Inception hikes Mercury prices 400%

SpaceX reported third-quarter revenue of $14.6 billion, and the traditional rocket business is no longer the main event. The company’s new "Super Intelligence" division contributed a staggering $7.5 billion to that total, effectively generating half of the aerospace giant's quarterly revenue.

This division houses Grok, the Cursor coding assistant, and a "neocloud" business dedicated to renting out massive AI computing power. By bundling consumer models, developer tools, and raw compute into a single division, SpaceX has quietly built an enterprise AI juggernaut inside a rocket company.

For the industry, this is a structural shift. Musk is proving that the real money isn't just in training frontier models, but in controlling the entire stack from the data center to the IDE. The legacy cloud providers should be terrified of a competitor that can generate $7.5 billion in a single quarter by treating AI compute as a primary utility.

OpenAI has acquired Orchestra Bio, a startup specializing in AI agents that identify how delays in drug programs impact overall timelines. The technology will be directly integrated into Rosalind, OpenAI's suite of life-science tools.

This is a sharp escalation from providing general-purpose models to building specialized, vertical-specific agents. By absorbing Orchestra Bio, OpenAI is signaling a direct push into the pharmaceutical industry, targeting the massive logistical inefficiencies that plague drug development.

Builders should note the playbook: OpenAI is actively acquiring domain-specific agent companies to reinforce its enterprise moats. If you are building a thin wrapper for life sciences, you are now competing directly with the Rosalind platform. The winner here is OpenAI's enterprise valuation; the losers are independent biotech AI startups who just found themselves in the crosshairs.

Anthropic has confirmed a draft AI safety bill that would grant the Commerce Department sweeping powers over model deployment. The proposed legislation allows the agency to seek a judge's order to block the release of any AI model deemed to pose an imminent risk of 50 or more deaths, or $1 billion in damages.

This is a brazen attempt to legislate a regulatory moat. By defining "imminent risk" in terms of massive casualties and financial ruin, Anthropic is pushing for unprecedented federal oversight that would disproportionately burden open-source developers and smaller labs who lack the legal armies to fight Commerce Department injunctions.

The real loser here is permissionless innovation. If this bill passes, releasing a frontier model will require a federal hall pass. Anthropic wins by cementing its status as the responsible incumbent, while the rest of the ecosystem gets tied up in bureaucratic red tape.

Security firm Claroc discovered that shared conversations from Meta's Muse app were entirely findable through public internet records. The chats were exposed via publicly searchable DNS records, meaning anyone could access them without needing the direct "unlisted" link. Meta has since paused the sharing feature to investigate the flaw.

This is a catastrophic failure of basic security architecture. Relying on obscure URLs to protect private user data is a known anti-pattern, but leaking those exact URLs into public DNS logs is amateur hour.

The incident highlights a glaring blind spot in how massive tech companies are rushing AI features to market. If you are building consumer AI, treat this as a cautionary tale: your users' private conversations are radioactive material. Meta loses massive trust here, while security researchers once again prove that the AI industry's infrastructure is built on sand.

DeepSeek has released DeepInfer, the inference engine that powers V4.1 Flash, along with detailed production-serving documentation. The open-source release includes the core inference engine and API-processing layer, featuring distributed inference, speculative decoding, SWA Bounded Replay, and structured output support for both NVIDIA and Huawei Ascend platforms.

This is a direct assault on the proprietary serving stacks of Western AI providers. By open-sourcing a highly optimized, production-ready inference engine, DeepSeek is effectively commoditizing the infrastructure required to run frontier models at scale.

For developers, this is a massive win. You now have access to the same serving architecture that a major Chinese AI giant uses in production. The losers are the proprietary inference gateways that charge a premium for orchestration. DeepSeek is playing a brilliant strategic game: commoditize your competitors' infrastructure while driving adoption of your own ecosystem.