NVIDIA Releases the 550B-Parameter "Nemotron 3 Ultra"

NVIDIA has released the large language model **Nemotron 3 Ultra**. It uses a Mixture of Experts (MoE) architecture with 550B total parameters, of which 55B are active, and supports context lengths of up to **1 million tokens**.

The model is optimized for long-running AI agents, and its standout feature is the ability to process massive contexts while improving inference efficiency. It is already available on Hugging Face and is drawing attention from the local LLM community as well. NVIDIA also announced Nemotron 3.5 Content Safety, a multimodal safety model, signaling that enterprise-focused safety enhancements are moving forward in parallel.

Cloudflare CEO: "The Future of the Web Is Pay-to-Crawl"

Cloudflare CEO Matthew Prince announced that bot traffic on the internet has overtaken human traffic — a significant pull-forward from the timeline he had previously predicted, "around the end of 2027."

Prince pointed to the explosion of AI agents as the cause, concluding that "going forward, it's clearly going to be pay-to-crawl." As a harbinger of a business model shift around how web content is collected and used, the change is likely to affect both content creators and AI companies.

The AI IPO Rush Is Accelerating

Top-tier AI companies are racing to prepare for their IPOs. According to WIRED, the momentum has reached the point where some real estate listings are soliciting payment "not just in cash, but in Anthropic shares."

The wave of listings by AI companies, led by OpenAI and Anthropic, suggests that the AI sector's money cycle is transitioning "from venture capital to the public markets."

Supabase Hits a $10.5 Billion Valuation on Vibe Coding Demand

Supabase, an open-source database platform, has raised $500 million in funding, bringing its valuation to **$10.5 billion**. According to CNBC, the driving force behind this is a phenomenon known as "vibe coding."

Vibe coding is an approach to development that relies solely on natural language instructions given to AI coding assistants. The spread of AI-driven development is pushing up demand for backend infrastructure, fueling the growth of platforms like Supabase.

An Airline Caught Faking "Empathy" with AI After a Prompt Was Sent by Mistake

An airline has been found to be using AI to generate fake "empathy" in its customer communications. The practice came to light when an AI prompt was accidentally sent to one of its passengers.

Using AI to perform emotional solidarity rather than actually solving problems has ignited debate over the ethical use of AI.

Canada Aims for 250,000 Jobs and a 3% GDP Boost in Its National AI Strategy

The Canadian government has announced a new AI strategy. According to Reuters, the strategy targets the creation of 250,000 new jobs and a 3% boost to GDP.

The acceleration of national-level AI investment is drawing attention as a move to establish Canada as an AI power on par with the US and China.

LM Studio Makes It Possible to Run Local LLM Models on the iPhone

LM Studio has released a new feature called "LM Link" that makes it possible to remotely run even the largest local LLM models from an iPhone.

The setup allows models running on a PC to be controlled and used via an iPhone, further advancing the practicality of local inference on mobile devices.

---