What happened
AMD, Cisco, and HUMAIN said on August 31 that infrastructure built around AMD Instinct MI355X accelerators and Cisco Silicon One networking was live in Saudi Arabia and serving HUMAIN customers.
The operating stack includes MI355X GPUs, EPYC CPUs, Cisco N9000 networking, Silicon One, and 800G optics. The companies did not disclose the megawatt capacity of this initial live deployment.
Live now, larger later
The announcement also described a planned phase of up to 250 MW using MI400 GPUs, EPYC CPUs, ROCm, and Cisco infrastructure. Deployment is planned to begin in 2027, with the first capacity expected in the second half of that year.
A separate target calls for the joint venture to reach up to 1 GW by 2030. Those figures describe a roadmap, not the size of the service operating today.
Why it matters
The live system is evidence of movement from an investment program to an operating service. It also shows what sovereign AI infrastructure requires in practice: accelerators, CPUs, optics, networking, software, local operations, and controls over where data and workloads reside.
The next question is delivery pace. A functioning initial cluster reduces one kind of execution uncertainty, but it does not answer whether hundreds of additional megawatts can be supplied, built, connected, and filled on the proposed schedule.
What to watch next
- Disclosure of the operating cluster’s current scale and customer mix.
- The start of the planned MI400 phase in 2027.
- Evidence that initial capacity comes online during the second half of 2027.