# AI weekly: 9–15 November 2026

Every article published from 9–15 November 2026, newest first, with a short summary of each.

## This week's articles

### [Voters back data center limits in 40 of 41 local measures: what it means for AI capacity](/articles/data-center-backlash-ai-capacity)

10 November · Trends

US voters backed data center limits in 40 of 41 local ballot measures. Why towns push back, how power and grid costs work, and what it means for AI compute.

### [The new LLM serving stack: cache-aware routing, continuous batching and expert pruning](/articles/llm-inference-serving-stack-cache-routing)

10 November · Trends

Buoyant's Kinference, Fireworks Forge and AWS research on MoE expert pruning show where LLM serving costs now go. How KV caches, routing and batching work.

### [Revolut, AMD and Mistral show coding agents moving to self-hosted open-weight models](/articles/self-hosted-open-weight-coding-agents)

10 November · Trends

Revolut, AMD and Mistral point coding agents towards self-hosted open-weight models. How the serving stack, routing, evals, licences and the cost maths work.

### [OpenAI's Control Room and the case for a real kill switch on AI workloads](/articles/ai-workload-kill-switches-control-room)

10 November · Trends

OpenAI built Control Room to stop research AI jobs within 35 seconds. How workload inventories, halts at each layer, safety cases and stop drills work.

### [METR finds AI roughly doubling Anthropic's research pace as its IPO filing flags a threshold](/articles/ai-speeding-up-ai-research)

10 November · Trends

METR estimates Anthropic's capability progress has roughly doubled, mostly because of AI help. How AI R&D acceleration is measured and why thresholds matter.

### [LLMjacking and shadow AI: how stolen keys and pasted data are leaking out of AI tools](/articles/llmjacking-and-shadow-ai-data-leaks)

10 November · Trends

Fireworks AI widened its breach disclosure as hospitals said staff pasted patient data into AI tools. How LLMjacking works and how to stop both kinds of leak.

### [GR00T 2 and Isaac Lab 3.0: robots that learn a new task from one demonstration](/articles/robot-foundation-models-one-demonstration)

10 November · Trends

NVIDIA unveiled GR00T 2 and Isaac Lab 3.0 at CoRL 2026 as ANYbotics shipped an explosion-safe robot. How robot models learn, and what deployment takes.

### [Confidential inference reaches TPUs with Google's TPU 8tx, TPU 8i Max and Cloud Pathways](/articles/confidential-inference-tpu-8tx-pathways)

10 November · Trends

Google added TPU 8tx for confidential inference and TPU 8i Max, and made Cloud Pathways GA. How confidential computing reaches accelerators, and its limits.

### [Companies cap AI spend per employee while model prices fall: a practical guide to AI FinOps](/articles/per-employee-ai-budgets-and-finops)

10 November · Trends

Apple and Accenture reportedly capped staff AI spending as Anthropic, Meta and Fireworks cut prices. How to measure agent costs and set budgets that work.

### [Clinical AI in late 2026: measured savings, shared liability and an AI scribe shakeout](/articles/clinical-ai-savings-liability-scribe-shakeout)

10 November · Trends

Health systems report AI savings, the AMA spreads clinical AI liability, and scribe firms Heidi and Abridge cut staff. How to read the numbers and the risks.

### [Transformer 2.0? Titans, MIRAS and memory that learns at test time](/articles/transformer-2-titans-miras)

9 November · Research

A plain guide to Google's Titans, MIRAS and Nested Learning, how memory that learns at test time works, and what it means for time series and tabular models.

### [Large tabular foundation models: how in-context learning works](/articles/large-tabular-foundation-models-in-context-learning)

9 November · Research

How large tabular foundation models predict on a new table without training on it, by reading labelled rows as context: theory, models, limits and practice.

## More weeks

[← Week 45, 2026](/weekly/2026-w45) · [Week 47, 2026 →](/weekly/2026-w47) · [All weeks](/weekly)
