# Codex Broadens, GPT-Rosalind Targets Biology, and Fairwater Comes Online *By AI News Digest • April 17, 2026* OpenAI dominated the day with a major Codex expansion and a new life-sciences model, while Microsoft and Google DeepMind offered fresh evidence that AI deployment is accelerating in datacenters and robots. ## OpenAI leads the day ### Codex moves further beyond coding OpenAI said major Codex desktop updates are rolling out starting today, adding macOS computer use, persistent automations, image generation, and support for 90+ plugins [^1][^2][^3][^4][^5]. It says Codex can use Mac apps in the background with its own cursor, learn from previous actions, remember work preferences, and handle ongoing or repeatable tasks [^2][^6]. Sam Altman also highlighted an in-app browser and described computer use as more useful than he expected because it can work across Mac apps in parallel without interrupting the user [^7][^8]. *Why it matters:* The update combines app control, memory, scheduling, image generation, and tool integrations in one release, extending Codex into broader desktop workflows [^6][^3][^5]. ### GPT-Rosalind targets biology with gated deployment OpenAI introduced GPT-Rosalind as a frontier reasoning model for biology, drug discovery, and translational medicine [^9][^10]. The company said the life sciences series is optimized for protein and chemical reasoning, genomics analysis, biochemistry knowledge, and scientific tool use, and is being delivered through ChatGPT, Codex, the API, and a life sciences research plugin with more than 50 templated skills [^11][^12][^13]. The research preview starts with qualified customers including Amgen, Moderna, the Allen Institute, and Thermo Fisher, while OpenAI says deployment is being handled carefully with differentiated access for verified researchers in light of dual-use risk [^12][^14][^13]. *Why it matters:* OpenAI is pairing domain-specific scientific capabilities with workflow tooling and restricted access in a high-stakes area [^13]. ## The physical AI stack keeps scaling ### Microsoft says Fairwater is live ahead of schedule Satya Nadella said Microsoft’s Fairwater datacenter in Wisconsin is going live ahead of schedule, bringing together hundreds of thousands of NVIDIA GB200s in a single seamless cluster [^15][^16]. Microsoft said the site is designed as an integrated datacenter, GPU fleet, and network system for exponential-scale training and inference jobs, with claimed 10x performance versus the world’s fastest supercomputer today [^16]. The company also said Fairwater uses a liquid-cooled closed-loop system that requires zero water for operations after construction, matches consumed energy with renewable sources, and is one of several similar sites under construction alongside AI infrastructure in more than 100 datacenters worldwide [^16]. *Why it matters:* The buildout shows how frontier AI competition is now being expressed through power, cooling, networking, and inference capacity as much as through model announcements [^16]. ### DeepMind and Boston Dynamics put Gemini reasoning on Spot Google DeepMind said it partnered with Boston Dynamics to power Spot with Gemini Robotics embodied reasoning models [^17]. DeepMind said Spot can better understand its surroundings, identify objects, follow simple commands such as tidying a room, and interact through plain English instead of complex code [^17][^18]. It also said the system bridges Gemini Robotics ER to Spot’s tools for movement, photography, and grasping so the robot can carry out more complex tasks [^18]. *Why it matters:* This is a concrete deployment example of language-guided robot control moving onto a commercial platform [^17][^18]. ## One enterprise signal worth noting ### Cohere says enterprise AI adoption broke out Cohere CEO Aidan Gomez said the past year was the company’s breakout year for enterprise AI adoption, with revenue growing 6x last year and expected to post another large multiple this year [^19]. He described North as infrastructure for agents in high-security environments, including on-prem and air-gapped deployments for manufacturing, public sector, finance, and energy, with enterprise controls and support for non-coders building automations [^19]. Gomez also emphasized privacy, sovereignty, and cloud-agnostic deployment as core to Cohere’s positioning [^19]. *Why it matters:* The comments add a rare revenue datapoint and reinforce how much enterprise demand is concentrating around secure deployments [^19]. --- ### Sources [^1]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2044828389929062874) [^2]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2044827932145897652) [^3]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2044828148890812538) [^4]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2044828015780343940) [^5]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2044828378147311990) [^6]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2044827705406062670) [^7]: [𝕏 post by @sama](https://x.com/sama/status/2044858863246323811) [^8]: [𝕏 post by @sama](https://x.com/sama/status/2044858862042591378) [^9]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2044861690911850863) [^10]: [𝕏 post by @gdb](https://x.com/gdb/status/2044891908213027032) [^11]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2044861694216900901) [^12]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2044861695911477643) [^13]: [Episode 16: Building AI for Life Sciences](https://www.youtube.com/watch?v=UZyH0nx5zgI) [^14]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2044938017530577210) [^15]: [𝕏 post by @satyanadella](https://x.com/satyanadella/status/2044767391293509761) [^16]: [𝕏 post by @satyanadella](https://x.com/satyanadella/status/1968677244861379012) [^17]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2044763625680765408) [^18]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2044763631858909269) [^19]: [Cohere's Aidan Gomez: The Canadian Founder Taking AI Models To The World, Not Just Big Tech](https://www.youtube.com/watch?v=RZ9I_CsDaoo)