Positron claims Oracle win

- Social posts claim Positron's Asimov AI chips were sold to Oracle, positioning the startup against Nvidia in datacenters. (x.com/i/status/2046417590252367968) - The posts frame Asimov as a challenger to Nvidia's H100/H200 class of datacenter GPUs. (x.com/i/status/2046434408400290196) - A confirmed hyperscaler or large cloud buyer purchase would be a rare commercial breakthrough for an AI‑ASIC challenger. (x.com/i/status/2046417590252367968)

Positron is telling investors and customers that Oracle bought systems built around its Asimov artificial-intelligence chips, but neither company has publicly confirmed an Oracle deployment. (positron.ai) (oracle.com) The claim surfaced in social posts that described Asimov as an Oracle win and cast the chip as an alternative to Nvidia hardware used in cloud data centers. Oracle’s public artificial-intelligence infrastructure pages still center on Nvidia and Advanced Micro Devices graphics processing units, with no Positron listing. (x.com) (oracle.com) Positron says Asimov is a custom chip for running transformer models after they are trained, the stage known as inference, and says each chip will carry more than 2 terabytes of memory at a 400-watt power envelope. The company says Asimov is scheduled for 2027 and that its Titan server will package four Asimov chips with more than 8 terabytes of accelerator memory. (positron.ai 1) (positron.ai 2) That design target is different from the Nvidia systems Oracle advertises today for both training and inference. Oracle says its Oracle Cloud Infrastructure Supercluster with Nvidia H200 graphics processing units scales to 65,536 GPUs, and Oracle said in March 2026 that it was expanding its Nvidia collaboration again. (oracle.com 1) (oracle.com 2) Positron’s pitch is that inference is increasingly limited by memory capacity and bandwidth rather than raw math throughput. On its site, the company says current accelerators often realize less than 30% of theoretical memory bandwidth on real inference workloads and argues that purpose-built silicon should work alongside, not replace, GPUs. (positron.ai) The company already markets an earlier product, Atlas, as “shipping today,” while positioning Asimov as the next-generation chip behind future Titan systems. Atlas is advertised as an air-cooled inference server with more than 4 times performance per watt and more than 3 times performance per dollar versus Nvidia, though those comparisons are presented by Positron and not independently benchmarked on the page. (positron.ai) Positron has raised enough money to get attention in a market dominated by much larger suppliers. Its press page links to coverage of a $230 million Series B and says the company is building United States-designed, fabricated, and assembled AI acceleration hardware. (positron.ai 1) (positron.ai 2) What is missing is the evidence that usually turns a chip startup’s customer claim into a cloud-market fact: a named Oracle contract, an Oracle product page, or an Oracle press release. Until one of those appears, the Oracle sale remains an unverified claim layered on top of a very real push by Positron to sell inference hardware into data centers now dominated by Nvidia systems. (oracle.com) (oracle.com)

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.