The LEAP 2026 announcements TechScoop is tracking

Category: Events

By Mo

Published: 2026-09-01T16:00:00.000Z

A running guide to the newest LEAP 2026 announcements and what they add up to: industrial AI moving into plant, HUMAIN filling in its stack, an operator scramble up the value chain, and localisation reaching software.

The second half of LEAP 2026 has produced a different kind of announcement from the first. The opening days delivered the headline numbers — cloud regions with dates attached, gigawatt ambitions, capital commitments. What has followed is narrower, more operational and, for anyone trying to work out what the Kingdom's technology sector will actually look like in two years, more informative. This is a running account of those announcements and what they add up to. It will be updated as more land before the show closes on 3 September. The clearest theme is industrial AI moving out of the demonstration hall and towards the plant. Aramco Digital used the show to set out a national industrial communications network running on the 450 MHz band , licensed by the Communications, Space and Technology Commission and built for factories, pipelines, utility grids and field crews rather than consumer phones. It is the least glamorous announcement of the week and possibly the most consequential, because it addresses the constraint that quietly limits every industrial AI project: models are useless without data, data comes from sensors, and sensors in remote or shielded industrial environments have historically had nothing reliable to talk to. A dedicated low-band network with national reach changes what is possible to instrument. The rest of Aramco's week fits that logic. Its technology arm announced a partnership with Avathon on industrial AI spanning Aramco's own operations, the Saudi market and global industrial customers — a proposition that rests on something few AI vendors possess, namely decades of instrumented operating history across an enormous equipment fleet. Its parent separately signed a memorandum with Microsoft covering industrial AI and a roadmap for cloud services under enhanced sovereign controls . Connectivity, models and compute, assembled as one argument rather than three procurement exercises. HUMAIN, meanwhile, has spent the week filling in the parts of its stack that were still missing. The PIF-owned company came into LEAP with capacity — AMD Instinct systems live with Cisco , 100MW under construction with DataVolt at NEOM , 250MW planned with Together AI — and two obvious gaps: what runs on it, and who builds the machines. Both have now been addressed. Reflection AI's open frontier models are to be deployed on HUMAIN's Saudi infrastructure once the lab ships its first model, with the two exploring customisation, a sovereign neocloud arrangement and compute offtake. And Compal has been named long-term ODM partner for successive generations of AI servers on both NVIDIA and AMD silicon, extending from data-centre systems out to edge and physical AI devices. Set alongside the operating system work with KORA and the AI PC built with Qualcomm , the outline of a full stack is now visible: silicon partners, manufacturing, data centres, operating system, models, devices. Whether it coheres into a business is a separate question from whether the pieces exist, and the honest answer today is that the pieces exist. The open-model choice deserves more attention than it has had. Hosting a closed model in a region is a hosting arrangement; the provider still owns the system. Running open weights on infrastructure you operate is a different proposition, and it is the only version of sovereign AI that survives contact with a regulator's questions about who can reach the data. That HUMAIN is doing this while also carrying its own ALLaM models — which Microsoft agreed to distribute globally — suggests a strategy of being the place regional customers come for whichever model suits the job, rather than a single-model bet. The data-centre pipeline continues to lengthen. Zain KSA and DataVolt signed a memorandum to co-develop facilities across the Kingdom , aimed at enterprise, government and cloud-provider demand rather than hyperscale AI campuses — a different and probably under-served part of the market, since not every workload wants to sit in a 100MW facility in the north-west. It should be read as what it is: an MoU with no site, no capacity figure and no completion date. Announced pipeline across the Kingdom is now growing considerably faster than grid connections to serve it, and proposed capacity is not operating capacity. That gap is the single most under-discussed risk in the Saudi AI build-out. In telecoms, the interesting movement has been towards networks that run themselves. Mobily and Ericsson signed a memorandum covering AI inside the radio access network , using rApps on Ericsson's automation platform, aimed at Level 4 autonomy in selected use cases — the level at which a network monitors, decides, acts and verifies without a human approving each step. The pair have a real basis for the claim, having demonstrated Level 4 automation during Hajj 2025, which is about as unforgiving a test environment as exists. Mobily is running a parallel programme with Nokia, having signed on AI-powered network efficien