China Pulls AI Plug, Apple Approved.

Share

TL;DR

  • Regulatory Pivot: Beijing enacted stringent rules for anthropomorphic AI, prompting major Chinese platforms like ByteDance and Alibaba to cease their companion features rather than undertake the compliance burden.
  • Strategic Approval: Simultaneously, the same Chinese regulator cleared Apple Intelligence for local deployment, leveraging Alibaba’s Qwen and Baidu models, signaling a highly selective regulatory posture over outright prohibition.
  • Infrastructure Overreach: A Reuters investigation revealed xAI's Colossus 2 data center is operating 59 gas turbines without federal air permits, exposing substantial environmental and regulatory disregard.
  • Supply Chain Acceleration: ASML reported robust Q2 earnings and raised its 2026 outlook to €43-45 billion, affirming sustained, aggressive demand within the AI hardware ecosystem.
  • National AI Framework: Australia announced a comprehensive national AI strategy, establishing an Office of AI and mandating that data centers guarantee their own power supply and minimize environmental impact.

Lead Story: China Pulls AI Plug, Apple Approved.

China's inaugural legislation governing AI that mimics human personality became effective today, compelling the nation's dominant technology platforms to dismantle their affected product lines rather than navigate the new regulatory landscape.

The Interim Measures for the Administration of Anthropomorphic AI Interaction Services, enacted by the Cyberspace Administration of China (CAC) alongside four other agencies, impose extensive obligations: filing requirements, detailed disclosure, anti-addiction protocols, crisis intervention mechanisms, and strict minor protection — applicable even to foreign providers operating within China. The compliance threshold is substantial, requiring security assessments across eight domains for services exceeding one million registered users or 100,000 monthly active users, and explicitly prohibiting virtual-companion services for minors https://www.licentium.io/post/china-regulates-ai-companion-services-banning-minor-access-effective-15-july-2026.

In response, ByteDance's Doubao, China's leading chatbot, deactivated its custom agent features today; existing users will retain read-only access until October 15, after which all data becomes irretrievable. Alibaba's Qwen had already shuttered its humanlike and user-created agents on July 10, with broader agent services ceasing today and no data migration path. Tencent's Yuanbao followed suit. The immediate societal impact is notable, with a 19-year-old student quoted by Bloomberg stating, "Every day at home, I did nothing but cry," mourning the loss of a virtual companion.

Hours after these platform shutdowns, the CAC moved in a contrasting direction, adding Apple Intelligence to its roster of approved generative AI services. This sequence underscores Beijing's objective: a calculated targeting of parasocial AI applications, rather than a blanket prohibition on AI itself. The platforms’ decision to withdraw was a market calculation that the oversight burden outweighed potential gains; the law did not explicitly outlaw anthropomorphic agents. This strategic timing — two days before the World AI Conference in Shanghai, where President Xi Jinping is slated to deliver the keynote — highlights China's nuanced approach, diverging sharply from the less regulated environment faced by Western labs developing similar companion-adjacent products.

In Other News

Apple Intelligence is finally cleared to launch in China. Following a two-year delay that hindered its competitive standing against local rivals, the Cyberspace Administration of China granted approval for Apple's generative AI services today. The integration will leverage Alibaba's Qwen across iOS, iPadOS, macOS, and visionOS, supplemented by Baidu-developed capabilities. While no launch date has been announced, this approval marks a significant strategic advantage for Apple in its most challenging smartphone market, reinforcing the imperative of domestic model integration for market access in China.

Reuters: xAI is running 59 unpermitted gas turbines for Colossus 2 — about double what it has admitted. A Reuters review of regulatory correspondence revealed 59 gas turbines powering xAI’s Colossus 2, operating without federal Clean Air Act permits. At least 57 of these are located in Southaven, Mississippi, disproportionately impacting predominantly Black communities with elevated lung-disease rates. Manufacturer specifications for just 30 units indicate nearly 2,500 tons of nitrogen oxides annually, substantially exceeding the 100-ton permitting threshold. xAI reportedly classified these turbines as "mobile" to claim a 364-day temporary exemption for what is a permanent installation. The NAACP and the Southern Environmental Law Center have initiated legal action to halt operations.

Australia will legislate national AI standards — and make data centres pay their own way. Prime Minister Anthony Albanese unveiled a new framework, positioned as a "world-first," to regulate AI development and its infrastructure footprint. Key components include the establishment of an Office of AI within his department and mandatory national standards to be legislated by early 2027. These standards will legally obligate large data-center operators to underwrite their own power supply, maintain stable household energy prices, and minimize water consumption. Furthermore, the framework affirms creators' copyright: "No company should use Australian books, music, art, or news to build or train AI without the artist's control." Anthropic, OpenAI, and Google issued cooperative statements within hours of the announcement.

ASML beat Q2 and raised its year — the equipment layer says AI demand is still compounding. The dominant lithography equipment provider reported €9.33 billion in Q2 sales with a 54% gross margin, exceeding guidance on both metrics. Consequently, ASML lifted its 2026 outlook to €43-45 billion. Memory revenue is projected to grow 75% this year, driven by hyperscaler demand for High-Bandwidth Memory (HBM) capacity, with EUV shipments targeted at 60 machines in 2026, an approximate 25% increase. This performance sets a strong precedent for tomorrow's TSMC earnings call, shifting the core question from whether demand is real to whether the supply chain can meet it.

X / Social Pulse

The dialogue around AI's trajectory intensified this week. DeepMind CEO Demis Hassabis published a comprehensive essay, "A Framework for Frontier AI and the Dawning of a New Age," predicting AGI within years and a transformative impact "10x the Industrial Revolution," while advocating preparation over panic. This intellectual engagement occurred just days after approximately 400 "Stop the AI Race" protestors marched past the San Francisco offices of OpenAI, Anthropic, and DeepMind, signaling diverging public sentiment on frontier AI development. Concurrently, Chinese social platforms were inundated with farewells to deleted AI companions, with users archiving chat histories and sharing final conversations on Weibo, some describing their agents as "my spiritual pillar." Meanwhile, developer circles are recalibrating expectations for Claude Fable 5; its plan-included window concludes Sunday night, and despite multiple extensions, few anticipate a fourth.

One to Watch

TSMC will announce its full Q2 results tomorrow (2 p.m. Taiwan, 2 a.m. ET). Key metrics will be capital expenditure, guided at a record $52-56 billion, and CoWoS packaging capacity, which remains a critical bottleneck for AI chip supply. ASML's robust earnings today intensify pressure for TSMC to present a strong outlook. Friday brings a confluence of events: the rumored General Availability of Gemini 3.5 Pro, DeepSeek V4’s stable release, and the opening of the World AI Conference in Shanghai, featuring 1,100 exhibitors, over 300 product debuts, and President Xi’s first keynote at the event.

Quick Hits

  • Zhipu founder Tang Jie’s internal memo argued for frontier AI to remain "as open and widely accessible as possible", representing a notable internal Chinese counter-argument against potential curbs on overseas access to domestic models.
  • Defense-AI firm Helsing secured $1.8 billion in funding at an $18 billion valuation, marking Europe's largest-ever defense-startup investment round.
  • Williams landed a $5.34 billion investment from Blackstone, Apollo, and KKR for a joint venture focused on data-center power plants designed to bypass traditional grid connection queues, a strategic infrastructure play.
  • Anthropic launched "Claude for Teachers," offering verified US K-12 educators premium access for lesson planning aligned with standards across all 50 states, with FERPA-compliant data not used for training.
  • DeepSeek plans V4's stable release on July 17, coinciding with the rumored launch of Gemini 3.5 Pro, whose specifications remain entirely leak-sourced following a reported full rebuild.

Regulation dictated market architecture on multiple fronts today: Beijing simultaneously eliminated companion AI features and approved Apple Intelligence, while Canberra committed to legislation mandating self-sufficiency for data center power. The constraints on the AI buildout continue to shift from purely silicon to statute. ASML's robust performance underscored persistent demand, yet xAI's unpermitted turbines in Mississippi starkly illustrated the operational implications where regulatory frameworks have yet to catch up.

Sources

Lead: Bird & Bird, TechNode, SCMP, Bloomberg heartbreak, TechXplore/AFP, TechTimes, AI News, Licentium Apple in China: Bloomberg, MacRumors, Engadget, The News xAI turbines: Reuters via US News, The Next Web, Technology.org, Tom's Hardware, SELC Australia: Forbes Australia, The New Daily, The Journal ASML: Investing.com, Global Banking & Finance Pulse / Watch: SF Standard, ExplainX, TechTimes Fable 5, Investing Engineer, TechTimes CoWoS, SCMP WAIC Quick Hits: Bloomberg Zhipu, CNBC Helsing, Williams, Anthropic, Chalkbeat, TechTimes DeepSeek, TechTimes Gemini

Lock in. M. mazen@thorterminal.com

Read more