Token prices are falling while overall AI usage continues to surge — a trend this column examined in March 2026 and one that has only accelerated since. The next generation of frontier models is increasingly being built around the AI agent paradigm. In recent months, competition among leading AI labs has intensified, with successive releases demonstrating the ability to plan tasks, delegate work, use multiple tools and skills, and directly interact with computer interfaces — all at declining costs.
As the debate over the US fiscal 2027 defense spending plan to dismantle the Space Development Agency (SDA) continues, the Proliferated Warfighter Space Architecture's (PWSA) transport layer is concentrating around SpaceX, seemingly narrowing the door for suppliers. In reality, however, it marks a shift in the business focus from selling a batch of satellites to supplying continuous replacement components and terminal opportunities that are still up for grabs.
For the past two decades, enterprise information governance has focused on protecting data through classification, encryption, backups, and internal containment. That remains essential, but in the AI era, competitive advantage increasingly comes not from data itself, but from how effectively companies turn it into operational improvement.
South Korea's technology policies—including the two semiconductor strategies discussed previously and the broader K-Moonshot initiative introduced more recently—have been drafted by different government bodies, including the Ministry of Trade, Industry and Energy (MOTIE), the Presidential Office, and the Ministry of Science and ICT (MSIT). Yet they share several common characteristics.
South Korean President Lee Jae-myung unveiled the country's Three Mega Projects for AI and Semiconductors in late June 2026, an ambitious national strategy designed to strengthen South Korea's global leadership in artificial intelligence and semiconductors. The initiative centers on three pillars—semiconductors, physical AI, and AI data centers—and aims to double the nation's DRAM output within five years while expanding capabilities in high-bandwidth memory (HBM), advanced packaging, AI processors, and next-generation memory technologies. It also seeks to extend South Korea's semiconductor footprint beyond the Seoul metropolitan region.
At a humanoid robotics summit in Tokyo in May 2026, I saw a consulting firm's global labor automation map for Physical AI. After returning, I recreated the same map using the firm's research on Digital AI job functions. Placing the two side by side revealed something unexpected.
The race to commercialize physical AI and autonomous robots is running into a fundamental challenge: existing robot safety frameworks were designed for deterministic systems operating in controlled environments, not for autonomous machines making decisions in dynamic, unstructured ones.
Over the past decade, annual venture capital invested in physical AI and robotics startups has surged from a few hundred million US dollars to nearly US$25 billion, more than a 10x increase concentrated in recent years.
In early June in Vienna, a robotics startup used its keynote at ICRA 2026 — the International Conference on Robotics and Automation — to show a robotic arm slowly and precisely shaving its founder's face.
Computex Taipei 2026, held from June 2 to 5 under the theme "AI Together," drew more than 1,500 exhibitors from 33 countries and set a new record in scale. The show underscored a new AI industry reality: competition has moved far beyond standalone chip compute and into a systems-level battle spanning compute, connectivity, power, and cooling.
At the close of his keynote address at the Humanoids Summit in Tokyo, Hiroshi Ishiguro — one of the pioneers of humanoid robotics — offered a candid assessment of the industry's progress: despite decades of investment and research, Japan has yet to produce a truly transformative, mass-market application for robotics.
Falling inference prices and tightening data regulations are pushing AI compute beyond the hyperscale data center — reshaping infrastructure decisions for enterprises, governments, and device makers worldwide
Over the past year and a half, reasoning in large language models (LLMs) has become a mainstream capability, with measurable gains across programming, mathematics, law, and healthcare. The robotics industry is now asking whether the same can be done in the physical world.
AI's rapid evolution — from AI servers to agentic AI and emerging physical AI — centers on high-performance computing, and integrating general fault-tolerant quantum computers into that stack could change what HPC can do. The transition, however, confronts deep technical mismatches between classical AI servers and quantum processors.
