01 SpaceX inks partnership with Cursor and reserves a $60B buyout option
SpaceX has entered a commercial arrangement with Cursor, the startup that builds AI tools for programming, taking an option to acquire the company for $60 billion or instead pay a $10 billion fee. The deal was framed as a way to combine Cursor’s automated coding platform with the broader group of Musk-linked companies ahead of an IPO for the SpaceX / xAI / X ensemble.
The agreement would give SpaceX a pathway to bring Cursor’s developer tools into xAI’s stack. Reports note both Cursor and xAI lack proprietary models that match leaders such as Anthropic and OpenAI, so the tie-up appears aimed at shoring up gaps in tooling and developer reach rather than instantly vaulting either firm to model parity.
The structure — an expensive option plus a fallback payment — highlights two facts at once: SpaceX is prepared to pay for strategic access to developer tooling, and the companies see value in a tighter commercial relationship even if an outright acquisition is not immediate. Observers say the move also signals how corporate combos are being used to assemble AI capabilities ahead of market events like IPOs.
- $60B option or $10B fee: SpaceX can buy Cursor for $60 billion or pay $10 billion as part of the deal.
- Tooling, not models: the arrangement appears aimed at combining Cursor’s developer tools with xAI’s products; neither firm currently matches top models from Anthropic or OpenAI.
- Strategic positioning ahead of IPO: the deal fits a pattern of bundling AI capabilities across Musk-linked companies before a public offering.
02 Google releases two TPUs — one tuned for training, one for inference
Google unveiled the eighth generation of its Tensor Processing Units as a pair: a TPU aimed at training and another engineered for inference. The company framed the chips as infrastructure tuned to more agentic, multi-step AI workloads.
Splitting the TPU line reflects a broader trend: specialized silicon for distinct model phases can improve performance and cost efficiency when running complex agents and persistent workflows. Google positions the chips as platform pieces for enterprises and cloud customers building next-generation AI systems.
The announcement signals how major cloud providers are adapting hardware to support emerging agentic use cases rather than relying on a one-size-fits-all accelerator. Expect the new TPUs to be marketed to customers that need scale for both model development and low-latency inference.
- Two chips: Google’s TPU gen-8 includes a training chip and a separate inference chip.
- Built for agents: Google framed the hardware for more agentic, multi-step AI workloads.
- Cloud focus: the TPUs are aimed at enterprises and cloud customers that need both training scale and inference latency.
03 OpenAI makes ChatGPT for Clinicians free to verified U.S. medical providers
OpenAI announced that ChatGPT for Clinicians will be free to verified U.S. physicians, nurse practitioners, and pharmacists. The offering is designed to support clinical care, documentation, and research workflows.
Making the clinician-specific ChatGPT free for verified professionals lowers the access barrier for tools that can draft notes, summarize clinical information, and assist with research tasks. OpenAI framed this as a targeted rollout to practitioners rather than a broad consumer change.
The move could accelerate adoption of AI in clinical settings by providing no-cost access to a specialist build of ChatGPT for eligible U.S. clinicians, while leaving broader deployment and integration decisions to health systems and providers.
- Free for verified clinicians: U.S. physicians, nurse practitioners, and pharmacists can access ChatGPT for Clinicians at no charge.
- Clinical use cases: the product is pitched to support care, documentation, and research tasks.
- Targeted rollout: OpenAI’s change applies to verified practitioners, not a general public release.
What moved around the edges
Gemini now takes notes for in‑person rooms and non‑Google calls
Google expanded its Gemini-powered Meet notetaker to generate transcripts and summaries for in-person meetings and for calls hosted on Zoom and Microsoft Teams; in-person support had previously been limited to an Android alpha release.
The Verge AIモデル開発で新しい論点が浮上
ChatGPT Images 2.0 introduces a state-of-the-art image generation model with improved text rendering, multilingual support, and advanced visual reasoning.
OpenAI BlogOpenAI shows WebSockets and connection-scoped caching speed up agent loops
In a technical blog post, OpenAI demonstrates how using WebSockets and connection-scoped caching in the Responses API reduced API overhead and improved latency for Codex-style agentic workflows.
OpenAI BlogOpenAI publishes an open-weight Privacy Filter to spot and redact PII
OpenAI introduced the Privacy Filter, an open-weight model designed to detect and redact personally identifiable information in text, which the company says achieves state-of-the-art accuracy.
OpenAI BlogAnthropic’s Mythos preview rolled out to agencies while CISA lacked access
Several U.S. federal agencies have begun using Anthropic’s Mythos Preview to find vulnerabilities, but Axios reported that the Cybersecurity and Infrastructure Security Agency did not have access to the preview at rollout.
The Verge AI