This observation covers roughly 08:00 on 31 August to 08:00 on 1 September 2026 in Australia/Sydney. It includes three publicly available and verifiable technical changes. They are at different stages—production operation, construction and collaborative development—and capacity or products that have not been delivered are not treated as current capability.
AMD, Cisco and HUMAIN bring production AI compute online
AMD announced on 31 August that AI computing infrastructure deployed by HUMAIN is now in production in Saudi Arabia and serving customers. The current system uses AMD Instinct MI355X GPUs and EPYC CPUs, interconnected through Cisco N9000-series equipment, Silicon One networking and 800G optics. It supports model training and inference through a GPU-as-a-service offering. Telecompaper also reported the operational and customer-service status. The companies have not disclosed the number of GPUs, operational power capacity, customer workload data or independent performance tests.
The change advances the joint venture and construction plan announced in November 2025 into actual operation. Reuters reported at that time that the companies planned an initial 100 MW project and were targeting 1 GW by 2030; construction had not yet begun. The production system announced on 31 August uses MI355X hardware and should not be conflated with the later expansion. AMD also says that a subsequent buildout of up to 250 MW, based on MI400-series GPUs, EPYC processors and ROCm, is planned to begin in 2027, with capacity expected to start coming online in the second half of that year. This remains planned capacity rather than compute available today.
HUMAIN and DataVolt begin construction at NEOM Oxagon
HUMAIN and DataVolt expanded their partnership on 31 August and announced that construction had begun on an AI data centre at NEOM Oxagon. HUMAIN is participating in 100 MW of the development, which sits within DataVolt's planned 360 MW first phase; the wider campus is planned at 1.5 GW. NEOM's official statement and Reuters both confirm that the first 100 MW is expected to become available in 2028. The intended design includes renewable energy, cooling systems, high-performance computing clusters and subsea-cable connectivity.
The newly confirmed change is the move from planning into construction, not the availability of 100 MW of operating capacity. The announcement does not identify a GPU supplier, accelerator model, server count, percentage of construction completed or contracted customers. The location, construction status, capacity scope and target year are therefore the facts currently confirmed.
NVIDIA invests in MediaTek and opens an NVLink Fusion path for custom XPUs
NVIDIA and MediaTek expanded their collaboration on 31 August. MediaTek will adopt NVLink Fusion as a design foundation for cloud providers, hyperscalers and model developers building custom XPUs that connect to NVIDIA's rack-scale NVLink systems. The disclosed technology includes the NVLink Fusion chiplet, NVLink-C2C, a custom memory architecture and integration with the MGX rack platform. The companies will also continue work on chips associated with RTX Spark and DGX Spark PCs and on automotive computing platforms.
NVIDIA is also investing US$3.5 billion in MediaTek convertible bonds. Reuters independently verified the amount and the platform-adoption arrangement. The announcement establishes the investment, collaboration scope and design path, but does not identify a customer chip, tape-out date, production schedule or measured performance. The agreement therefore cannot yet be described as delivery of a custom XPU.
Sources
AMD, AMD, Cisco and HUMAIN Expand Saudi Arabia’s AI Infrastructure as AMD Instinct Systems Go Live, 31 August 2026
https://ir.amd.com/news-events/press-releases/detail/1298/amd-cisco-and-humain-expand-saudi-arabias-ai-infrastructure-as-amd-instinct-systems-go-live
Telecompaper, AMD, Cisco and Humain say Saudi AI infrastructure goes live, 31 August 2026
https://www.telecompaper.com/news/amd-cisco-and-humain-say-saudi-ai-infrastructure-goes-live–1581231
Reuters, AMD, Cisco and Saudi's Humain launch AI joint venture, land first major customer, 19 November 2025
https://www.reuters.com/world/middle-east/amd-cisco-saudis-humain-launch-ai-joint-venture-land-first-major-customer-2025-11-19/
NEOM, HUMAIN partners with DataVolt to jointly develop 100MW of a 360MW AI data center project in NEOM, 31 August 2026
https://www.neom.com/en-us/newsroom/HUMAIN-and-DataVolt-to-develop-AI-data-center-in-NEOM
Reuters, Saudi AI firms Humain, DataVolt partner for data center on kingdom's Red Sea coast, 31 August 2026
https://www.reuters.com/business/energy/saudi-ai-firms-humain-datavolt-partner-data-center-kingdoms-red-sea-coast-2026-08-31/
NVIDIA, NVIDIA and MediaTek Deepen Long-Standing Partnership to Build AI Edge to Cloud Computing Platforms, 31 August 2026
https://nvidianews.nvidia.com/news/nvidia-and-mediatek-deepen-long-standing-partnership-to-build-ai-edge-to-cloud-computing-platforms
Reuters, Nvidia to invest $3.5 billion in chipmaker MediaTek, expand partnership, 31 August 2026
https://www.reuters.com/world/asia-pacific/nvidia-invests-35-billion-mediatek-convertible-bonds-2026-08-31/
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.