← Back to topic list

AI Daily News Rundown March 17 2025: Nvidia’s $1 Trillion Bet, Meta’s Local Agent Move, and the OpenAI "Side Quest" Purge

enoumen | 2026-03-17 21:49 | 8 views

https://preview.redd.it/oj55rr0veopg1.jpg?width=3000&format=pjpg&auto=webp&s=683d562610375aeb2d323b0915d78c23f7567a26 [](https://substackcdn.com/image/fetch/$s_!Dia_!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9688054d-c500-438a-a81b-47276d0ef44b_3000x3000.jpeg) LISTEN TO ADS-FREE Audio of this episode at [https://djamgamind.com/daily](https://djamgamind.com/daily) 🚀 **Welcome to AI Unraveled.** Today, we unravel the biggest numbers in the history of the industry. Nvidia has signaled the era of the $1 trillion chip cycle, while OpenAI admits that “side quests” are no longer an option in a competitive market. **This episode is made possible by our sponsors:** 🛑 **AIRIA:** As agents like **Manus** move onto your local machine to manage files and run terminal commands, the risk of “Agentic Drift” is real. **AIRIA** provides the governance layer to ensure these powerful local tools stay within your enterprise compliance boundaries. 👉 **Govern the desktop:** [Airia.com](https://airia.com/request-demo/?utm_source=AI+Unraveled+&utm_medium=Podcast&utm_campaign=Q1+2026) [](https://airia.com/request-demo/?utm_source=AI+Unraveled+&utm_medium=Podcast&utm_campaign=Q1+2026) 🎙️ **DjamgaMind:** Tired of the ads? Get the forensic version of this news. Join our **Ads-FREE Premium Feed** at **DjamgaMind**. Technical, deep, and uninterrupted. 👉 **Switch to Ads-Free:** [DjamgaMind on Apple Podcasts](https://podcasts.apple.com/us/podcast/djamgamind-audio-intelligence-ads-free/id1864721054) **In Today’s Briefing:** * **The $1T Projection:** Jensen Huang’s GTC vision for Blackwell and Rubin dominance. * **Inference-First Hardware:** The launch of the Groq 3 LPX rack for high-speed token decoding. * **Manus: My Computer:** Meta’s new desktop app that gives AI agents terminal-level control. * **The OpenAI Pivot:** Why Sora and Agent Mode were deprioritized to save the core business. * **Orbital Data Centers:** Nvidia’s Vera Rubin Space-1 Module and the challenge of zero-convection cooling. * **Data Center Sentries:** Boston Dynamics’ Spot dogs become a standard $200k security expense. * **The “Atoms” Filter:** Why Google and Accel are rejecting 70% of AI startups for being “wrappers.” **Credits:** Created and produced by **Etienne Noumen**. **Keywords:** Nvidia GTC 2026, Vera Rubin Space-1, Groq 3 LPX, $1 Trillion AI Chips, Manus My Computer, Meta AI Desktop, OpenAI Sora Pivot, Boston Dynamics Spot Data Center, NemoClaw, Digital Optimus, Amazon 1-Hour Delivery, Atoms AI Program, DjamgaMind, AI Unraveled. Contact us [etienne\_noumen@djamgamind.com](mailto:etienne_noumen@djamgamind.com) **Connect with Etienne Noumen**: [https://www.linkedin.com/in/enoumen/](https://www.linkedin.com/in/enoumen/) **Partner with us**: [https://djamgamind.com/partners](https://djamgamind.com/partners) ⚗️ **PRODUCTION NOTE**: **We Practice What We Preach.** *AI Unraveled* is produced using a hybrid **“Human-in-the-Loop”** workflow. While all research, interviews, and strategic insights are curated by Etienne Noumen, we leverage advanced AI voice synthesis for our daily narration to ensure speed, consistency, and scale. # Nvidia projects $1 trillion in AI chip sales by 2027 [**LINK**](https://elink640.dupple.com/ss/c/u001.QT9-cN0mZn-HYCKnMP3yhcgVBJAjHghBdrIcrZ5lOdEVtpIL32OsZ1Y5CpFoSI_VhK_yKyE1m9lA1-jDERT1-3jyO2PTwjbxJ3heNABKiqAveys75EfH4dD5nuZ5794ZYtA2frgOcBcQwI7wJlLoNXUYKkNW2ns0rND1TAJgTM2zpDOyA-ZiojUfGum8ktsf-PnUHsz9iaYSi9lDff7KKjfsDK41rkFi0WRg2J-_hFwPiX7G5lqVoYwD-REkE1V2eAMXfwIbBQ1AwQrFcd6W1Fg7QUtjkonnJCs6FvCvaP4/4p0/lsZsIDvcT4eTojCeqx1FFw/h7/h001.IVpeFG1Qh_HWR0JJSUCBdLqXj6AZOKmEVYjb8-iA4-w) * Nvidia CEO Jensen Huang said at the company’s GTC conference that the company expects to sell $1 trillion worth of Blackwell and Rubin chips by the end of 2027. * Nvidia unveiled the Groq 3 LPX rack, combining 72 Vera Rubin servers with 256 new language processing units, designed specifically for inference computing rather than training AI models. * The company also announced partnerships for autonomous driving with BYD, Geely Auto, Hyundai, and Nissan, plus a coalition of software companies working on frontier open-sourced AI models. # Nvidia unveils AI chip for orbital data centers [**LINK**](https://elink640.dupple.com/ss/c/u001.QT9-cN0mZn-HYCKnMP3yhWtS15OZmF4EHk-SJz9hjXAHVEZRCRDSNDHVYh2a5SuLcAZLL6efUdajbUuUAraCks-xShfRjWLoV6BiSIHsmtz4tjeMBUURoD9jjIuXDigHNaRCaXIckSW-Z9iQ8ZyhhD0quEVrcuV2z-oNQXF7C9czLgiO15KvA_WdcUJs0Awo/4p0/lsZsIDvcT4eTojCeqx1FFw/h8/h001.Gxc-2y9oRPtrohf4ElnFII-BOxWuGYwlpjV1wxTKKPY) * Nvidia announced computing platforms designed for orbital data centers at its GTC 2026 conference, marking a significant move to bring artificial intelligence processing into space environments. * The company’s Vera Rubin Space-1 Module, which includes IGX Thor and Jetson Orin chips, is engineered for size-, weight- and power-constrained environments and will fly on missions led by multiple partners. * Cooling remains a key engineering hurdle because space lacks convection, while SpaceX — which acquired xAI for $1.25 trillion — has asked the FCC to launch 1 million satellites for AI centers. # [**Nvidia unloads at GTC with NemoClaw and more**](https://link.mail.beehiiv.com/ss/c/u001.u02qJFHqR61XIkDbYtOHoCl0xTplzSza_HzHQveBS2vgcpukVQKS_Vjhu_tpky7-VpoGirQMuGFgbU0Snapo2JK9-QUxvmWtNZvl3LHSl3_sRDifjSdb88-NH2eL_s3sQdzN72L6twEo-p2s9lKg5J6VHQ0kBqZ4Cqz4X40QZhJU1o-H2RFPB-c1uC3NoiEf982F1qkIJiOh6bMa6sJ1cSVEolLDXp28NpXuH65XIvTnikI01a5udKALzKLkX1Eg/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h5/h001.DLoxvQl7M2e3CHmE0XlAXyGC1-Rykp3LwXvJzEPob9U) https://preview.redd.it/u3r10gyzeopg1.png?width=1456&format=png&auto=webp&s=661659e562d3aec610fa7e0fe1eb6e53c9d210e1 [](https://substackcdn.com/image/fetch/$s_!fL3T!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F1274c66f-70f0-4bf1-9458-bf1c360c28d5_1456x816.jpeg) *Image source: Nvidia* Nvidia CEO Jensen Huang [**made**](https://link.mail.beehiiv.com/ss/c/u001.u02qJFHqR61XIkDbYtOHoCl0xTplzSza_HzHQveBS2vgcpukVQKS_Vjhu_tpky7-VpoGirQMuGFgbU0Snapo2JK9-QUxvmWtNZvl3LHSl3_sRDifjSdb88-NH2eL_s3sQdzN72L6twEo-p2s9lKg5J6VHQ0kBqZ4Cqz4X40QZhJU1o-H2RFPB-c1uC3NoiEfY4vPrk9252MpbjYqBRedYfV5CTPdzBxJP2iqSE0XL6zCAy1X5ehJxEVwW9naaw7Q/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h6/h001.IPgmI3IGRc4xnoiupiiaJOBVaW1WKsTVwrE0IpyatyM) a wave of announcements at GTC 2026, including an open-source NemoClaw for agents, next-gen Vera Rubin platform, DLSS 5 for photorealistic game graphics, and new enterprise and robotics tools. **The details:** * NemoClaw [**brings**](https://link.mail.beehiiv.com/ss/c/u001.DrYDwug-xrpEbNqFhzTCdazBw7hee4iWsRnwFW5F65SGpV3t6ogQVf5jGueRy5RZp0YMPDmhcOG-n2K71ubT-udIRlACGuy-CBZq6K2EcBy5jb_Xnahpp2P0MbXuKKdk5Hf1_6j0UyUdcz6nAsXKslv3TGK_IepF_B86yh9_3gFyT4cGt1UAs19FijSLEHmEgRBFwxA4aSTiEJMj5mGIhi0B0z0w8soah_kVi6AULb5IiCgjhKdE_nLtqKn_sZxcmJWN6QWm0gDcv08YGbdYaw/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h7/h001.Ron4con9RZ5e-AuD8DcTefNbS3C9f3AnkZcWFgE4dmg) security and privacy guardrails to OpenClaw agents, with an emphasis on growing the agentic tech’s use across enterprises. * The Vera Rubin platform puts seven new chips into production to power AI training and agents, with Huang also [**teasing**](https://link.mail.beehiiv.com/ss/c/u001.DrYDwug-xrpEbNqFhzTCdazBw7hee4iWsRnwFW5F65Tur59--Url0vwpyngtLDzzg3vjW-MMxTaSMsBIGEAxHIJJJLVP4OMxpCwlxb8kkzAtl-GKdRQugT1MaLbOq31HKfb4IZnoW86YXxYXFz9bm052OYPUCe9AdvzgMdx0LXvmmXs3UULoweHljq8ay5OFmP9ToacgN0VFsSrvrQIA11FigmoJbKXglKNMk0d2K1opmIv1VDz68HGrG-Z8SOhg89mj-2aFvN51jU5qVmT49Q/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h8/h001.Qvg9KfQcibN8Y6X-MSHYKh3fPaXvnmfofEUfyqRHdhc) space-based data centers. * DLSS 5 [**uses**](https://link.mail.beehiiv.com/ss/c/u001.dSnm3kaGd0BkNqLYPjeMfz-OaGtlrSa6VCor0GzRl-L_e3NjjE52__GyhlBQ8rNIxSREhIRazf3Up2ZGIiD0ErCPzsgQvi_9Zv39PHvf9ikCynUGQaccvDfHX_6TtrJ5Sai2z_501oUInTs_5vbbkLDYXDMuEIJyhlE3xEZPfgqRYDDc3hTkix0YecbrGJWMVKwPwE-fmraSsFYSrLZ1y81yO7bYPt7pK0g1shKf4NEQIh0S1RZhXkwWfqYhqy4qCYfqs_XPZihd-zF7MZ8pLxTPd9LOhTHvSHUNTa7-RZ3xLQ7_IYx6DGf8mc-PlPLj/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h9/h001.2m3VCVqb3LlT_lVceTqBofmusy3g88v181v0sUgV_iM) AI to add photorealistic lighting and materials to games in real time, with Bethesda, Capcom, and Ubisoft among the first studios on board. * A new open-source Agent Toolkit lets enterprises build secure AI agents, along with new AI platforms and partnerships for [**vehicles**](https://link.mail.beehiiv.com/ss/c/u001.DrYDwug-xrpEbNqFhzTCdazBw7hee4iWsRnwFW5F65T5pR-GEuKDaBk0FfONvZS5x155I3fOFP64ilsgLpogxnY-KDVIAxCDcGzO6CYBAOuQGJNs4PXxOBFqhfbmkZrMWPfrpUQrcPRC0Hdk29KNGaf5GeabcGkiPJPoShRl8Yh2D55J8T9wpfFNYFy9njK84R2jtYKtCV1tQdrESxT6pVDQBZfpW_IBtZlqHkVf3m5prciethIUgJG5Oq8NdsXXm2e5ORLdwvMsB5q8bRZmQg/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h10/h001.jKFbWGSQLKNsK0vzbs8cXTDx-yvWqYg3w_dsBNgiOkw), [**robots**](https://link.mail.beehiiv.com/ss/c/u001.DrYDwug-xrpEbNqFhzTCdazBw7hee4iWsRnwFW5F65SGpV3t6ogQVf5jGueRy5RZIoW11d_RRvBso4C-YUOJ-Fu3spMkWZYYnUqdoI53ZNVjIzA6jD-p-KSrruzSu9CzRsWz_cy9kcOreD68AHpukmeczwctkkPrT0c_u4UfuvPMSoBCoBfUjWn_jtFd2tMd7CUUsH1PyRk2eFc_oTE09WM3aWEyah7SGR0xntzt-XSdndCuZz322zqE3tYmYuVWIt2Sd_gpodcM6lbENl-bLntM4DgIthAhqn12co8JA3YVxX0QbYbVD5vW5r0VA2Y_7o5hYD4OdjwJZL-yXfehpw/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h11/h001.rRf-P6AWn6kOdHeagxWNbZuuyTbNpsFKAdAFuZIrWFc), and more. **Why it matters:** Huang pitched Nvidia as ‘the first vertically integrated but horizontally open company,’ and GTC makes that case hard to argue. Chips, agents, game graphics, robotics — every announcement pointed to the same play: own the infrastructure layer beneath all AI workloads, and let everyone else build openly on top. # [**Manus brings its AI agent to the desktop**](https://link.mail.beehiiv.com/ss/c/u001.BKH0F2yLXfXXfZz4rVL6METPtw3uGuvkRWPxqNNWuN45tZG1IsR4AWZWMBVmyDE6siEWGmHT2tYxE_5dlg1btkYOgun8CI6MfMVanDQCblP3KGbA8OJcrqfmqLiP1cpbg7QBOXnFrQLvWogBEjbBIRX9_p3F04i8G3adxosc1QFiv8VoE7MXxNCm0_vJopnBsXc7b3v8KhMmxITRtUkYjzNuen5djg9FHWDh1a9lxq8iC3W93nEn43xCjJHpX7YF_hmHY19oK01I5EVmBlZgpA/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h27/h001.SkG_IYmT6VM4tMukYtV6GKtlWAr0Al9KXoy0DMFcjo4) https://preview.redd.it/ms3yxio1fopg1.png?width=1456&format=png&auto=webp&s=b7258ed3d14865580697a19adf7dbda42f8f5b26 [](https://substackcdn.com/image/fetch/$s_!tnPV!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F22caf3b0-6f5a-4f03-ab0a-06fff5421e26_72x72.png) [](https://substackcdn.com/image/fetch/$s_!aiRy!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdce5459d-1551-481d-9e2a-808f4721af69_1456x816.jpeg) *Image source: Manus* Manus just [**launched**](https://link.mail.beehiiv.com/ss/c/u001.BKH0F2yLXfXXfZz4rVL6METPtw3uGuvkRWPxqNNWuN45tZG1IsR4AWZWMBVmyDE6siEWGmHT2tYxE_5dlg1btkYOgun8CI6MfMVanDQCblP3KGbA8OJcrqfmqLiP1cpbg7QBOXnFrQLvWogBEjbBIRX9_p3F04i8G3adxosc1QFiv8VoE7MXxNCm0_vJopnBstEEEppG46T9Pg4tvCWappnVBYBfRAPXvC4LF3jaU56yII6zjM7WVQUDSUUHqcilpypW47RP7xwZw8Wf93mCiQ/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h28/h001.azwOczf3GbA2x4bKV0FtHmaBo4s4nWfjFt2Zd1p7d5U) My Computer, a new desktop app that moves its cloud-based AI agent onto users’ local machines to manage files, run terminal commands, build apps, and more. **The details:** * My Computer works through the local terminal, giving the agent direct access to read, sort, and edit files stored on a user’s machine. * Use cases range from organizing unsorted photos into labeled folders, batch-renaming invoices, to building and packaging apps autonomously. * Meta [**acquired**](https://link.mail.beehiiv.com/ss/c/u001.dSnm3kaGd0BkNqLYPjeMf022yDnKIUiG8l1A8gWu3xu_DTOkaYrvywWa72fLbanUgQzoU2YFI1_adPdLRCc2Hzy8cQ9aaRBbUYeHDc2VGwiItqI7b5zwYmifg1nlFZ6JCuBy4Y1uodlnROUI7i05kNYKew3YkX68vUayxBQfIhRTtrK5oXMZsFcrna1PAd5e1d7O04iaKsoStaokPantUPWYH992xq0AVSqzJFc4vaN_Z0kKfAzlpmhn4bXWZHd88NfhaFuB7Enoca0yAGjDEg/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h29/h001.pxfhazee1yxgETEVWJI24-HvJsL1VgPrZcZtlus7_bQ) the Chinese agentic startup in December for $2B, with its team joining the company, and CEO Xiao Hong coming in as a VP. * The agent can also tap into a machine’s hardware when it’s sitting idle, running jobs in the background, or completing tasks assigned remotely from a phone. **Why it matters:** Manus was already one of the more capable AI agents in the cloud, and now it’s making the desktop move we’ve seen from OpenClaw, Perplexity, and others. The race to be the orchestrator of users’ computers is on, and Manus is a good opportunity for Meta to gain a foothold without a current frontier model of its own. # OpenAI cuts side projects to focus on core business [**LINK**](https://elink640.dupple.com/ss/c/u001.i8ligEAMODugkGN-qAs51T96-F3mXwedXWzV3XFf-pg1XEMmUwlfqbDiRjMRZKnRp6T5QakdXl0YJgmfTlwJJri-eBw29npSiyQTzobs9IKTxTJRyQ3X29aCWLxCqDWGbtSKb5lyOfljI3TXEEokRjH6NR5HuhLqm5XTeD9NURXk-T1MJMDhgVIeacx_Zt9GL75xNKVaLuc4SW_WUC4tvd1jRRUgFb7cnyKJDu2NXICP6d_vKdClLe0mOKCgITtW/4p0/lsZsIDvcT4eTojCeqx1FFw/h9/h001.TMHaSsvGLwr9PPocV4HTdYZlnYST8Ch_Z8NjCAiqofw) * OpenAI is shifting away from launching many products at once and will now focus its resources on two core areas: coding tools and business customers, according to a Wall Street Journal report. * Fidji Simo, OpenAI’s CEO of Applications, called Anthropic’s success in enterprise and coding a “wake-up call” and told employees the company “cannot miss this moment because we are distracted by side quests.” * Products like the Sora video generator and an agent mode struggled after launch, with Sora’s usage going flat after briefly hitting number one in the Apple App Store and the agent losing most users. # Amazon launches 1-hour delivery across the US [**LINK**](https://elink640.dupple.com/ss/c/u001.QT9-cN0mZn-HYCKnMP3yhWtS15OZmF4EHk-SJz9hjXAsjdavlhIbaujP5PYIJQgt5QcAiZjdCehBnFWg9SrTgQtO2XCOt9-MRMxcuULAXQ3ApDIopo5-aRhhEpEA_RIwagrIugpBO-snVfFvNcSmmHYksmimDyjACyqjpTZzbn1eAZvlo3RJ_hAKpuAfOub50Lol9Cjdd5T9HdYb2qYNXNmOhVcGTWcOYcDrziJ4Mww/4p0/lsZsIDvcT4eTojCeqx1FFw/h10/h001.2RhpjIOGFxqucVx6WWpxs25Hx3vZK4HMGQq6zyQXj_U) * Amazon announced it is now offering one-hour and three-hour delivery options across parts of the U.S., covering about 2,000 cities and towns for three-hour service and hundreds for one-hour. * More than 90,000 products are eligible for three-hour-or-less delivery, including pantry items, cleaning supplies, over-the-counter medications, clothing, and toys, with no change to everyday low prices. * Amazon added a storefront shopping page and search filters so shoppers can find products available for one-hour or three-hour delivery, and plans to expand the service to more areas soon. # Data centers are turning to $200,000 robot dogs to guard the facilities [**LINK**](https://elink640.dupple.com/ss/c/u001.QT9-cN0mZn-HYCKnMP3yhQaef6i08JVqfevH1xwmey5dOdWU6xCAsUlrTzlp2QleZBbRvFBuRfdP__7NEHZACwc0DCPvumgr57BvJCyMeRlVhOIxICdrdgVBVPd3mJLGEPLdu6n_KBMwX_G0L5oGoTfzOq-3SXxrlZspNIHBYJsQC1IFVwYRvGgn-R4UX5Y6cHXMX-LPsLRc0svixddaDg/4p0/lsZsIDvcT4eTojCeqx1FFw/h11/h001.NCKofn55PrBTWAFky0jyYnYgLH4I0Qbis5qkdRIq0O8) * Companies like Boston Dynamics and Ghost Robotics are selling four-legged robots, priced from $165,000 to $300,000, to patrol and inspect AI data centers around the clock. * Boston Dynamics says interest from data center clients has jumped sharply in the past year, with its Spot robot detecting temperature changes, leaks, and unusual noises across server halls. * Ghost Robotics’ Vision 60 handles external perimeter security, and both companies say the robots are meant to augment human guards rather than replace them entirely. # Whistleblowers say Meta and TikTok profited from ragebait [**LINK**](https://elink640.dupple.com/ss/c/u001.XGyucGaZXoGCX9sMMuMKgtY4kCqoYPszq50yFftduOdk8U48LvsTGz4-mvyrhOp1r4-qXo2UcqErbSvck2__HR78fTZDTFI5lEm4QktkLO-AkEZCRMxXmcID0J6o-c6lO5ZN6XqCnlC6XdNbpSQEOoRQRO_XNNeKMl1uEG6oYZKlT4A_3MyvpPDpJvo5p4dhydhLFL6oETGbWYuSTd78kwI6QojKiMjvOUqLqN8K_p3-anbsgbx6qFpfFFCMmqU1SBxcCL3lOq1gOSOCcs3IKde_a1BeMob0VOwt6T7dO_w9Xd4u62h3ksv57AlmkTgKEuPBb5KwREMO-apaMBsZsmgugr660RQ-iGIMMbRZ9s0/4p0/lsZsIDvcT4eTojCeqx1FFw/h12/h001.h5cj79Qrsk43MyW90pxIItLUahNYehgvMrKk7A3jixo) * Whistleblowers from Meta and TikTok told a new BBC documentary that both companies knowingly allowed harmful ragebait content on their platforms because it drove engagement and growth. * A Meta engineer said staff were told to keep borderline material like misogynistic posts and conspiracy theories visible in feeds to compete with TikTok, partly because the stock price was down. * Internal TikTok dashboards reportedly showed complaints from politicians were prioritized over reports from young users, including a 16-year-old girl who flagged fake sexualized images of herself on the platform. # [**Atoms program elides AI wrappers**](https://link.mail.beehiiv.com/v1/c/HLb01%2FHfQZ2YuDH%2Bb2CODXxt9ok1uZlWYlzBoemUvq009zCJynp7T9rB%2FSGT%0AVKfkfxdxfp6%2FyoFE1CparOGxhVRYnzOh7X9YyJlZB22%2Ftv%2BAqSyWWGmkPOU8%0APnBq2q%2FNt%2B28psVmb62L3jgNomHvvb87sW1EGd1K%2BZC0lvPlj4k%3D%0A/ed1940cd6f27ef2b): Google and the venture firm Accel started the Atoms program to fund early-stage AI startups in India. But in a sign of the times moment for 2026 tech, they found that “AI wrappers” dominated the submissions. (That is, startups based around relatively superficial features that are built atop of powerful third party AI models.) Accel partner Prayank Swaroop told TechCrunch that this definition fits 70% of the overall applicants for the program, but 0% of the five final selections. Those lucky startups will receive $2 million in funding from Accel and Google’s AI Futures Fund, along with $350,000 in Google compute credits. # [**Alibaba preps “AI agent” product**](https://link.mail.beehiiv.com/v1/c/l7qOIo1WygdzEPfnSOnZCXdcnQVzoZa%2BgI5%2FsSRPieRZ3ld1ZyFNnIM4cZrK%0AWktkTCKqkr6oy9olL8WrA7wmf%2Fbt2KrD7qC671VDo3C2XW%2BPfVy9S1SU6qC9%0AJ1H6AQbjAComAtn%2FJgYisw6EdT2W27l6H2qVwUj4mhSzOA%2B3VTg%3D%0A/8dd0bbe226c59633): \\Bloomberg reports that the Chinese giant plans to ride the country’s [OpenClaw/agentic AI trend](https://link.mail.beehiiv.com/v1/c/T7tXTX8Y8eIr7JYFrwmjNRX2KsDpK5Q1kiEoO1QRyrAvnTpXWXgXwGuu6Zg5%0A0yEH7Nhi%2BnHpvNeSqJKjcEAjwgf%2FzVxgSRVQfJrsD2rfkScr%2Fk4GsMHloVUt%0Ale5tnPMhxZce%2BRnX0TfCqRMuZ2xu8eNCl1G5wS8Rq4SlreA0nQ8%3D%0A/bc1a8e05773c27f8) by releasing its own enterprise-ready AI assistant, built atop its Qwen family of models. The agent gets built-in access to the Alibaba galaxy of platforms, including shopping destination Taobao and fintech service Alipay. How will this integrate with existing enterprise platforms and services? What does Alibaba plan to charge companies that want custom agents for all their executives? Great questions, for which Bloomberg has zero answers. # What Else Happened in AI on March 17th 2026? **Microsoft AI Envisioning Days** – Free video series helping software companies build, deploy, and monetize AI apps and agents. [**Watch now**](https://link.mail.beehiiv.com/ss/c/u001.zulPMwBCNL5ahjB956nrIrUZ60WIQpTDf3tzUbjaBI9x_UuVnx9UWQRb3X7D8DV4D-FtgDQ4XnTRsktrajD6SSo2b6btPEvxvdWHnXSdcV1_C6rT4-WNBZvxmUVrmrpe0M3kex4Csa7-AodEGtCt3gQdiIgmo-g-XBCEBJQvTIGplsjew49-QoB8LbXib1q3o2O-fx6Ur9QvhgVZJgLahdzjc-7tg-CzWx-5RJHAmkgNsgcv7izoOI4l8h8yMB4yS9pEgmNfopAH0vuvxmT60AjcuGl39b9eBPJfUVcnPXxg1lqW-eavT0QCmpfNOHPf7f_PEa8v7QGTP0lK4_qIbQ/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h36/h001.hvJfEzUfvVbRoHg0ehiYkr0xzO_RxuRrOLTGylEOTvA).\* **Encyclopedia Britannica and Merriam-Webster** [**sued**](https://link.mail.beehiiv.com/ss/c/u001.KT4rQsO6sHS_v2VASG2xukYrcBLmr-VWvDqpbYLTfcTrTc-mCaGIOGLp3TguJnUuql794OlkSfMlu1hWFzvUF2a52EysWYUbauPgyRfBDC3YonnqU6sIbNwEL3p02l75rUziW1vFvthj7Uqn-JJqofCA8VNAjwf_1bHlkP10hen99BDwRbzi8L_BcPo7D5hncmoKJA6SC9MBnzYV59iHEVBzAo2gOB9M-u-UYGKVKx43OW3WE9h7qveJkzWi9XwsyYlUdEQO-7L0yESZ9JC2Q0WBNzrdn7wGLCE2DbkacUB3OCe67sSRmyhHV0X0VV8O/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h37/h001.aWSz_uKYE1iYBu2jS_CAeBR8tR2n0L7PBAoLip3DJlY) OpenAI, alleging it scraped their articles, produced competing outputs, and attributed hallucinations to them. **Physical Superintelligence** [**dropped**](https://link.mail.beehiiv.com/ss/c/u001.siHJl2oxYc5G1hfeCOvt3R64uvprik-D0czR2qX9nWfLH8rI5RON-3Vhi3hLpCnhQPfgKR2TQceJwlN_F-viDUf6gWe44MOHRqj4l3lmUblWs8e8elHFBMhNxNrqNf5BHYUkvROIRWa7jgEZdwMumNL0mS4vP4_pW3SaKNysngFLJhU6rErBpQG37xRI-ES3qQOz_cXjA67jntsDHu2VRb9glAw6p0_ugfCfQ_ql1YEF4p3cENrk_JSqGQJNDuSm/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h38/h001.SCV1Vlgk4etDGOmD_xEXzwtUlZZQRTAKjIXOZHqtnUs) Get Physics Done, an open-source AI agent that can scope research problems, run experiments, verify results, and draft papers. **Moonshot AI** [**published**](https://link.mail.beehiiv.com/ss/c/u001.6k0_SAz8nrOuu_-LoNX1HePrmh_LLjC15Re46y_-ArLmp5JEFC34WRhzt931x1jVE5VKV0f9uEMoSPBq9_uR1PLZO5YwPOh0seKpNy7vamnA1ZYnYm0Xa5ojAlkJ7l3p_H9UQ6K1x4dsFHlGj_R-rthwq48jl_yFpUDCn-JKzXxmjC5nSUGvL5pBiXoE3hlQBZtwXADJw3S_250N_GEexdo1q7p_7Cg4plg1kI7t4lm5GPEpbbdARMgBeL9K_3Jl011MjNTXLbJADF0E3Pmmcg/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h39/h001.HhvxvFy0mav_8dqIB264JMXVlC4-11jjcD3WHVCr5gg) Attention Residuals, a technique that lets models pull from their earlier layers instead of stacking them, delivering 1.25x more compute efficiency. **OpenAI** is reportedly [**restructuring**](https://link.mail.beehiiv.com/ss/c/u001.dSnm3kaGd0BkNqLYPjeMf4_pVCvgnnuS9csEAS3pWhukjN77GK_f-3nAppvJRQHBvii2f2MKkXR4x0hrXy4-a7rAF1RlzeEajRMTGQFLULEaJjNEKqg0oQEmi6Ro5jKHqc_SRBiYclHp6UIagHOvjEX7xfQjZ5UWJJeLXJTIWHak6v8nuRN1All50Joz3LF9j8M-JRW-9NADiz0UXATxEHGaJL3dnSdmZiumIPu0cfWMCx7a4mMubyJn6tMPlOdfNutEwvdHp7nQPHfXFL-xwzkAm3P1itTgfY2h5XAcp1CkAr3uoa3suCDrdkE_vhIEgIBJXGCHDsusCskpSFP2aMVKD_5Dq7VPQOVLMHXcblE/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h40/h001.q3QIdx63WR1SyQyjrdYNRY2UHLkTbQkDeYRmfgrWa7k) its Stargate computing team, coming alongside a shift in strategy to renting more AI servers rather than building its own data centers. **Meta** [**signed**](https://link.mail.beehiiv.com/ss/c/u001.DrYDwug-xrpEbNqFhzTCdc03qoPsRZZOK9vttjsrDro4b5GTpiJFRGs4zkdWYoWx5LByTfEFJvYCtIQeZf8W77YFWsT-f_WrYBuWCGaMHtH9wBm0Jc6j1-8QaPIO6GCO96VFlTUB8ynhxg4LXm78zrJTSG5xRrBX5VnkuYCHYxjw6sImvy2MnvlsUbMiVsuvbZ3061nF4xex14o1OW1U3PIvjqb5WXVu3X0DdWX3h1RjhiFaD8I_3qEM8FNOYXUwOeSrq65oFSyx0w7z1gHtkqZh6b7GDFBCCpExKqrgli07AH2zTx6L85CzYhd-VNbI/4p0/WEV-qaA8SHaEYRgQ6Mt7Cg/h41/h001.DkmalNZcQAmt4XPolr56QbPIw-pzTkC0T1zYfCRnsVA) a $27B deal with Nebius to deploy AI cloud infrastructure, including one of the first large-scale deployments of Nvidia’s Vera Rubin platform.

Comments (0)
No comments yet.
Add comment

Login is required to comment.

Login with Google