Today in AI – 11-20-2025

AI compute racks stretching toward a skyline with U.S. and Gulf motifs, symbolizing the global buildout of GPU infrastructure

Key Stories (past 48 hours)

Nvidia’s $57B quarter puts AI compute in overdrive

Nvidia reported record Q3 FY26 revenue of $57.0B (+62% YoY), with data center revenue at $51.2B (+66% YoY). Guidance for Q4 FY26 is $65B. CEO Jensen Huang said “cloud GPUs are sold out,” and management cited visibility toward roughly $500B of Blackwell and Rubin platform demand through 2026. For practitioners, the prints underscore the continued constraint on GPU supply, likely affecting training/inference capacity, pricing, and AI project timelines through 2026.

$57.0B
Nvidia Q3 FY26 revenueSource: nvidia-q3-fy26-press

Read more analysis: Nvidia’s $57B Quarter Puts AI Compute in Overdrive.

U.S. OKs Blackwell exports to UAE and Saudi as HUMAIN targets 600k GPUs

The U.S. Commerce Department authorized exports of up to 35,000 Nvidia Blackwell (GB300) chips each to UAE-based G42 and Saudi-backed HUMAIN, under strict security and reporting conditions. Separately, HUMAIN announced plans to deploy up to 600,000 Nvidia GPUs across Saudi Arabia and the U.S. over the next three years. Expect accelerated “sovereign AI” build-outs, shifting compute gravity toward the Gulf and adding competition for hyperscaler capacity. Read more analysis: U.S. OKs Blackwell Exports to UAE and Saudi as HUMAIN Targets 600k GPUs.

Google ships Gemini 3 and Antigravity IDE, pushing agentic coding into the mainstream

Google introduced Gemini 3 Pro across its ecosystem (Gemini app “Thinking” mode and Search’s AI Mode) and launched Antigravity—an “agent‑first” coding environment that gives AI agents controlled access to the editor, terminal, and browser, while producing verifiable “Artifacts” (plans, screenshots, recordings). Early developer feedback is mixed, citing speed, tool‑use, and stability issues. For engineering leaders, Antigravity signals where IDEs are headed: multi‑agent workflows with auditable traces. Read more analysis: Google Ships Gemini 3 and Antigravity IDE to Speed Agentic Coding.

Microsoft Ignite: agent governance arrives (Agent 365, Work IQ; Security Copilot for E5)

At Ignite, Microsoft unveiled Agent 365—a control plane to authorize, monitor, and secure autonomous agents across Microsoft and third‑party ecosystems—and “Work IQ,” an intelligence layer for Copilot. Microsoft said Security Copilot will be included for Microsoft 365 E5 customers and rolled out new collaboration agents in Teams. The move formalizes enterprise agent governance and hints at near‑term ROI tracking requirements for autonomous workflows.

EU’s “Digital Omnibus” proposes easing and delaying parts of the AI Act; clarifies data use under GDPR

The European Commission proposed a “digital omnibus” to simplify and align digital rules. It would phase application of high‑risk AI provisions until necessary standards/tools are published (up to ~16 months of additional runway), pursue central oversight for some AI systems, modernize cookie consent, and clarify when personal data can be used to train AI under GDPR’s legitimate‑interest basis—with safeguards and a right to object. The package faces opposition from civil society groups who call it a rollback of protections, and still requires approval by Parliament and Council. For AI builders with EU exposure, this could rebalance compliance timelines and training‑data options if adopted.

OpenAI launches “ChatGPT for Teachers,” free through June 2027

OpenAI announced ChatGPT for Teachers, a secure workspace with admin controls for U.S. K–12 educators, free until June 2027. Early adopters include large districts like Houston ISD, focusing first on prep, IEP documentation support, and compliance workflows. The move exemplifies domain‑specific packaging to drive safe adoption in regulated sectors.

Emerging Trends

Sovereign AI accelerates—and shifts compute geography

Signals: U.S. approvals for advanced Nvidia exports to G42 and HUMAIN; HUMAIN’s plan for up to 600k GPUs; Nvidia’s cited $500B demand visibility through 2026. Impact: Expect more state‑backed AI “factories,” regional energy/land‑use deals, and multi‑sourcing beyond U.S. hyperscalers, tightening GPU markets and power availability for private workloads.

Agent control planes become enterprise table stakes

Signals: Microsoft’s Agent 365 for authorization, quarantine, telemetry, and policy around autonomous agents; Security Copilot bundled for E5; multi‑agent Teams workflows. Impact: CISOs and IT will demand agent inventories, audit trails, and kill‑switches before scaling agentic automation—driving convergence of SecOps, MDM/EDR, and AI governance.

IDEs evolve for agentic development with verifiable traces

Signals: Google’s Antigravity (Artifacts, multi‑agent orchestration) and Gemini 3 “Thinking” mode; mixed developer reports on tool reliability and latency. Impact: Expect rapid iteration on planning modes, tool‑use sandboxes, and provenance logs across IDEs. Early adopters should pilot on low‑risk repos while vendors harden execution controls.

Public preview
Antigravity availabilitySource: theverge-antigravity-2025-11-18

Regulatory recalibration in Europe

Signals: EU Digital Omnibus proposes delayed application for high‑risk AI requirements until standards exist, cookie simplification, and GDPR clarifications enabling AI training under legitimate interest with safeguards—amid sharp pushback from digital‑rights groups. Impact: If enacted, model developers gain clearer pathways for data use and more time to align with conformity assessments; risk teams should track amendments and Member State positions.

Sector‑specific AI packaging to spur adoption

Signals: ChatGPT for Teachers (free through June 2027) with FERPA‑aligned controls; district pilots aimed at documentation and planning tasks. Impact: Expect more domain‑tuned SKUs (healthcare, public sector, finance) that bundle controls, connectors, and templates to reduce procurement friction.

Conversations & Insights

“Is the AI bubble overblown?”—Nvidia print reignites the debate

Where: Financial media, X, investor notes. What’s being said: Strong revenue and guidance eased bubble fears; analysts argue the boom is still early as data center demand compounds. Skeptics caution about capex saturation and energy constraints. Takeaway: Budget for continued GPU scarcity and rising TCO; hedge with workload portability and efficiency work.

“Agentic coding, day 1 reality check”

Where: Developer subreddits and forums. What’s being said: Early users of Gemini 3 and Antigravity report both promise (multi‑agent testing, browser runs) and pain (timeouts, over‑eager code edits, inconsistent plans). Others stick with GPT‑5.1 or Claude Sonnet for complex refactors. Takeaway: Treat agentic IDEs as controlled pilots with robust versioning and CI gates; demand transparent plan/approve steps.

“EU Digital Omnibus: simplification or rollback?”

Where: Brussels policy circles, NGO statements, European press. What’s being said: The Commission frames proposals as simplification with safeguards; EDRi and others warn of weakened rights and delayed enforcement, especially around high‑risk AI and personal‑data use for training. Takeaway: Compliance leaders should scenario‑plan for both outcomes: a more permissive but still enforced regime, and potential amendments during Parliament/Council negotiations.

Quick Takeaways

  • GPU constraints persist into 2026; book capacity early and explore heterogeneous inference to control costs.
  • Expect more state‑backed AI build‑outs; watch export‑control conditions and data‑sovereignty clauses in cross‑border deals.
  • Pilot agent governance now (inventories, RBAC, telemetry, kill‑switches) before scaling autonomous workflows.
  • Treat agentic IDEs as “beta” in production contexts; require plan/approve stages and tight repo permissions.
  • Track the EU Digital Omnibus closely; it could alter AI training data strategies and compliance timings if adopted.

Sources