The takeaway
Compute politics set the agenda: export controls harden, Nvidia answers with board oversight, distillation industrialises, and inference routing consolidates into a single API layer.
Why it matters for builders
Routing layers are becoming infrastructure, compliance is a runtime concern, distillation erodes model moats, retention defaults change privacy assumptions, and capital keeps flowing to vertical AI stacks with proprietary data loops.
AI News Roundup: November Seventeen and the Compute Chokepoint
Overview: Today's AI news ran through one bottleneck: who gets to use advanced compute, and on what terms. Washington moved to legislate chip exports harder, Nvidia answered with board-level national security oversight, Reuters documented industrial-scale distillation of US models in China, Perplexity collapsed multi-provider inference into a single routing API, and Google quietly extended how long it keeps Gemini conversations even when history is off. Money kept flowing too: Isomorphic Labs closed a $5 billion round to push AI-designed drugs toward clinical trials.
AI Chip Curbs Harden as Huang Scandal Reaches Congress
The congressional push to block advanced AI chip sales to China picked up bipartisan momentum after former Nvidia CEO Jensen Huang's resignation over an affair with a business executive US intelligence believed could be a spy. Senate Minority Leader Chuck Schumer renewed calls from the floor for three bills, the AI Overwatch Act, the MATCH Act and the Chip Security Act, to be folded into the stalled National Defense Authorization Act. House China Select Committee Chair John Moolenaar said the incident "highlights the need for export controls." The AI Overwatch Act would codify existing bans on exporting top-tier chips to China, Iran, Cuba and North Korea, and give Congress review power over individual export licenses.
Nvidia Answers With a Security Committee, Not a Strategy
Under interim CEO Colette Kress, Nvidia formed a standing board committee to oversee national security policy, export control compliance and global trade risk. Chaired by director Suzanne Nora Johnson, it meets at least quarterly, and Nvidia's existing security and trade teams now report into it and to Kress. It landed hours before the company's third-quarter earnings call, which was expected to be dominated by the scandal rather than by demand. Huang has agreed to testify before Congress later this month.

Perplexity Puts Every AI Model Behind One Routing API
Perplexity shipped Orchestrator, an API that places OpenAI, Anthropic, Google and open-weight models behind a single endpoint with automatic failover, per-turn routing and unified billing. The practical shift is that model selection moves out of application code and into configuration. Teams stop hardcoding a provider and start declaring a policy: this route for latency, that route for cost, a fallback when a provider degrades.
China Industrialises Distillation of US Models as Curbs Tighten
Reuters documented more than 300 Chinese papers and filings describing US frontier models being distilled into domestic systems, even as Washington tightens rules on remote access to AI compute. Distillation converts borrowed answers into durable training data, which is why export restrictions on hardware and cloud access do not by themselves keep Chinese models behind. For teams building on frontier APIs, output provenance and provider terms become an engineering constraint.
OpenAI Puts NATO's Cyber Defenses on Daybreak Models
NATO now runs OpenAI's Daybreak cyber models across its institutions, consuming up to 300 billion tokens a month to hunt and remediate vulnerabilities. The interesting part is the operating model: value comes from high-volume AI triage confirmed by human maintainers, not from autonomous patching. That is the same division of labour most enterprises are still trying to define.
Google Extends Gemini Chat Retention to Two Months
Google is extending retention of temporary Gemini chats, and chats where history is off, from 72 hours to roughly two months, rolling out globally over the coming weeks. It cites safety, security, feedback processing and potential system failures, and says temporary chats are still not used for training. OpenAI and Anthropic both retain temporary or incognito chats for 30 days by default, so Google is now the outlier. The same update splits chat history and AI training into separate settings.
What to Watch Tomorrow
- Nvidia's Q3 earnings call fallout - guidance and the new board committee's mandate will be read as a referendum on export policy. TechCrunch
- NDAA amendment fight - whether the three AI chip bills survive conference is the real test of the Huang aftershock. CNBC
- Commerce Department interim rule - new controls on remote access to AI infrastructure outside the US are still being operationalised. TechCrunch
- Anthropic's Nasdaq debut - pricing is expected near or above the $70 to $78 range, with trading slated for November 19. CNBC
Builder Impact
- Route, do not bet. Perplexity's Orchestrator is the clearest signal yet that multi-provider inference is becoming infrastructure. If your agent hardcodes one model, a routing layer is now the cheaper design.
- Compliance is becoming a runtime concern. Chip export rules, remote-compute rules and distillation disputes all land on the same question for builders: where does this request execute, and whose terms apply.
- Distillation breaks the moat argument. Hardware curbs do not stop model convergence. Assume competitors can approximate your behaviour and compete on data loops, product and distribution instead.
- Retention defaults are a design input. Google's move from 72 hours to two months for history-off chats means privacy assumptions in your prompts, logs and demos need re-reading, not just your policy page.
- Vertical stacks are where the capital goes. Isomorphic's $5 billion round shows investors pricing domain-specific model plus simulator plus proprietary data, not another chat surface.
The Automation Brief
Read 5 AI stories instead of 50.
The essential moves in AI agents, models, automation and infrastructure — filtered for builders and operators, with the part that actually matters.
No noise. Unsubscribe anytime.
Editorial notes
Stefan Trbojevic
n8n Lab Editorial
17 November 2026
17 November 2026
Sources
AI disclosure: AI assisted with research and drafting. Factual claims are reviewed by an editor.




