This Week in AI: March 9-15, 2026 (Week 11)

News of the week: March 9-15, 2026 (Week 11)

A lawsuit dominated the week: Anthropic sued the US government over being labeled a “supply chain risk” while continuing to expand its enterprise business, including through Microsoft and a new partner network. Billions flowed into AI infrastructure: Nscale and Nebius each landed two billion dollars, and Oracle showed how fast its cloud is growing. Yann LeCun raised more than a billion dollars for “world models” as an alternative to today’s language models. Google embedded Gemini more deeply in Maps and its office apps, while Meta and xAI dealt with problems with their models and teams.

Regulation and Law

Anthropic sues the Pentagon over being labeled a security risk

Anthropic, the maker of the Claude models, filed a lawsuit in federal court in California against the Trump administration on March 9. It challenges the Pentagon‘s decision to designate the company a “supply chain risk,” which in practice bars government agencies from using its services (Fortune).

The company argues that the move is retaliation for its public advocacy of AI safety rules and that the government violated its constitutional rights. The lawsuit also targets other agencies, such as the Treasury Department and the State Department, which ordered employees to stop using Anthropic’s services (eWeek).

Over the course of the week, the dispute became part of a broader debate about who should decide on military uses of AI and how that will affect the race with China. On March 12, Fortune noted that the case could set a precedent for the relationship between the government and private AI labs (Fortune).

Business and Investment

Yann LeCun raises more than a billion dollars for “world models”

The startup AMI Labs, founded by Meta‘s former chief scientist Yann LeCun, announced on March 10 an investment of roughly 1.03 billion dollars. It is one of the largest seed rounds in history, meaning the first investment in a young company (Tech.eu).

The Paris-based company is developing so-called world models, which are meant to learn from camera and sensor data and understand the physical world, rather than just predict the next word like today’s language models. The CEO is Alexandre LeBrun, while LeCun serves as executive chairman. Investors reportedly include Bezos Expeditions, Nvidia and Temasek (The Star / Reuters).

Billions for European “neoclouds”: Nscale and Nebius

British company Nscale, which builds data centers for AI, announced on March 9 a Series C round of 2 billion dollars at a valuation of 14.6 billion dollars. The round was led by Aker ASA and 8090 Industries, with participation from Nvidia, Citadel and Dell, among others (Built In).

Two days later, on March 11, Nvidia announced a 2 billion dollar investment in the Amsterdam-based cloud company Nebius for a stake of about 8.3%. Nebius aims to bring more than 5 gigawatts of data center capacity online by the end of 2030 (The Star / Reuters).

Oracle: cloud grows 44%, shares jump

Oracle reported its fiscal third-quarter results on March 10. Revenue rose 22% to 17.2 billion dollars, cloud revenue grew 44% to 8.9 billion, and cloud infrastructure alone was up 84% (Oracle / SEC).

The volume of contracted but not yet billed business (known as RPO) reached 553 billion dollars. The company’s shares rose more than 9% on March 11, although they remain well below their September high (SiliconANGLE, The Motley Fool).

Atlassian lays off 1,600 people as it shifts toward AI

Atlassian, the Australian maker of Jira and Confluence, announced on March 11 that it is cutting about 1,600 jobs, or roughly 10% of its workforce. CEO Mike Cannon-Brookes said the company needs to free up money for investments in AI and enterprise sales (Bloomberg).

CTO Rajeev Rajan is also leaving the company as of March 31. Atlassian expects one-time costs of 225 to 236 million dollars (TechRepublic).

More business news

OpenAI announced on March 9 the acquisition of the startup Promptfoo, which develops tools for security testing of AI applications and agents. The technology will become part of the OpenAI Frontier enterprise platform, and the open source project is set to continue (OpenAI).
Anthropic launched the Claude Partner Network on March 12 with an initial budget of 100 million dollars for 2026. The money will go toward training, technical support and joint marketing with partners such as Accenture, Deloitte, Cognizant and Infosys (Anthropic).
The startup Wonderful, which offers AI agents for customer support, raised 150 million dollars in a Series B round on March 12 at a valuation of 2 billion dollars. The round was led by Insight Partners (TechCrunch).

New Models

Nvidia releases the open Nemotron 3 Super model

Nvidia released Nemotron 3 Super on March 11, a model with 120 billion parameters and open weights, so companies can download it and run it themselves. It is designed mainly for AI agents, meaning systems that carry out multi-step tasks on their own (Dataconomy).

The model combines the Mamba architecture with classic attention and the Mixture of Experts technique, in which only part of the model is activated for each query. This is meant to speed up responses and lower operating costs (TechBriefly).

Google: Gemini Embedding 2 for text, images, video and audio

Google introduced Gemini Embedding 2 on March 10, its first natively multimodal model for embeddings. Embeddings turn content into numerical vectors, which let search find similar things by meaning rather than just by matching words (Google).

The new model places text, images, video, audio and PDF documents in a single shared space. It is available in public preview through the Gemini API and Vertex AI.

Meta delays its Avocado model

According to The New York Times on March 12, Meta has pushed back the release of its next major model, code-named Avocado, from March to May at the earliest. In internal tests it reportedly lagged behind models from Google, OpenAI and Anthropic in reasoning, coding and writing (The Decoder).

The model reportedly outperformed the older Gemini 2.5, but not Gemini 3. Meta’s leadership even reportedly considered temporarily licensing Gemini for its products, but no decision was made.

Products and Agents

Microsoft Copilot Cowork built with Anthropic

Microsoft introduced Copilot Cowork on March 9, an agent in Microsoft 365 Copilot that can handle an entire multi-step job from a single request, such as preparing a presentation for a customer meeting and scheduling time to prepare. It uses Anthropic’s Claude model for reasoning and runs inside the customer’s corporate environment (Fortune).

At the same time, Microsoft announced a new Microsoft 365 E7 bundle priced at 99 dollars per user per month, available from May 1. The company’s shares had previously fallen on investor concerns that AI agents will replace traditional enterprise software.

Google Maps gets Ask Maps and new navigation

Google launched Ask Maps in Google Maps on March 12, a feature that lets users search for places with complex questions in everyday language, such as where nearby they can work quietly on a laptop. For now it is available in the US and India on Android and iOS (Google).

The second new feature is Immersive Navigation, redesigned navigation with a three-dimensional view of buildings, overpasses and lanes. Google calls it the biggest change to navigation in Maps in more than a decade (TechSpot).

More product news

Google expanded Gemini in Docs, Sheets, Slides and Drive on March 10. Gemini can now create a document or spreadsheet from a prompt, rewrite entire passages in the style of the text, and show a summary of search results in Drive. The features are currently in beta for Google AI Pro and Ultra subscribers (The Next Web).
OpenAI added interactive visualizations for math and science to ChatGPT on March 10. For more than 70 topics, such as Ohm’s law or compound interest, users can change variables and watch how the result changes. The feature is available on all plans (OpenAI, Engadget).

Companies and People

Musk: xAI “was not built right” and is being rebuilt

Elon Musk admitted on X on March 12 that his AI company xAI “was not built right the first time around” and will be rebuilt from the ground up. The statement came amid departures of co-founders, only a minority of whom remained at the time (Sherwood News).

According to The Information, xAI also hired two executives from the startup Cursor, which develops an AI code editor for programmers. They are expected to help with a new attempt at a coding tool, an area where Grok lags behind the competition (The Decoder).