China's MiniMax Prepares New 2.7-Trillion-Parameter Model "M3 Pro" — Scaling Race Heats Up Further
According to a report by The Information, Chinese AI company MiniMax is planning to release its next-generation large language model, the "M3 Pro." The model is said to reach 2.7 trillion parameters, a figure that highlights how competition over model scale is intensifying further among Chinese players such as DeepSeek and Alibaba.
Chinese AI companies have been rolling out large-scale models one after another over the past few months. In addition to DeepSeek's V4 series and Tencent's Hy3, MiniMax's plan to introduce an even larger model shows that the race for frontier-class performance continues on both the U.S. and Chinese sides. While parameter scale does not necessarily translate directly into higher performance, the 2.7 trillion figure symbolizes the momentum of the current scaling race.
Former DeepMind Executive Verity Harding Warns of the Dangers of an AI Arms Race
Verity Harding, a former Google DeepMind executive and expert on AI ethics, warned in an interview with WIRED that the situation now surrounding AI development represents "the worst-case scenario taking shape."
Harding in particular pointed to the U.S. government's nation-state-like posture toward AI as a cause for concern. If AI development proceeds within a nationalist framework, she argues, security risks will grow and international coordination will become more difficult. The view that the AI development race is increasingly taking on the character of an "arms race" has been voiced by a number of experts, but the fact that someone with inside knowledge of DeepMind is publicly sounding the alarm carries significant weight.
While Harding is not definitively pessimistic about the future of AI, she stressed that changing the current course will require a shift in mindset on the policy front.
French Competition Authority Orders Meta to Negotiate Copyright with News Media
France's competition regulator (Autorité de la concurrence) on July 8 ordered Meta Platforms to enter into negotiations with news organizations over copyright licensing fees. The measure came in response to two complaints alleging that the company abused its dominant position.
Meta is obligated, within 15 days, to provide news organizations with the information necessary for calculating copyright fees. Criticism had long persisted that the company was using news content on its platform without paying appropriate compensation.
This decision is likely to influence debates over the value of content and copyright protection in the AI era. As news articles are increasingly used as training data for generative AI, the ruling is drawing attention as a precedent for rebalancing the power relationship between platforms and content producers.
Apple and Broadcom's U.S.-Made Chip Deal to Exceed $30 Billion
Apple has expanded its component procurement agreement with Broadcom, bringing the total value to more than $30 billion. The company says it will procure over 15 billion U.S.-made chips and support hundreds of jobs. It will also help fund the upgrade of production equipment at Broadcom's Colorado plant.
Moves to strengthen semiconductor manufacturing in the United States are accelerating in parallel with surging AI infrastructure demand. Apple's newly announced procurement expansion materializes an earlier commitment, but it can also be read in the context of supply chain restructuring aimed at meeting rapidly growing AI chip demand. Together with U.S. investments by TSMC and Samsung, "Made in USA" semiconductors are gradually becoming a reality.
Singapore's Temasek Surpasses $400 Billion in Net Assets, Boosted by AI Investments
Temasek Holdings, Singapore's state-owned investment company, announced that its net portfolio value reached 518 billion Singapore dollars (approximately $401 billion) as of the end of March. Raising its allocation to AI and U.S. markets contributed to the improvement in returns.
Total shareholder return for the one-year period was 10.5% in local currency terms. The company attributes the direct improvement in returns to its shift toward AI-related investments, a sign that institutional investors increasingly view the AI sector as a major source of gains.
At a time when some voices warn of an AI bubble, a real-world investment result showing the AI sector lifting a portfolio could have a measurable effect on market sentiment.
Anthropic Expands Eligibility for "Claude for Open Source" Program for OSS Developers
On July 8, Anthropic expanded the scope of its "Claude for Open Source" program for open source software (OSS) developers. Eligible developers will receive the top-tier subscription plan, "Claude Max 20x," free of charge for six months.
The strengthened support for OSS developers can be seen as a move to prioritize building relationships with the community as AI tools become deeply embedded in development workflows. Anthropic has long emphasized its engagement with the OSS community, and this expansion is an extension of that policy.
Summary
China's MiniMax and its planned 2.7-trillion-parameter model show that the competition over model scale is far from over. The warning about an arms race from a former DeepMind executive offers an important perspective on where this race may lead. France's order for Meta to negotiate copyright signals a regulatory trend reshaping the relationship between content and platforms. Apple's massive deal with Broadcom and Temasek's AI-driven investment results demonstrate that AI infrastructure buildout is proceeding on both the semiconductor and capital fronts. And Anthropic's expanded free offering for OSS underscored that the race to win over developer communities is still going strong.