Saudi technology localisation has moved from offices to factories, and now to software

Saudi technology localisation has moved through three phases, and the announcements at LEAP 2026 sat across all of them. The first phase was presence. Regional headquarters programmes brought international technology companies to establish licensed entities in the Kingdom, which created local employment, local tax presence and local accountability. It did not create capability. A regional office sells and supports products designed and built elsewhere. The second phase is manufacturing, and it is the one currently most visible. Lenovo unveiled the first laptop manufactured in Saudi Arabia at LEAP 2026, produced in Riyadh at a facility developed with Alat covering around 200,000 square metres at Riyadh Integrated near King Khalid International Airport, designed for laptops, desktops, servers and Motorola smartphones, and described as the company's largest plant outside China. Mass production of a limited range is due to begin in late 2026, and the first cohort of Saudi engineers has completed specialised training at Lenovo's global facilities. The third phase is the one that has arrived quietly and matters most. HPE has been extending its Saudi Made, Developed and Deployed initiative from hardware into software, naming ShaheenAI, MOZN and OmniOps as founding software partners contributing locally developed platforms, and signed a memorandum with OmniOps at LEAP on a sovereign AI stack combining HPE's Saudi-made infrastructure with OmniOps' platforms, anchored on Bunyan, its inference platform that can run across cloud, on-premises and air-gapped environments. The value capture across these three phases is very different, and it is worth being explicit about why. An office captures a sales margin. An assembly plant captures manufacturing labour and some logistics, while the design, the components, the intellectual property and most of the margin remain elsewhere. Software captures the intellectual property itself, because in software the product and the intellectual property are the same object. A locally built inference platform, sold to local customers, developed and maintained by local engineers, keeps almost all of the value in the country. Which is why the software phase is more consequential than the factory, and gets a fraction of the attention. A plant is photogenic and a codebase is not. The verification problem inverts across the same sequence, and this is the harder point. A regional headquarters is trivially verifiable: the entity exists or it does not. Manufacturing is verifiable in principle through local content percentages, though neither Lenovo nor Alat has published one, and the honest description of the Saudi laptop on current information is assembly of a product designed elsewhere with the depth of localisation still to be established. Software localisation has no equivalent measure at all. There is no accepted way to state what proportion of a platform was built in a country. Was the architecture designed locally or adapted? Are the engineers employed here or contracted from elsewhere? Does the product depend on foreign components, models or services that would break if access were withdrawn? A company can describe a platform as locally developed while most of the engineering happened somewhere else, and nobody outside can currently check. OmniOps' air-gapped capability is interesting precisely because it is the one claim in this space that is testable. A system that runs with no external network connection either does or does not. That is a demonstrable property rather than an assertion about where code was written, and it is the kind of claim procurement can actually evaluate. The dependency underneath all three phases is unchanged and worth naming without alarm. Every accelerator in every Saudi AI deployment is designed and manufactured abroad, under export control regimes outside Saudi influence. No amount of assembly or software development changes that in the near term, and no country outside a handful has solved it. The realistic goal is not independence but capability plus optionality: enough domestic skill to operate, adapt and maintain, plus enough diversity of supply that no single decision elsewhere is decisive. HUMAIN running NVIDIA and AMD estates in parallel is that logic applied to procurement. The measures that would make the third phase legible are not exotic. For any platform described as Saudi developed: how many engineers work on it and where are they employed, what proportion of the codebase is written in the Kingdom, which foreign components or services does it depend on, and can it operate if those become unavailable. Four questions, all answerable by the company, none of them currently asked in announcements. The factory has a date in late 2026 that can be checked. The software phase has no date and no measure, which makes it both the more valuable phase and the easier one to claim.