NVIDIA Now Powers 81% of the TOP500 Supercomputers
The latest TOP500 rankings, announced at ISC 2026 in Hamburg, reveal that NVIDIA technology powers more than 400 of the world's top 500 supercomputer systems. That works out to an 81% share — an increase of 17 systems over the previous list.
Notably, roughly 90% of the newly ranked systems adopt NVIDIA technology. In addition, 26 systems on the TOP500 now run on NVIDIA Grace CPUs, up eight from the previous list. On the Green500, which ranks power efficiency, the top eight systems are all equipped with NVIDIA GPUs, with the number-one system, KAIROS, driven by a single Grace Hopper Superchip. Furthermore, 376 systems are connected via NVIDIA networking, underscoring the company's growing footprint in communication infrastructure as well as compute.
The numbers confirm that GPU demand driven by the AI boom is further cementing NVIDIA's dominant position in the HPC space.
OpenAI Releases GPT-5.5-Cyber — From Vulnerability "Detection" to "Automated Remediation"
OpenAI has expanded its cybersecurity initiative "Daybreak" and released the full version of GPT-5.5-Cyber, a security-specialized model. Alongside the model, the company announced an update to the Codex Security plugin and the launch of a partner network that includes more than 25 security firms and several government agencies.
The biggest change is that AI's role is shifting from vulnerability "detection" to "automated remediation." Whereas conventional security tools stop at flagging problem areas, GPT-5.5-Cyber is designed to generate and apply patches as well. OpenAI claims it outperforms Anthropic's Mythos model on benchmarks, though its effectiveness in real-world operations remains to be seen.
With the cybersecurity talent shortage deepening, practical AI-driven remediation could reshape the industry's very structure.
ASML's ¥40 Billion EUV Machine — The Frontline of Semiconductor Miniaturization
MIT Technology Review has published a detailed report on ASML's latest EUV (extreme ultraviolet) lithography machine. Roughly the size of a double-decker bus and a precision block of aluminum weighing more than 150 tons, the machine carries a price tag of about $400 million (around ¥60 billion).
At the heart of the machine lies "the mechatronics that hold the mirrors with atomic-level precision." Developed over more than a decade under the leadership of ASML's Jos Benschop, this technology is the key to miniaturizing next-generation semiconductors. As demand for AI chips keeps growing, pushing the limits of lithography is the only way to manufacture faster, more efficient chips.
The machine stands as a symbol of the astonishing concentration in modern technology, where a single piece of equipment can determine the future of the global semiconductor industry.
Sakana.ai Releases "Fugu" Model — Taking on Anthropic's Fable
Sakana.ai, a Japanese AI startup, has released Fugu, a new model positioned to compete with Anthropic's Fable. Details are available on the release page, and the announcement has also been generating buzz on Hacker News.
Sakana.ai has built its reputation on distinctive approaches such as evolutionary algorithms and natural language processing. While specifics on Fugu's architecture and training methods await the official announcement, an AI lab from Japan drawing worldwide attention with a frontier-class model release is a meaningful development for the diversity of the ecosystem.
Stripe Announces "MPP," a Payment Protocol for AI Agents — HTTP 402 Is the Key
Stripe, a major payments company, has introduced MPP (Machine Payments Protocol), a protocol that enables AI agents to make payments autonomously. The starting point is a simple observation: "The current financial system was built for humans. That's why AI agents can't use it."
The technical core is a machine-to-machine payment mechanism built around HTTP status code 402 (Payment Required). When an AI agent calls an API and encounters a paid endpoint, it receives HTTP 402, automatically processes the payment, and retries the request. A world where AI can procure and consume resources without human intervention holds the potential to dramatically expand agent autonomy.
At the same time, the risks of unintended spending and abuse by AI remain issues that warrant discussion. While the technical framework is coming together, governance maturity still has a long way to go.