NVIDIA Achieves 100% Liquid Cooling with "Rubin" — Efficiency Improves Even with 45°C Coolant

NVIDIA has announced the world's first design for its next-generation "Rubin" AI infrastructure platform in which every chip and network component is cooled by liquid alone. The standout feature is that the allowable coolant temperature has been raised to **45°C (Celsius)** — hotter than a typical warm-water pool (38–40°C).

Historically, data center cooling has accounted for up to 40% of power consumption. Under NVIDIA's DSX reference design, direct-to-chip cold plates and closed-loop circulation make it possible to operate without chillers (mechanical cooling) even in warm climates. As a result, compared with cooling tower setups, **annual water usage of up to 2.6 million gallons per MW can be cut to nearly zero**.

At a 50MW-scale hyperscale facility, annual cooling-related cost savings of more than $4 million are projected. Cooling vendors such as Motivair — part of Schneider Electric — are also accelerating their shift to liquid cooling as rising power densities push air cooling to its limits.

Preferred Networks Launches "PLaMo 3.0 Prime," with a Free API Plan on the Way

Preferred Networks (PFN) has begun offering "**PLaMo 3.0 Prime**," a from-scratch LLM built entirely in Japan. The company is putting its cost-performance front and center, and a free API plan is slated to be available.

For Japanese companies, it matters that the options keep growing for models optimized for Japanese — in terms of data handling and linguistic characteristics. See the official announcement for detailed specs and pricing.

OpenAI's New Role "FDE" — at One Point, "70% of the Job Vanished" Within Six Months

Colin Jarvis, OpenAI's Global Head of Forward Deployed Engineering, shared what the company's new role, "**FDE (Forward Deployed Engineering)**," actually looks like in practice.

FDEs embed deep in customers' environments and own AI solutions end to end, from deployment through continuous improvement. What's striking is that the work looks nothing like it did six months ago — there was even a period when "70% of the job disappeared." It highlights how the role itself is transforming at high speed to keep up with the pace of AI evolution.

Google DeepMind Publishes Its "AI Control Roadmap"

On June 18, 2026, Google DeepMind published its "**AI Control Roadmap**." As agentic AI shifts from the demo stage to real-world deployment, the need for safe control frameworks is growing.

The roadmap spans a broad range of domains — code changes, research planning, cyber defense, document creation, and product operations — laying out monitoring and governance policies for agents acting autonomously. The discussion is advancing from a practical standpoint on what's needed before agentic AI becomes a "colleague" within organizations.

"ChatGPT Never Mentions Our Company" — The LLMO Problem in the Age of AI-Assisted Job Hunting

In job hunting, top students increasingly take their career questions to LLMs. As a result, a problem has come into focus: companies with little web visibility and no **LLMO (LLM optimization)** strategy simply aren't being found by students.

As a new challenge succeeding traditional SEO (search engine optimization), interest in LLMO is rising in the recruiting and HR space. In an era when AI acts as the filter for information, companies are being forced to rethink what their digital presence should look like.

---