AMD aims to deliver first Helios racks to data centres this quarter

AMD aims to deliver the first Helios racks to data centre customers before the end of this quarter. That is what CEO Lisa Su said during a discussion of the company's most recent quarterly results. In the fourth quarter and throughout 2027, AMD intends to scale up deliveries further to meet demand that Su described as "very strong".

Helios is AMD's integrated rack solution for AI compute workloads in data centres. The system combines processors, memory, networking hardware and cooling in a single pre-assembled rack, allowing data centres to order a complete compute unit rather than assembling individual components themselves. AMD is targeting the growing market for large-scale AI infrastructure with this offering.

Su did not provide specific figures on initial delivery volumes or the associated revenue outlook. AMD does not typically publish exact revenue numbers broken down by product line.

What Helios entails

Helios is built around AMD's MI300-series AI accelerators and is designed as a ready-to-deploy building block for large AI data centres. By integrating hardware and software at the rack level, AMD aims to reduce installation and configuration time for data centres. Customers face less system integration work themselves, which accelerates deployment.

The approach aligns with a broader industry trend: Nvidia is also bringing integrated rack solutions to market with its NVL platforms. Vendors are seeking to add more value than discrete chips alone while simultaneously embedding customers' platform choices into their infrastructure.

Scale-up central to AMD's planning

Su emphasised during the quarterly presentation that deliveries will increase substantially in the second half of this year and in 2027. The timing is not coincidental: major hyperscalers and cloud providers are actively expanding their AI infrastructure, and AMD wants to capture a share of that wave of investment.

Su explicitly described customer demand as "very strong", though her remarks did not make clear exactly which customers have already signed contracts or in what volumes. AMD typically keeps such details out of its public communications, in part to protect competitively sensitive information.

Investors and analysts are closely watching the extent to which AMD is gaining ground on Nvidia in the AI hardware segment. Concrete delivery timelines and scale-up plans give that audience a more tangible picture of commercial progress than product announcements alone.

AMD's position in the AI hardware market

AMD has been steadily building market share in the AI accelerator segment in recent years, a market long dominated by Nvidia. The MI300X chip, the predecessor to the current generation, already attracted serious interest from cloud providers such as Microsoft and Meta. Helios represents the next step: moving from discrete chips to complete systems.

Nevertheless, Nvidia remains the largest player for now, both in terms of market share and ecosystem. AMD's CUDA alternative, the ROCm software stack, enjoys broader support than it did a few years ago, but many AI developers still work primarily with Nvidia's tooling. The hardware-software integration in Helios is partly a way to lower that barrier: customers receive a working system rather than separate components they have to make interoperate themselves.

Whether deliveries this quarter are actually on schedule and the scale at which the 2027 ramp-up will take place will become clear from AMD's subsequent quarterly reports and from the investment plans of major data centre operators.

For the European and Dutch AI infrastructure market, the availability of a second serious supplier of integrated AI rack solutions is significant. European data centres and cloud providers expanding their AI capacity benefit from having multiple vendors to avoid dependence on a single party. AMD's scale-up broadens the options available to those buyers, which also affects the negotiating position of European customers relative to Nvidia.