Inference Now Drives Two-Thirds of AI Workloads, Reshaping Chip Deployment
This summary and analysis were generated by AI from the original article at AI – SiliconANGLE and may contain errors (how Viqus works). Read the source for full details.
4
What is the Viqus Verdict?
We evaluate each news story based on its real impact versus its media hype to offer a clear and objective perspective.
AI Analysis:
The inference-shift thesis is credible and widely discussed, but this sponsored interview offers mainly one vendor's unverified figures and no independent evidence of market impact.
Article Summary
An interview segment from SiliconANGLE's theCUBE at SC26 features Positron AI sales chief Cameron McCaskill arguing that inference now accounts for about two-thirds of AI infrastructure workloads, up from roughly 15% in 2023. He says this shift is pushing buyers toward smaller, quickly built sites, where power, cooling and memory supply matter as much as raw chip speed. Positron, which raised $875 million at a $5 billion valuation in September, builds inference appliances that use commodity memory instead of scarce, expensive high-bandwidth memory (HBM). Its first-generation Atlas server runs on a programmable chip, and the company plans to tape out its own silicon early next year for the Titan server. Oracle is bringing about 340 Atlas servers live in its cloud this quarter, and a neocloud provider has signed on for Atlas and committed to Titan. A single Positron server draws about 4,500 watts, and a Titan rack fits within 40 kilowatts, suiting air-cooled modular sites. The article is sponsored content in part and offers no independent data on the workload split, so the figures are the vendor's own claims.
Key Points
- McCaskill says inference now represents about two-thirds of AI infrastructure workloads, versus roughly 85% training in 2023.
- Positron avoids HBM by using commodity memory, keeping power draw low enough for air-cooled modular data centers.
- Oracle is deploying about 340 Atlas servers this quarter, and Positron plans its own Titan silicon tape-out early next year.
Why It Matters
This is a vendor executive's view of a real trend: inference is becoming the dominant economic driver of AI buildouts, which favors power-efficient hardware and fast-deploy sites over large training clusters. The argument is plausible and matches broader industry commentary, but the workload split is unverified and the piece is a promotional interview with a funded startup. Its significance is moderate; it signals where a niche of chip vendors is betting, not a shift in the market leaders' position.

