Top story: OpenAI’s $110B mega‑round
OpenAI has closed a record‑breaking 110 billion dollar funding round, valuing the company at about 730 billion dollars and making it one of the most valuable private firms in history. Amazon is investing 50 billion dollars, while Nvidia and SoftBank are each contributing 30 billion dollars, with additional investors expected to join as the round remains open. OpenAI says the money will fund massive compute and infrastructure build‑out, with long‑term plans for hundreds of billions in AI infrastructure spending through partners like Amazon Web Services and Nvidia.
Frontier models and research breakthroughs
Google DeepMind’s AlphaEvolve, a Gemini‑powered coding agent, is being credited with advancing theoretical computer science by discovering new mathematical structures and more efficient algorithms, including improvements to long‑standing open problems. Deployed across Google’s infrastructure, AlphaEvolve’s algorithms have already recovered about 0.7 percent of Google’s worldwide compute capacity on an ongoing basis and sped up a key matrix‑multiplication kernel in Gemini’s architecture by roughly 23 percent.
Google’s Gemini 3.1 Pro is now one of the strongest frontier models, scoring 77.1 percent on the ARC‑AGI‑2 reasoning benchmark (more than double Gemini 3 Pro) and 94.3 percent on the GPQA Diamond graduate‑level science benchmark, reportedly the highest score yet. The model offers a one‑million‑token context window, enabling analysis of very large codebases and document collections in a single run.
China’s DeepSeek V4 is emerging as a major open‑source challenger: it is a trillion‑parameter Mixture‑of‑Experts model with about 32 billion active parameters per token, a one‑million‑token context window, and innovations like Engram conditional memory and sparse attention aimed at cutting inference cost dramatically. Analysts note that V4’s pricing undercuts Western proprietary models by a wide margin, potentially reducing AI adoption costs by 50–80 percent in some enterprise use cases.
Major product and assistant launches
OpenAI’s GPT‑5.4, launched on March 5, combines strong reasoning, coding, and agentic capabilities in a single family of models with up to a one‑million‑token context window. It adds native computer‑use abilities and a tool‑search mechanism that reportedly reduces token usage by around 47 percent on complex tasks, positioning it as a flagship model for both coding and general automation workloads.
Microsoft has introduced Copilot Cowork, an enterprise AI “coworker” that can run long‑lived, multi‑step tasks across files and applications, built in close collaboration with Anthropic and leveraging Claude’s agentic capabilities. Cowork uses Microsoft’s Work IQ layer to reason over an organization’s documents, emails, and other assets, and is currently in testing with a limited set of customers via the Copilot Frontier program.
On the startup and vertical side, new launches like Zest AI’s CU Lending Collective (bringing credit‑risk models to small credit unions) and Basis’s agentic accounting platform (now backed by a 100‑million‑dollar Series B at a 1.15‑billion‑dollar valuation) show how specialized “agent” products are moving into traditional sectors such as lending, audit, and tax. Analysts see these as early examples of domain‑specific AI agents becoming embedded in everyday back‑office workflows.
Policy, regulation, and AI governance
In the United States, a key March 11 deadline from the White House AI executive order requires the Commerce Department to publish an evaluation of state AI laws, flagging those it considers “onerous” or in conflict with federal policy—especially laws that require AI models to alter “truthful outputs” or mandate disclosures that may raise First Amendment issues. At the same time, the Federal Trade Commission is due to issue a policy statement on how Section 5 of the FTC Act (unfair and deceptive practices) applies to AI models, including when state requirements to change outputs might be preempted.
State‑level activity is accelerating: recent legislative trackers highlight wide‑ranging bills such as Florida’s proposed “AI Bill of Rights” (covering chatbot use, minors’ access, and data selling), workers‑comp rules that prohibit AI‑only claim decisions, and multiple bills regulating AI in mental health, insurance, deepfakes, pricing, and professional services. Several states are also advancing “AI non‑sentience” and “no legal personhood” bills to explicitly bar AI systems from being treated as legal persons.
Globally, new frameworks are coming into force in 2026, including California’s AI Transparency Act and Generative AI Training Data Transparency Act, which require labeling of AI‑generated content, public summaries of training datasets, and controls around detection and provenance tools. South Korea’s Basic AI Act, Japan’s principles‑based AI law, and Vietnam’s digital technology law all add transparency, labeling, and human‑oversight expectations for high‑impact AI systems, often with extraterritorial reach when foreign systems affect their citizens.
Hardware, infrastructure, and Nvidia GTC
Nvidia’s GTC 2026 conference—often described as the “Super Bowl of AI”—kicks off today in San Jose and runs from March 16 to 19, with investors and developers watching for major announcements on next‑generation GPUs, AI “factories,” and software for physical AI. Jensen Huang has previewed a “five‑layer” AI stack (energy, chips, infrastructure, models, applications) that Nvidia aims to serve end‑to‑end, underscoring its shift from pure chipmaker to full AI‑infrastructure platform.
The upcoming Rubin GPU platform—an evolution of Nvidia’s AI chips—has been flagged by analysts as potentially reducing inference token costs by up to tenfold and cutting the number of GPUs needed to train Mixture‑of‑Experts models by about four times. Nvidia’s sovereign‑AI business (selling full AI data‑center stacks to governments) reportedly tripled year‑over‑year to more than 30 billion dollars in fiscal 2026, driven by countries such as Canada, France, the Netherlands, Singapore, and the United Kingdom.
Notable industry trends to watch
Agentic AI is rapidly becoming the mainstream narrative: frontier models like GPT‑5.4, Gemini 3.1 Pro, Claude Opus 4.x, and DeepSeek V4 are all being positioned not just as chatbots but as the reasoning engines behind agents that can use tools, operate computers, and manage long‑running workflows. Microsoft’s Copilot Cowork and Anthropic’s Claude Cowork concept, in particular, signal a shift from single‑prompt interactions to ongoing delegated work inside enterprise systems.
Cost and openness are another major theme: DeepSeek V4’s trillion‑parameter open‑source design and aggressive pricing, along with competitive offerings like Gemini 3.1 Pro and lower‑cost models from various vendors, are pressuring U.S. labs to compete on price as well as quality. At the same time, new regulations on transparency, provenance, and safety—especially around deepfakes, child protection, and high‑risk applications—are pushing companies to build stronger governance, disclosure, and human‑oversight layers into their AI products from day one.
