The AI industry may be riding a wave of momentum, but two clouds have recently unsettled the outlook: growing public opposition to data center construction across the United States, and a dramatic leap in the rankings by Chinese-developed model Kimi K3, which has renewed fears of AI overinvestment. NVIDIA CEO Jensen Huang addressed both concerns directly on July 21.

Nvidia's newly detailed Vera Rubin NVL72 is being pitched less as a faster chip and more as a cheaper unit of AI output. This shift matters most for the power-constrained data centers now defining the ceiling of the industry's growth. In its July 21 announcement, Nvidia said the rack-scale system delivers roughly 10 times more tokens per megawatt than its current Grace Blackwell NVL72, and about one-tenth the cost per million tokens of the GB200 NVL72. This framing positions electricity, not silicon, as the scarce resource.

Intel and Fortinet have expanded their long-standing partnership to develop the Fortinet Security Processor 6 (SP6), a move expected to shape how cybersecurity hardware is designed, produced, and distributed globally. The effort is also intended to make the supply chain for critical security chips more resilient, potentially benefiting enterprises, governments, and consumers worldwide.

China is considering tighter export controls on artificial intelligence and semiconductor technologies amid intensifying competition with the US in frontier AI. The proposed measures, reported by Financial Times, would also target overseas acquisitions of advanced technology startups and could appear in the next revision of Beijing's export control catalogue.

Google is reportedly developing a new chip that embeds information from its Gemini AI models directly onto the silicon to boost efficiency. If the anticipated performance gains materialize, this new design would mark a deeper integration of the tech giant's hardware and AI capabilities.

Chinese AI developer Z.ai has completed a giant data center running exclusively on domestically made chips, a milestone in Beijing's drive to replace restricted Nvidia silicon for future AI development.

TSMC has announced an additional US$100 billion investment in the US on top of its existing US$165 billion commitment, expanding its Arizona footprint to 12 fabs and one R&D center. While the market has focused on the size of the spending, investors are more concerned that overseas expansion could squeeze profitability.

Samsung Electronics is cutting over 800 jobs in the US from its smartphone, display, and consumer electronics business units. The division, part of Samsung Group, is facing a challenging market this year, in contrast to the soaring performance of its chip unit.

Samsung Electronics' latest multi-billion-dollar equipment procurement reflects the company's accelerating investment in AI memory capacity while raising a broader question about what is driving one of the industry's most aggressive fab expansion strategies.

Rising demand for AI data centers and high-bandwidth memory (HBM) is pushing semiconductor competition further upstream into silicon wafers and critical materials. Micron and GlobalWafers recently announced a 10-year agreement, with Micron also providing US$500 million in support, marking the longest and potentially largest long-term supply agreement in GlobalWafers' history.

AI development is driving larger data transfers and higher GPU efficiency demands, pushing memory toward customization and prompting South Korean industry watchers to call for a shift to a "memory foundry" model. Sungkyunkwan University professor Seokjoon Kwon said at the Nano Korea 2026 forum that memory makers must move from mass production to order-driven design tailored to customer needs.

SiPearl has powered on and begun validating Rhea1, its first-generation server CPU designed in Europe. Its next challenge is convincing Taiwan's server manufacturers to turn the chip into systems that data centers can order and deploy.