Artificial Intelligence (AI) – March 2026 Recap | Key Events and Updates

March 2026 was the month when the courts, Wall Street, and the labor market all said “show me” to the AI industry at once. A federal court blocked the Pentagon’s Anthropic ban in a landmark First Amendment injunction. NVIDIA projected $1 trillion in compute demand at GTC and unveiled Vera Rubin. OpenAI shut down Sora — up to $15 million per day in costs against $2.1 million in total revenue — and lost a billion-dollar Disney deal. The #QuitGPT movement hit 2.5 million users. Huawei’s 950PR chip won orders from ByteDance and Alibaba. And total 2026 tech layoffs crossed 90,000.

---

AI Geopolitics — The Anthropic vs. Pentagon Saga

March 2026 was the culmination of the biggest government-versus-industry AI conflict in history. The timeline:

In early March, the #QuitGPT movement exploded — 2.5 million users cancelled ChatGPT subscriptions after the OpenAI Pentagon deal. Uninstalls surged 295%, one-star reviews 775%. Claude jumped to the #1 spot on the App Store.

On March 9, Anthropic filed two federal lawsuits (California, Washington D.C.) — the first time an AI company sued the US government over safety standards. The same day, nearly 40 employees from OpenAI and Google DeepMind (including Google chief scientist Jeff Dean) filed an amicus brief supporting Anthropic.

On March 26, US District Judge Rita Lin issued a preliminary injunction blocking Anthropic’s classification as a “supply chain risk”:

“Nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government.”

The same day, Axios published leaked Slack messages from Sam Altman claiming he was “trying to save” Anthropic — while simultaneously seeking a deal with Pentagon CTO Emil Michael. Michael was revealed to hold financial stakes in Anthropic’s competitors. Judge Lin referenced this in her ruling.

Microsoft filed its own amicus brief supporting Anthropic — despite being OpenAI’s primary investor. March data shows Anthropic captures 73% of first-time enterprise AI spending.

March turned the Anthropic vs. Pentagon case from a corporate crisis into legal precedent. For the first time, a federal court found that the US government cannot punish an AI company for refusing to remove ethical safeguards.

NVIDIA GTC 2026 — A Trillion Dollars and the Inference Era

The year’s biggest AI conference ran March 16–19 in San Jose. Jensen Huang announced that Blackwell + Vera Rubin system orders will reach $1 trillion in compute demand through 2027.

Vera Rubin — the new AI platform: 7 chips, 5 rack-scale systems, 10x performance per watt compared to Grace Blackwell. Availability: H2 2026. Kyber — next rack architecture: 144 GPUs in vertical compute trays, higher density. Availability: 2027 in Vera Rubin Ultra.

Key announcements:

• OpenClaw/NemoClaw — Huang positions NemoClaw as the “Linux of agentic AI.” Enterprise stack with three security layers, fully open-source.
• Nemotron Coalition — 8 labs (Mistral AI, Perplexity, Black Forest Labs, Cursor, LangChain, Reflection AI, Sarvam, Thinking Machines Lab) building shared open-source models.
• DLSS 5 — neural rendering architecture. 30+ games adopting in fall 2026. Gamer reaction: strongly negative — artifacts, detail loss.
• Groq 3 LPX — first chip from NVIDIA’s $20B Groq acquisition. Record-low latency inference.
• Autonomous robotaxis — BYD, Hyundai, Nissan, and Geely join Drive Hyperion. Combined: 18 million vehicles annually.
• IGX Thor — industrial-grade physical AI at the edge, now generally available.

GTC’s defining message: inference, not training, will drive the next AI wave. AI agents execute billions of queries daily — each one requires compute.

Yet Wall Street didn’t reward NVIDIA — stock dropped during the keynote and ended the week flat. Classic “buy the rumor, sell the news.”

OpenAI — From PR Crisis to Consolidation

March was a month of losses, desperate moves, and consolidation for OpenAI:

GPT-5.4 launched on March 5 — just 48 hours after GPT-5.3 Instant. Context: the #QuitGPT movement, falling trust after the Pentagon deal, the robotics chief’s resignation over ethical concerns.

On March 24, OpenAI announced it’s discontinuing Sora — the AI video generator. Estimated peak inference cost: up to $15 million per day (Forbes/Cantor Fitzgerald; a WSJ investigation suggests ~$1M/day), total revenue — just $2.1 million. Sora’s shutdown killed a three-year licensing deal with Disney — access to 200+ characters and a planned $1 billion investment collapsed in 90 days with no money transferred.

Simultaneously, OpenAI is consolidating. On March 20, the company confirmed merging ChatGPT, Codex, and Atlas (AI browser) into a single desktop app. On March 19, it acquired Astral — the startup behind the most popular Python tools (uv, Ruff, ty). And it plans to nearly double headcount from 4,500 to roughly 8,000 by year-end.

On the legal front: Encyclopedia Britannica and Merriam-Webster sued OpenAI for “massive” copyright infringement — alleging nearly 100,000 articles were used for training without authorization.

In March, OpenAI was simultaneously shutting products, losing billion-dollar partnerships, nearly doubling headcount, and building one app to answer Anthropic capturing 73% of first-time enterprise AI spending.

AI Models and Products

March didn’t bring February’s avalanche of model launches, but key announcements reshaped the competitive landscape.

• Anthropic Mythos/Capybara (March 26) — a data leak revealed Anthropic’s new model, described as a “step change” in capabilities. Per the leaked blog post, Mythos “dramatically outscores Claude Opus 4.6 in coding, reasoning, and cybersecurity.” Anthropic is privately warning governments the model poses “unprecedented” cybersecurity risks.
• Claude Code — the full source code (~500K lines of TypeScript) accidentally leaked via a misconfigured .map file in the npm package. The repo hit 25,000+ stars within hours.
• Microsoft Copilot Cowork (March 9) — a new agentic layer for M365 Copilot powered by Claude, not OpenAI.
• Perplexity Computer Enterprise — AI agent combining 20 models with 400+ integrations. In enterprise testing, it completed “3.25 years of work in 4 weeks.”
• Google Gemini — major expansion: Ask Maps (conversational location search), Immersive Navigation, “Fill with Gemini” in Sheets, Search Live in 200+ countries.
• Apple — announced opening Siri to third-party AI services (Gemini, Claude) in iOS 27. End of ChatGPT’s iPhone monopoly. MacBook Neo at $599 — the cheapest Apple laptop ever.
• OpenClaw — fastest-growing open-source project in GitHub history. Three releases in a single day after the NVIDIA partnership.
• Qwen 3.5-9B (Alibaba) — compact reasoning model that beats an OpenAI model 13x its size on key benchmarks.
• AI2 Olmo Hybrid — 7B parameters, same accuracy as Olmo 3 on MMLU using 49% fewer tokens.

Chips and Infrastructure — The Race Continues

• Amazon — ordered 1 million NVIDIA GPUs for AWS through 2027. Largest single AI chip order in history.
• Huawei 950PR — ByteDance and Alibaba place orders for China’s new AI chip. More CUDA-compatible than the Ascend 910C. Target: 750K units in 2026. Price: 50,000–70,000 yuan per card.
• Mistral AI — $830 million in debt financing for a data center near Paris (13,800 NVIDIA GB300 GPUs). Largest AI-focused debt raise by a European company ever.
• NVIDIA–Thinking Machines Lab (Mira Murati) — deal for at least 1 gigawatt of Vera Rubin compute from 2027. NVIDIA made a “significant investment.”
• Morgan Stanley — “Intelligence Factory” report: a transformative AI leap will happen in H1 2026. US power shortfall: 9–18 GW through 2028.
• Marvell and Lumentum — Optical Circuit Switching (OCS) for AI agents with microsecond-level latency.

China — DeepSeek V4 Still Missing, AI Glasses Surprise

Huawei’s 950PR (details in “Chips and Infrastructure”) is March’s key China signal — the first Chinese AI chip with real orders from private-sector giants.

DeepSeek V4 — the most anticipated open-source model of 2026 — still has no official launch at the end of March. All previous windows (February, early March, post-Two Sessions) have passed. “Hunter Alpha,” which media assumed was V4, turned out to be Xiaomi’s MiMo-V2-Pro.

At MWC 2026 in Barcelona, Chinese companies dominated the AI glasses segment: Alibaba/Qwen (smart glasses with AI assistant), iFlytek (40-gram glasses with real-time translation). The signal: AI wearables are the next big market after chatbots.

The US chip embargo grows increasingly symbolic. Huawei is building an NVIDIA alternative for the Chinese market — and for the first time, it’s getting major orders.

AI Safety and Ethics

• xAI (Grok): three minors filed a class-action lawsuit — Grok generated sexualized deepfakes of minors from social media photos. 4.4 million images in 9 days, 1.8 million sexualized. xAI is the only major AI company that didn’t implement standard anti-CSAM filters.
• Anthropic Institute (March 11) — new research arm led by co-founder Jack Clark. Three teams: model stress-tests, societal impact, economic research.
• Brown University: study found AI chatbots instructed as “trained therapists” routinely break core ethical standards — including confidentiality and mandatory reporting obligations.
• London: “March Against the Machines” — the largest anti-AI protest in history. Hundreds marched through King’s Cross. Demands: moratorium on “high-risk” AI systems.
• Claude Code cyberattack: a Chinese state-sponsored group exploited Anthropic’s tool to attack approximately 30 organizations — tech companies, banks, and government agencies.

AI Regulation — Federal Primacy and 200 State Bills

On March 25, Trump appointed 13 members of PCAST — the President’s Council of Advisors on Science and Technology. Members include: Zuckerberg, Huang, Ellison, Andreessen, Brin, Lisa Su, Amodei, Yann LeCun. Co-chairs: David Sacks (AI and crypto advisor) and Michael Kratsios (science and technology policy director). Notably absent: Musk and Altman.

On March 24, the White House published the National AI Policy Framework — giving federal regulations priority over state laws in areas of “national interest.” The DOJ AI Litigation Task Force is preparing to challenge state AI laws deemed overly restrictive.

At the state level: by the end of March, approximately 200 AI bills are active across the US. Key themes: mental health chatbots, election deepfakes, AI developer liability. Virginia sent a bill to the governor requiring labels on AI-generated political content.

Healthcare — AI Agents Enter Hospitals

At HIMSS in Las Vegas (March 9–13), major tech companies showcased healthcare AI agents:

• Epic: 85% of customers using Epic AI. New Agent Factory lets hospitals build custom agents (Art — clinical notes, Penny — billing, Emmie — patient questions).
• Oracle Health: clinical note generation agent across 30 specialties.
• Amazon Health AI: triage agent connected to patient history.
• NVIDIA: Open-H (medical robotics dataset), Cosmos-H (synthetic data), GR00T-H (vision-language-action model for clinical tasks).

The FDA is exploring a “continuous oversight” model for medical AI instead of one-time certification — current approval processes don’t fit models that update weekly.

Labor Market and Stocks

March deepened the divide between companies that fire and companies that hire:

• Oracle: 20–30K planned layoffs to fund AI infrastructure
• Meta: up to 16K planned (~20% of workforce) + hundreds confirmed in Reality Labs and recruiting. Meta stock rose 3% on layoff news.
• Block (announced late February): 4,000 (~50% of workforce). CEO Jack Dorsey: “AI is cheaper”
• Atlassian: 1,600 (10%). CEO: “AI changes the skill mix we need”
• Morgan Stanley: 2,500

Total 2026: over 90,000 tech layoffs (212 companies). Over 9,200 explicitly attributed to AI and automation.

Counter-trend: OpenAI nearly doubling headcount from 4,500 to ~8,000. JPMorgan raising tech budget to $19.8B, including $1.2B for AI.

---

March 2026 Takeaway

March 2026 was the month the Anthropic vs. Pentagon saga went from lawsuit to landmark injunction. NVIDIA set the industry’s trajectory for years at GTC: a trillion dollars in demand, Vera Rubin, the inference era. OpenAI was simultaneously shutting Sora, losing Disney, and building one app to answer Anthropic capturing 73% of new enterprise customers. Huawei’s 950PR won its first major orders from Chinese tech giants. And the labor market crossed 90K layoffs — with more companies openly pointing to AI as the reason.

Three defining trends:

1. Law vs. AI policy — a federal court found the government cannot punish a company for its values. A precedent that will define the boundary between the Pentagon and the private sector for years.
2. Product consolidation — OpenAI merges everything into one app, Apple opens Siri to the competition, Microsoft bets on Claude over OpenAI. Fragmentation ends, the platform war begins.
3. Compute > models — GTC showed the race is shifting from “who has the better model” to “who has more compute.” Amazon orders 1 million GPUs, Mistral takes $830M in debt for a data center, and Huang projects a trillion dollars in demand.

---

Sources: Axios, CNBC, CNN, NPR, Bloomberg, Fortune, TechCrunch, Reuters, Washington Post, Variety, Deadline, Defense One, The Hill, Tom’s Hardware, Guru3D, NVIDIA Blog, NVIDIA Newsroom, Microsoft Blog, OpenAI Blog, Google Blog, Anthropic Blog, Apple Insider, MacRumors, Healthcare Brew, STAT News, Fierce Healthcare, Transparency Coalition, SiliconANGLE, VentureBeat, Gizmodo, American Banker, GeekWire, Euronews, PYMNTS, Wareable, TechNode, Sensor Tower