Daily AI News Roundup – July 15, 2026
China’s humanlike AI rules hit today, TSMC just posted the biggest quarter in its history, and the AI chip race is heating up on both sides of the Pacific. Here’s everything that matters on this Wednesday before what could be the biggest day of the AI year.
1. China’s Humanlike AI Rules Take Effect — Doubao and Qwen Shut Down Millions of Agents
Today marks the enforcement date for China’s Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services — the world’s first national framework specifically targeting AI that simulates human personality, emotion, and sustained interaction. The rules were jointly issued in April by five Chinese government ministries and took effect today, July 15.
The immediate fallout is significant. ByteDance’s Doubao and Alibaba’s Qwen have both shut down their humanlike agent and companion features, affecting millions of users. Doubao is offering read-only data access through October 15 before permanent deletion; Qwen has given users no grace period at all — agent configurations and conversation histories are being permanently deleted with no migration path. Customer service bots, workplace assistants, and educational tools are exempt, as long as they don’t involve sustained emotional interaction. For everything else, Beijing has drawn a hard line.
2. TSMC Posts the Biggest Quarter in Its History
AI’s infrastructure bill keeps climbing — and TSMC is cashing the checks. TSMC reported record Q2 revenue of $39.62 billion, up 36% year-over-year — the largest quarter in the company’s nearly four-decade history. June alone was up 68% year-on-year, the biggest single-month jump ever recorded. First-half 2026 revenue hit $2.4 trillion NTD, reflecting 35.6% growth.
The driver is unambiguous: AI chip demand has overridden the consumer electronics cycles that once governed TSMC’s seasonality. Its leading-edge N3 process and CoWoS advanced packaging are both sold out through year-end. Full Q2 earnings with management guidance drop Thursday — the same day Gemini 3.5 Pro is expected to launch. Adding to the bullish signal, chipmaking equipment giant ASML lifted its annual sales forecast above Wall Street expectations today, citing AI demand and announcing a 30% expansion of its production capacity — easing some fears about equipment bottlenecks in the chip supply chain.
3. Anthropic Eyes Samsung for Custom 2nm Inference Chip
Anthropic is in early-stage talks with Samsung to co-develop a custom AI inference chip targeting Samsung’s 2-nanometer process, according to TechCrunch and Bloomberg. The company is still determining the processor’s specifications — what it should do, how powerful it should be, and how it would fit into a server — but the strategic logic is clear: custom chips can shave 10–30% off inference costs, and Anthropic is spending roughly $1.25 billion per month on compute.
The move is a direct response to OpenAI, which recently announced its own custom inference processor — dubbed Jalapeño — developed with Broadcom. Anthropic also recently poached a key OpenAI chip engineer, Clive Chan, who was one of the first people on OpenAI’s custom chip program. The AI chip arms race, once a battle between Nvidia and everyone else, is increasingly becoming a race to build proprietary silicon.
4. OpenAI’s GPT-5.6 Family Is Now Fully Available — Sol Leads on Coding
OpenAI’s GPT-5.6 family — a three-tier model lineup named Sol, Terra, and Luna — rolled out broadly on July 9 after a limited government-preview period. All three models are now available across ChatGPT, Codex, and the OpenAI API, as well as GitHub Copilot.
The breakdown: Sol ($5/$30 per million tokens) is the flagship, best suited for complex reasoning and long-horizon tasks. Terra ($2.50/$15) is the balanced everyday model, offering GPT-5.5-level performance at roughly half the cost. Luna ($1/$6) targets high-volume, latency-sensitive workloads. On benchmarks, Sol posts an 80 on the Artificial Analysis Coding Agent Index — a new state of the art, edging out Claude Fable 5 — and scores 53.6 on Agents’ Last Exam, a 13-point gap above its nearest rival. Sol’s position at the top of coding leaderboards will pressure both Anthropic and Google heading into Thursday’s expected Gemini 3.5 Pro launch.
5. All Eyes on Thursday: Gemini 3.5 Pro + Xi at WAIC + TSMC Earnings
Tomorrow, July 17, is shaping up as one of the most consequential single days in recent AI history. Three major events are converging:
- Gemini 3.5 Pro is widely expected to launch — Google’s rebuilt frontier model targeting a 2-million-token context window and Deep Think reasoning mode. No official confirmation yet, but the date has been reported across multiple outlets citing unnamed Google sources.
- Xi Jinping will deliver the keynote at the 2026 World Artificial Intelligence Conference in Shanghai — his first ever appearance at China’s flagship AI summit, with geopolitical implications for the global AI race.
- TSMC’s full Q2 earnings call will provide management guidance on the second half of the year and the chip supply outlook — critical reading for anyone tracking AI infrastructure investment.
We’ll have full coverage of all three. Check back Thursday.
That’s your AI news for Wednesday, July 15. Big day tomorrow — don’t miss it.