China's open-weight AI models have surged in usage and have now overtaken proprietary models for the first time. South China Morning Post (SCMP), citing data from US web development platform Vercel, reported that on August 25, Chinese open-weight AI models accounted for 54% of total token usage across its own calls and the agent layer AI Gateway, surpassing proprietary models for the first time.

Amazon Web Services (AWS) and Nvidia are expanding their collaboration to add millions of GPUs and new infrastructure for agentic and physical AI. The move signals rising global demand for advanced computing, with implications for cloud customers, governments, and robotics developers seeking faster, more secure access to AI tools worldwide.

Tesla CEO Elon Musk recently said the company's upcoming AI5 inference chip could deliver two to three times the performance per watt of Nvidia products when deployed in autonomous vehicles and Optimus humanoid robots, potentially at just around 10% of the cost.

As the humanoid robot supply chain moves closer to mass-production readiness, Taiwan's precision manufacturing sector is gaining new opportunities to enter the US market, according to DIGITIMES observations.

The pinnacle of Elon Musk's far-reaching ambitions in orbital compute has come into better view following another major SpaceX announcement to turn the space giant's expanding plans for Starlink V3 and Starmind into a grand Starbase at the heart of Louisiana.

Chinese GPU vendors are beginning to extend their applications beyond terrestrial computing and into space. The recently launched Yunjian Muxi satellite (Jitianxing A-04) carries an intelligent computing payload powered by a domestically developed MetaX GPU, marking an effort to validate Chinese AI chips under real-world space conditions.

The 2026 World Robot Conference (WRC 2026) has concluded in Beijing. While the commercial prospects of humanoid robots remain a subject of debate, they once again took center stage at the event. With China continuing to lead the world in humanoid robot shipments, more than 100 robotics-related companies showcased new technologies and unveiled their latest products, making the event not only a magnet for capital but also a window into where the industry is heading.

For the first time in this cycle, Nvidia has told investors where its gross margin will bottom out and how far short of demand its supply will fall - two admissions that shift the question from whether the AI buildout continues to who absorbs its rising input costs. Management guided gross margin down from 75.0% to a trough of 71-72% by the fourth quarter, directly blamed memory pricing, and capped fiscal 2028 revenue growth at about 70% against demand it put at nearer 100%. Neither disclosure is a demand warning. Both say the company is now managing a shortage rather than a market.

Growth has stopped being the interesting question at Nvidia. Revenue for the quarter ended July 26, 2026 was US$96.2 billion, up 106% from a year earlier, and Data Center accounted for US$89.0 billion of it, up 117%. Gross margin held at 75.0% even as the Vera Rubin platform moved into volume—unusual, because a new rack-scale architecture normally costs a quarter or two of margin while yields settle. On the operating line, the quarter was close to flawless.

Nvidia has put a figure on its exposure to the AI labs it helps fund. On its second-quarter earnings call, chief financial officer Colette Kress said the company expects "demand from the AI labs for which we expect to leverage our balance sheet to contribute toward roughly a quarter of our business next year." That single sentence reframes the financing question. The issue is no longer whether Nvidia supports its customers' balance sheets, but that a quarter of forecast revenue now depends on structures the company itself has arranged.

The sharpest question on Nvidia's second-quarter call came a Bank of America analyst, who noted that OpenAI and Anthropic - both major beneficiaries of Nvidia's ecosystem investment - are designing their own chips, and that OpenAI had, days earlier, claimed its Jalapeno part outperforms Blackwell.