Samsung Starts Mass Production of Latest SSD for Nvidia's "Vera Rubin"

Samsung Electronics has begun mass production of its latest enterprise SSD, the PM1763, designed for Nvidia's next-generation AI platform "Vera Rubin." Announced at this year's Nvidia GTC conference, the product is positioned as part of a comprehensive storage solution for AI data centers, alongside HBM4 memory and low-power SOCAMM2 modules.

As AI workloads surge, eliminating bottlenecks in the data center storage layer has become an urgent priority. Vera Rubin is regarded as Nvidia's next-generation platform, and the start of mass production for components that support it suggests the transition to next-generation AI infrastructure is steadily underway.

EU Unveils New Plan to Address Cybersecurity Risks of Advanced AI

The European Commission has announced a new plan to address the risks and opportunities that advanced AI brings to cybersecurity. With the threat of AI-enabled cyberattacks on the rise, the EU is laying out a two-pronged approach: strengthening its defensive capabilities while also leveraging AI to improve security.

Separately, with the **EU AI Act taking effect on August 2**, practical compliance checklists for engineering teams are also drawing attention. The EU AI Act is the world's most comprehensive AI regulatory framework, and once enforcement begins, companies in violation could face fines of up to 7% of their revenue. For those on the implementation side, figuring out how to meet the technical requirements is an urgent challenge.

How OpenAI's Low-Latency Voice AI Serves 900 Million Users

A technical deep dive from ByteByteGo analyzes how OpenAI achieves low latency in the voice AI that serves its roughly 900 million users. Delivering real-time voice conversations requires optimizing the entire pipeline—from speech recognition to inference to speech synthesis—making innovations across multiple layers, such as streaming processing and lightweight models, essential.

In voice AI, response latency can make or break the user experience, so backend architecture design holds the key to competitiveness. This analysis should serve as a useful reference for engineers interested in designing voice AI systems that scale.

Rising Costs Push US Companies Toward Chinese AI Models

According to CNBC, as the cost of using US AI models from OpenAI, Anthropic, and others continues to climb, US companies are starting to turn to Chinese AI models. Chinese models such as Kimi (月之暗面) and MiniMax, which offer high performance at low cost, are becoming a realistic option for companies scrambling to cut costs.

Meanwhile, some platforms are moving to offer free access to Chinese models. For example, Dahl Global's inference platform is running a campaign offering 100 million free tokens for Kimi and MiniMax models, drawing interest from users.

This trend signals that "cost per unit of performance" is emerging as a new competitive axis in the AI model market. That said, concerns about data privacy and geopolitical risks remain, and companies will need to strike a careful balance.

Anthropic Expands Its Manhattan Footprint, Accelerating New York's AI Boom

According to the New York Times, Anthropic is expanding its office in Manhattan. New York has seen a growing concentration of AI companies in recent years, and Anthropic's expansion looks set to further accelerate this trend.

AI talent and capital, once concentrated in the San Francisco Bay Area, are spreading to New York as well. New York's mix of diverse industries—finance, media, fashion, and more—is seen as a unique strength when it comes to opening up new application areas for AI.

---