Question archive
Each day, we pose a question inspired by the daily news brief and answer it using our database of indexed AI/ML articles. Browse and search past questions below.
Why did Apple need Alibaba’s Qwen to ship Apple Intelligence in China across four device lines?
China rollout required a compliant local model plus on‑device feasibility: PrismML compression made Qwen small enough to run across Apple OSes, easing cloud/regulatory constraints.
2026-07-14Which project stage did New York’s 50‑MW data-center moratorium actually freeze: permits for new builds or already-approved sites?
Policy moratoria often target the discretionary environmental-permit gateway, freezing only new approvals while grandfathering projects past that review stage.
2026-07-13Where does Alphabet’s AI growth story collide hardest with margins: ads funding capex, or cloud absorbing it?
Alphabet’s AI margin pinch is structural: ads are the cash engine for centralized compute capex, while Cloud’s AI demand pulls that spend into COGS—squeezing both FCF and operating margin.
2026-07-12Why did Meta’s Muse Image trigger backlash despite using only public Instagram posts?
Backlash came from the consent architecture: default opt‑in let “public” photos become reusable AI training/replica material without notice, collapsing visibility into permission.
2026-07-11When 400 ex-Apple staff work at OpenAI, how does Apple separate “talent” from “trade secrets” in court?
In trade-secret cases, firms separate “talent” from “theft” by anchoring claims to breach vectors—unauthorized access, copying, and disclosure—rather than employee movement.
2026-07-10What does Musk’s Anthropic praise reveal about compute deals reshaping who gets called an “AI leader”?
“AI leadership” reputations now track control of compute pipelines: leasing hyperscale capacity (e.g., Colossus 1) structurally boosts capability signals and status.
2026-07-09Why did Microsoft’s emissions rise 25% even as it signed more renewable power contracts?
Renewable deals didn’t offset the 25% rise because emissions accounting changed: shifting from RECs to long‑term PPAs (and stopping credits) reclassifies electricity as higher‑carbon.
2026-07-08How does Russia’s “sovereign AI model” label turn data-localization into a control mechanism, not just privacy?
Data-localization becomes leverage when fused to a single AI regulator that gates deployment via high‑risk registration, conformity checks, and mandated transparency.
2026-07-07How does building an inference chip—rather than a training chip—actually reduce DeepSeek’s dependence on Nvidia and Huawei?
Inference is the scalable “deployment bottleneck,” so owning inference silicon cuts supplier risk even if training stays outsourced—and it sidesteps export bans aimed at top-end training GPUs.
2026-07-06Why does TeraWulf sell 50.1% of its joint venture while locking in a 20-year Anthropic lease?
Selling JV control turns a capex-heavy build into self-financing “project finance,” while a 20-year AI lease de-risks cashflows to fund HPC expansion.