Urgent.News

What's breaking now, across hundreds of outlets.

AI

Inference now drives two-thirds of AI workloads, changing where chips get deployed

AI spending is shifting from giant training clusters toward smaller, quickly built sites, and modular data centers are becoming a natural home for inference workloads. Power, cooling and memory supply now shape where tokens get generated as much as raw chip performance does. Inference-first chipmakers want to capture that shift. Positron AI Inc., which raised […] The post Inference now drives…

Inference now drives two-thirds of AI workloads, changing where chips get deployed

AI spending is moving away from large training clusters, with modular data centers becoming a common place for inference workloads. The power, cooling, and memory supply now impact where tokens are generated. Inference-first chipmakers aim to capitalize on this trend. Positron AI Inc., valued at $5 billion after raising $875 million in September, designs inference appliances around common memory rather than expensive high-bandwidth memory.

Cameron McCaskill, Positron's head of sales and business development, said inference now accounts for two-thirds of AI workloads, compared to one-third for training in 2023. This shift drives economic decisions for building out AI infrastructure. Inference has overtaken training as the driving force behind infrastructure purchases.

Positron's Atlas server, running on a programmable chip, is set to have its own silicon released early next year. Oracle is deploying about 340 Atlas servers in its cloud this quarter, and another neocloud provider has also signed up for both Atlas and Titan. To keep power low, Positron avoids high-bandwidth memory. Their servers draw around 4,500 watts per unit and fit within a 40-kilowatt rack.

This design suits sites with cheap power and few regulatory hurdles. Modular data centers, where operators can quickly deploy half to 2.5 megawatts, are attractive for inference providers seeking rapid deployment. These data centers, combined with Positron's technology, can enable monetizing new capacity within 90 to 120 days.

Rewritten by urgent.news; may contain errors — check the original.

Read the original at siliconangle.com →

Report a problem with this story

More in AI

Agentic AI for Payment Exceptions in LangGraph: Deadline Triage, Sanctions Hits & a Release the LLM Must Never Make

Payments is the domain where "the happy path is already automated" is most literally true — straight-through processing has existed for decades, and a clean payment never touches a human.

  • Agent classifies exception types and urgency using a small model
  • Escalates high-value exceptions with <30 minutes remaining deadline
  • Drafts discrimination narrative for sanctions hits but cannot clear false positives

Upscaling AI-Generated Video to 4K: Resolution, Bitrate, Frame Rate and Artifacts

Disclosure: I'm writing this as part of the VideoUpscalerAI team. The product-specific numbers below come from our own published docs, and the rest is general video engineering that applies to any…

  • 4K upscaling requires consideration of resolution, bitrate, frame rate and artifacts
  • Integer scale factors (2×) are cleaner than higher fixed factors
  • Generated footage often shows artifacts like smeared textures and blocky compression

Small Company Rides Anthropic Hype With A $657K AI Bet, Not $800M

Genius Group , a small education company listed in New York, is telling shareholders that Anthropic is now the biggest holding in its AI investment pot after Anthropic's record IPO. The catch: the only purchase it has disclosed for that pot so far is $657,100, set against a headline plan of $800 million.

More from Saturday 21 November →