavily on physical AI in the “world of atoms” over digital-only solutions.

Starcloud · Source Space-compute pioneer Starcloud has revealed a rapid scaling timeline, raising $200 million and hitting a $1 billion valuation just 17 months after graduating from Y Combinator. Starcloud successfully trained the first large language model in orbit in November 2025 using a single Nvidia H100 GPU and has now filed with the FCC to deploy a constellation of 88,000 compute satellites. The company’s business model is built on taking advantage of abundant solar energy in orbit, bypassing earthbound cooling and regulatory bottlenecks despite the massive transport and comms costs associated with orbital hardware.

Also Noted

• Google Maps (Source): Expanded its AI-powered “Ask Maps” conversational experience to six new countries, introducing real-time transit updates and secure Gmail integration to automatically reference hotel and flight bookings.
• Cursor (Source): Released details on Cursor Router, an intelligent classification system that dynamically routes user queries across models like Grok, GPT-5.6, and Opus to minimize developer costs and latency.
• SpaceFlow (Source): Launched its YC S26 autonomous procurement engine, managing $400 million in supplier spend within customized, on-premise SAP enterprise environments.
• Olio Labs (Source): Unveiled its in vivo drug discovery platform designed to predict human clinical outcomes from a single high-throughput experiment.
• Relativity Space (Source): Commenced Stage 2 testing at NASA Stennis, completing cryoproof and pressurization milestones ahead of its integrated stage hot fire campaign.

🎧 This digest would make a solid audio briefing if you’d like a quick podcast-style breakdown of today’s developments.