AMD Helios Takes On Nvidia: Rack-Scale AI System Arrives in Late 2026

AMD is preparing to ship Helios, a rack-scale AI system built on Instinct MI400 accelerators that competes directly with Nvidia's server platforms. According to a TechCrunch report dated July 23, 2026, first customer deliveries will begin before the end of 2026.

What Helios is

Helios is not a single chip but an entire server rack, designed by AMD to function as one accelerator for training and running large AI models. At its core are AMD Instinct MI400-series accelerators, linked into a single unit by high-speed interconnects.

Each MI400 accelerator carries memory of the new HBM4 standard, which is critical for loading models with hundreds of billions of parameters without constantly reaching out to external nodes. Helios continues the line started by the MI300 and MI350 series: AMD has been steadily increasing memory capacity and bandwidth to catch up with Nvidia in the data center.

According to AMD, the system's key specifications are:

• Customer shipments: before the end of 2026
• Built on AMD Instinct MI400 accelerators
• 72 GPUs in a single rack
• 31 TB of HBM4 memory and about 1.4 PB/s of bandwidth
• Open interconnect standards instead of proprietary NVLink

The rack-scale format means the customer gets a ready-made compute node rather than a set of cards to assemble themselves. This is the approach that cemented Nvidia's position in data centers: its GB200 NVL72 racks are sold as a single system, and AMD is now answering with a product in the same class for the first time.

How Helios differs from Nvidia

Helios's main differentiator is its bet on open standards. AMD uses UALink and Ultra Ethernet to connect accelerators, while Nvidia builds its racks on the closed NVLink. By AMD's design, this lets customers avoid lock-in to a single supplier and mix hardware from different vendors.

"Helios is designed to operate as one giant compute engine," said

Lisa Su, CEO of AMD, when announcing the system.

Openness is a strategic answer to Nvidia's closed ecosystem, where chips, networking, and CUDA software are tied together. AMD counters with a combination of its own hardware, open interconnects, and the ROCm platform.

Why it matters for the market

Nvidia controls more than 90% of the AI accelerator market, and Helios is AMD's most serious attempt to take a share of that demand. The biggest buyers of compute (Microsoft, Meta*, Oracle) are looking for a second supplier to reduce dependence and bring down prices, and AMD's rack-scale system gives them a comparable alternative for the first time. Even a few percent of this market would mean billions of dollars in AI accelerator revenue for AMD.

Demand for training large models continues to grow faster than Nvidia can ship hardware. Nvidia is already preparing its answer, the next generation of racks, Vera Rubin, so Helios arrives in the middle of a head-to-head platform race, not an empty niche. For AMD, this is a window of opportunity: competitive performance and total cost of ownership could secure the company a place in hyperscaler procurement.

What this means

The race for AI hardware is no longer Nvidia's monopoly. The rack-scale system is becoming the new unit of sale, and with Helios the largest data centers get a real second supplier for the first time, which means pressure on prices and delivery times.

Frequently asked questions

When will AMD Helios be released?

First customer shipments will begin before the end of 2026, according to TechCrunch on July 23, 2026. The system is built on AMD Instinct MI400 accelerators.

How does AMD Helios differ from Nvidia's systems?

Helios uses the open interconnect standards UALink and Ultra Ethernet instead of Nvidia's proprietary NVLink. It is a 72-GPU rack-scale system that AMD positions as a single compute node for training large AI models.

*Meta is designated an extremist organization and banned in Russia.