Chips & Supply · Europe

Arm Calls for Rack-Level CPU Benchmarks for Agentic AI, Claims Twice AMD's Performance

Credited to TheElec · thelec.net · 4h ago

British semiconductor intellectual property (IP) company Arm said on Oct. 15 that it has developed a new central processing unit (CPU) benchmarking framework for agentic artificial intelligence (AI), with support from Fujitsu, Meta and OpenAI.

The company said the framework is designed to evaluate real-world agentic AI workloads and performance across the entire computing process.

Arm claimed its AGI CPU delivered approximately twice the task throughput per rack of AMD's fifth-generation EPYC Turin 9755 server CPU under the new framework. The measurements were conducted by analyst firm Signal65 at Arm's request. The AGI CPU, unveiled in March, is Arm's first internally designed server CPU and was jointly developed with Meta.

Unlike chatbots that respond to a question and stop, AI agents repeatedly invoke external tools, retrieve data and execute code until a task is completed. GPUs handle model inference, while CPUs manage tool calls, data movement and task scheduling. As the number of agents increases, CPU workloads also grow. Arm said conventional CPU benchmarks and model inference metrics fail to capture these performance demands.

The new framework evaluates tooling, including calls to external applications, databases and services and data processing; reasoning and planning for breaking down complex tasks and organizing execution; the time required to create isolated code execution environments, or sandboxes; and the ability to complete real-world coding tasks and operate in command-line interface (CLI) environments.

Meta is also adopting Arm's AGI CPU in its next-generation open server rack design, Yosemite YV5.

Chips & SupplyEurope

ASML patents point to a hybrid bonder as chip toolmakers crowd into packaging

Credited to DigiTimes$ · digitimes.com4h ago

Useful

As it becomes harder and costlier to shrink transistors, advanced packaging has become the place chip equipment makers look to for their next leg of growth, and ASML appears to be preparing its second move into it. Patent filings analyzed by research firm SemiAnalysis suggest…

View the article · $DigiTimes

Front Method

Capital · Europe

Nscale’s $35 Billion IPO Plans Muddied by AI Data Center Delays

Reading from TechCrunch · Credited to Bloomberg $ · bloomberg.com · 8h ago

When British neocloud Nscale goes public, it will test public investors’ appetite for a stock whose revenue is tied primarily to two customers.

Since it was spun out of Australian cryptocurrency mining company Arkon Energy two years ago, Nscale has amassed over $103 billion worth of contracts, according to its IPO filing. But there’s a catch: Most of that, about 85%, comes from a deal to supply Microsoft with $43.8 billion worth of compute through 2033, and another supply agreement worth $44.6 billion with Anthropic.

Moreover, Anthropic’s agreement is contingent on Nscale obtaining financing, and the AI lab retains the right to walk away from or cancel the deal if Nscale fails to hit milestones that the filing explicitly categorizes as “stringent.”

Nscale’s customer concentration is a reminder of just how interconnected the AI industry has become. A recent paper by credit hedge fund Sona Asset Management, featured in Financial Times, found that many AI infrastructure providers heavily depend on a limited number of customers. Nscale’s competitor CoreWeave, for example, generates 67% of its revenue from Microsoft, and data center builder Applied Digital derives 67% of its revenue from Oracle, and 30% from CoreWeave.

While Sona noted that such interconnectedness is not necessarily a bad thing, it pointed out that a single setback or strategic shift by a major player can easily affect the entire industry.

Nscale, which plans to list on the NYSE, expects to be valued at $35 billion, Financial Times reported, and is seeking to raise $3 billion in the offering, according to Bloomberg.

The company reported revenue of $140.6 million for the six months ended June 30, up significantly compared to $10.4 million a year earlier. Net losses jumped to $1.02 billion from $369 million in the same period.

Earlier this month, one of Nscale’s major investors, Nvidia, agreed to provide the company with $1 billion in convertible debt as part of a larger $3.1 billion financing deal. The startup was valued at $14.6 billion when it raised a $2 billion Series C led by Aker ASA and 8090 Industries.

Aside from CoreWeave, Nscale’s competitors include Nebius, Lambda, and Crusoe — the latter last week said it raised $3.9 billion at a $30.9 billion valuation.

Nscale operates data centers in Norway, Portugal, Texas, and West Virginia. Its board of directors includes former Meta executives Sheryl Sandberg and Nick Clegg, as well as former OpenAI executive Fidji Simo.