The US government is investigating Chinese artificial intelligence developer Moonshot AI over allegations that its Kimi K3 model improperly used outputs from Anthropic's Fable model and was trained with restricted Nvidia GB300 chips accessed outside China.
China's research and academic community appears to be drifting away from international collaboration, even as the UK, Germany, and the US have steadily expanded cross-border partnerships in recent years. The contrast has drawn fresh attention after Moonshot AI's flagship 2026 model Kimi K3 and other Chinese technology products, including Unitree Robotics's H1 robot, prompted questions over how much of China's advanced innovation is truly homegrown.
SpaceXAI is laying the groundwork for at least one large-scale AI data center in Texas, a move that could significantly expand its computing footprint beyond its existing Memphis hub and strengthen its ambitions to become a major AI infrastructure provider. According to The Information, the company has evaluated multiple sites across Texas, transferred data center staff to the state, and posted job openings related to site acquisition, construction, and operations. A Texas buildout could match or exceed the roughly 1GW of compute capacity already deployed across its Memphis-area facilities.
Samsung is in talks to invest about EUR1 billion (approx. US$1.14 billion) in French AI startup Mistral as part of a wider fundraising round, the Financial Times reported, citing people familiar with the matter.
Fitch Ratings has adjusted its Taiwan GDP growth forecasts for 2026 and 2027 to 9.4% and 4.8% respectively, up from its earlier estimates of 6.9% and 4.0%. At a forum held in Taipei on July 21, Fitch noted that impacts from the US–Iran conflict are expected to be manageable. With the easing of US tariff pressures, Taiwan's economy is expected to outperform others in the Asia-Pacific region, driven primarily by strong global demand for semiconductors and AI-related products, as well as growing investment by Taiwanese companies.
Nvidia is moving to lock down the networking layer of the world's largest AI data centers, unveiling its Spectrum-6 Ethernet switch as the backbone for what it calls "gigascale" AI factories built around its next-generation Vera Rubin platform. The launch, detailed on the company's blog on July 21, is a reminder that networking — not just GPUs — has become a central battleground as operators scale toward clusters of more than 100,000 accelerators.
