Beijing Claws Back Meta's Manus.

Mazen Al Sabbagh

— 5 min read
Share

TL;DR

  • Beijing's clawback: Meta unwound its $2B acquisition of AI agent startup Manus under Beijing's directive, with Tencent now orchestrating a consortium of original Chinese investors to reclaim the asset at its prior valuation.
  • Compute strain: OpenAI's GPT-5.6 launch doubled prior traffic peaks, prompting usage resets for 500,000 accounts and a second extension for Anthropic's Claude Fable 5, underscoring systemic capacity constraints.
  • Meta's infrastructure play: Meta's stock rose 7% on plans to expand computing capacity to 14 gigawatts by 2027, commencing production of its Iris AI chip in September, and committing $10B to a new Canadian data center.
  • Memory scarcity alert: SK Hynix CEO Kwak Noh-jung forecasts the memory industry's most severe supply shortage in 2027, with AI-driven demand projected to outstrip supply beyond 2030.
  • Agent standard challenge: A coalition including Google and Microsoft is backing a new standard for AI agent integration, directly contesting Anthropic's Model Context Protocol, which has emerged as an incumbent default.

Lead Story: Beijing Claws Back Meta's Manus

Beijing has compelled Meta to reverse its $2 billion acquisition of Singapore-based AI agent startup Manus, with Tencent now leading a consortium to repurchase the company. The move effectively repatriates a critical AI asset, signaling intensified state control over strategic technology (Bloomberg, The Decoder, DigiTimes).

The consortium, comprised of Manus's original Chinese investors—Tencent, HSG, and ZhenFund—alongside the management team, will execute the buyback at the initial $2 billion valuation (Tech Startups). While Tencent is positioned to become the largest shareholder, Manus asserts it will maintain operational independence (Tom's Hardware).

Chinese regulators initially blocked the deal in April, citing investment rule violations and characterizing Meta's purchase as a "conspiratorial attempt to hollow out China's technology base." Founder Xiao Hong faced an exit ban as part of the enforcement. Meta commenced the unwinding process in June under Beijing's demand (TechCrunch), culminating this weekend in the full reversal and the asset's return to Chinese ownership at the original price.

This marks Beijing's first forceful reversal of a completed U.S. acquisition of an AI startup, mirroring Washington's own recent efforts to block outbound technology transfers. It establishes a clear precedent for state intervention in cross-border AI M&A.

This specific clawback operates within a broader context of tightening geopolitical control. Reuters reports China's Ministry of Commerce is engaging major tech firms like Alibaba, ByteDance, and Z.ai on restricting overseas access to advanced Chinese AI models, both proprietary and open-weight (Fortune, Forbes). Coincident with the U.S. easing license requirements for chip exports to the UAE, these parallel actions from Washington and Beijing underscore a global consensus: AI assets are fundamental instruments of statecraft.

In Other News

Usage Limits Strain Model Providers. The launch of GPT-5.6 propelled OpenAI traffic to double its previous peak, necessitating repeated usage-limit resets, a temporary lift on usage caps for premium subscribers, and new inference optimizations for its Sol model (Neowin). OpenAI subsequently initiated a fresh reset for approximately 500,000 ChatGPT Work and Codex accounts, addressing a reset bug and acknowledging suboptimal execution on the ChatGPT Work launch (Windows Report, The Decoder). Concurrently, Anthropic extended free Claude Fable 5 access for a second time, now through July 19, with boosted Claude Code limits; post-extension, Fable 5 transitions to a prepaid credit model (BleepingComputer, Implicator). This underscores ongoing infrastructure challenges across leading AI platforms.

Consortium Challenges Anthropic's Agent Protocol Dominance. Major enterprise players, including Google, Microsoft, Salesforce, Snowflake, and ServiceNow, have reportedly aligned to support a unified technical standard for integrating AI agents into business software (The Information). This initiative directly counters Anthropic's Model Context Protocol (MCP), which has quietly gained traction as the industry's de facto default over the past 18 months. While publicly framed as complementary to the Agentic Resource Discovery (ARD) spec unveiled in June (Crypto Briefing), industry analysis suggests this coalition aims to fundamentally disrupt MCP's market position.

Meta Accelerates Infrastructure Investment. Meta's stock saw a 7% surge following an internal disclosure detailing plans to nearly double its total computing capacity to 14 gigawatts by 2027, with 2026 AI infrastructure spending potentially reaching $145 billion (Motley Fool). Its proprietary AI chip, Iris—co-designed with Broadcom and produced by TSMC—is slated for September production (CNBC). Furthermore, Meta allocated $10 billion for a one-gigawatt data center in Alberta, marking its first Canadian facility and its 33rd globally (Bloomberg), reinforcing its commitment to independent, large-scale compute.

SK Hynix CEO Predicts Unprecedented Memory Shortage. Kwak Noh-jung, CEO of SK Hynix, warned Reuters that 2027 is poised to be the memory industry's most challenging year from a supply perspective, projecting AI demand to exceed manufacturing capacity well beyond 2030 (Reuters via US News). This stark forecast, issued on the day of the company's $26.5 billion Nasdaq debut, aligns with similar concerns articulated by Samsung and Micron (Wccftech), signaling a fundamental supply-demand imbalance driven by AI's compute requirements.

X / Social Pulse

The public dispute between Elon Musk and Sam Altman escalated into its fourth day, with Musk tweeting "Scam Altman strikes again" in reference to Apple's lawsuit. Altman retorted, "homeboy, you're the one selling public market investors on short-term space datacenters," underscoring the ongoing, personal rivalry between the two tech figures (CNBC, Quartz). Developers voiced frustration over "access whiplash" concerning Claude Fable 5, as its latest extension materialized hours after the previous deadline, marking the second such last-minute reprieve within a week (Digital Applied). Meanwhile, SK Hynix CEO Kwak Noh-jung's "worst year ever" memory shortage prediction catalyzed significant discussion across hardware forums, with Slashdot threads actively debating the potential for AI infrastructure demand to impact consumer RAM and SSD pricing (Slashdot).

One to Watch

Gemini 3.5 Pro Launch. Google's flagship model is reportedly slated for release on July 17, following a comprehensive architectural rebuild that delayed its June debut. The updated version features a 2M-token context window and a "Deep Think" reasoning layer (BigGo, TechTimes). Despite the recent departure of four senior Gemini researchers to Anthropic (TechCrunch), the launch aligns with a uniquely opportune market. Siri's forthcoming autumn relaunch relies on Gemini, and OpenAI is currently contending with unprecedented demand, positioning Google to capitalize on significant market receptivity.

Quick Hits

  • SambaNova Secures $1B Funding: SambaNova completed a $1 billion Series F round at an $11 billion valuation, led by General Atlantic, with JPMorgan Chase integrating its inference systems on-premises; a 2027 IPO is reportedly under consideration (TechCrunch, CNBC).
  • OpenAI Responds to Apple Lawsuit: OpenAI issued its initial rebuttal to Apple's trade secret litigation, stating, "We have no interest in other companies' trade secrets." Fortune concurrently revealed an internal "LOL" email from OpenAI that allegedly contributed to the dispute's origins (MacRumors, Fortune).
  • China's Companion-AI Regulations Enact: China's Interim Measures for Anthropomorphic AI Interaction Services, governing companion AI, become effective Wednesday, July 15 (AIToolsRecap).
  • Meta's Watermarking Flaw: Meta conceded its Content Seal detector struggles to reliably identify heavily cropped outputs from its Muse Image AI, thereby compromising its claims of robust, crop-resistant watermarking (Digital Trends).
  • OpenAI IPO Outlook: OpenAI is reportedly aiming for a September public debut at a private valuation between $730 billion and $850 billion, with Goldman Sachs and Morgan Stanley managing the offering, placing it on a comparable timeline to Anthropic's market entry (Crypto Briefing).

The market's focal point today was control: Beijing demonstrated its capacity to reclaim a strategic AI asset from a U.S. technology major, while a consortium of enterprise leaders moved to challenge Anthropic's emerging dominance over AI agent protocols. These power plays unfolded against a backdrop of escalating compute demand—evidenced by weekend usage overloads, a dire memory supply forecast, and Meta's expansive $145 billion infrastructure commitment—underscoring that access to critical resources remains the fundamental determinant of competitive viability.

Sources

Lock in. M. mazen@thorterminal.com

Read more