AMD shows another Helios after OpenAI lab visit
AMD posted another Helios lab photo on Oct. 1, 2026, tagged OpenAI, and described a tour. The post does not say OpenAI runs that rack.
The October 1 post tags OpenAI and describes a lab tour. It does not say OpenAI runs the rack in the photo.
On October 1, 2026, at 11:00 a.m. in Brasília (14:00 UTC), AMD posted on X a photo of a lab corridor next to a dense rack. It wrote that “another AMD Helios system enters the wild,” tagged OpenAI, and said that Vamsi Boppana and Uday Ruddarraju joined a lab tour the week before. This is not a new contract announcement. The text does not say the system in the photo is installed, operated, or in production at OpenAI.
That distinction matters because the two companies already have a large GPU deal on the calendar. A tagged photo in the same half-year as the first planned gigawatt is easy to read as a go-live. The post itself does not make that claim.
Quick recap
- AMD said on X that another Helios system “enters the wild” and that OpenAI people toured the lab the week before.
- Uday Ruddarraju replied the same day. AMD sells Instinct and Helios; he speaks for OpenAI.
- Helios, in AMD’s FAQ, is a reference design for a full AI cabinet, not a product for sale.
- It is not confirmed that OpenAI operates the rack in the photo.
First: what AMD Helios is
In AMD’s material, Helios is the blueprint for a full AI server cabinet, not a single chip and not a boxed product you order like a laptop. A GPU is the graphics processor that does the heavy math for the models. AMD describes Instinct GPUs, EPYC processors, and Pensando networking in the same rack. Each customer still has to assemble and run the hardware.
The FAQ, still up on October 9, 2026, says Helios “is a reference design, not a product for sale,” with volume in the second half of 2026. The cabinet uses Open Rack Wide (ORW), an open format Meta presented at the 2025 Open Compute Project (OCP). The published design is the recipe. The customer still has to build the kitchen.
In an October 14, 2025 blog post, AMD says a Helios rack with 72 Instinct MI450 Series GPUs delivers up to 1.4 exaFLOPS in FP8 and 2.9 exaFLOPS in FP4, 31 TB of HBM4 memory, and 1.4 PB/s of bandwidth. An exaFLOPS is a billion billion operations per second. FP8 and FP4 are 8-bit and 4-bit numbers. HBM4 is stacked memory sitting next to the GPU. The figures come from AMD Performance Labs, with a September 2025 footnote, as ceilings (“up to”), and with a note that results may change. The October 1 post does not cite specs.
The product page, on October 9, 2026, lists the MI455X GPU (432 GB of HBM4, up to 23.3 TB/s) and the same rack ceilings. Investor material dated July 23, 2026, called Helios “now in production.” The FAQ still calls it a reference design. Those are two statements from the same company. The lab post does not settle which one applies to the rack in the photo.
What AMD posted
The X post said another Helios system “enters the wild,” tagged OpenAI, and noted that last week Vamsi Boppana and Uday Ruddarraju joined the team for a lab tour. AMD did not give Vamsi’s title or the lab’s address in that post. On July 23, 2026, AMD’s newsroom described him as senior vice president of the company’s Artificial Intelligence Group.
Uday Ruddarraju replied the same day: “It’s always great spending time with the @AMD team. Some deliveries you don’t want to miss!” According to his LinkedIn profile, he has been Chief Technology Officer, Compute at OpenAI since July 2026, the job that oversees the computers that train and serve the models. “Deliveries” is his wording, not a rack count or a power figure.
On October 5, 2026, the newsroom published the series “Building AMD Helios” and wrote that “Customers are testing AMD Helios at their own sites.” The page embeds the October 1 post and does not name OpenAI as the operator of the rack in the photo.
The 2025 AMD-OpenAI deal
On October 6, 2025, AMD and OpenAI announced a plan to deploy 6 GW of AMD GPUs, with the first 1 GW of Instinct MI450 Series slated for the second half of 2026. A gigawatt, here, is electrical capacity for the install, not a measurement of the photographed rack. The announcement also cited rack-scale AI solutions and a warrant for up to 160 million AMD shares, the cap on a right to buy stock, tied to milestones. Those are the two companies speaking, a year before the lab post. They are not the same event as the October 1 photo.
What the post does not confirm
It is not confirmed that OpenAI operates this photographed rack, or that it is already in production there. RUMOR: the lab’s location, the exact tour date beyond “last week”, the number of racks, power draw, whether the chip is MI450 or MI455X, and whether “enters the wild” means a commercial shipment.
On July 23, 2026, AMD wrote that OpenAI “has had access to Helios systems for several months” and that it expects to bring Helios online through multiple partners in the second half of 2026. Access and expectation, according to the vendor, do not identify the rack in the photo.
In the half-year when the first gigawatt is due, AMD showed “another” Helios and an OpenAI visit. We still do not know whether this is lab validation, a customer test, or a relationship post. The photo exists. The post does not say the rack is in production at OpenAI.