Saudi Vision 2030's AI reality check: what has actually shipped

Three years on from the season of announcements that made Saudi Arabia the loudest voice in global artificial intelligence, the right question is no longer what the Kingdom intends to build. It is what it has actually built. Vision 2030 enters its final five-year phase in 2026 with a scoreboard that, depending on who is holding the pen, reads either as triumph or as a quiet reset dressed in the language of triumph. Both readings contain truth. The work of an investor, a policy analyst or an executive deciding where to commit is to separate the two, and to do it before the next summit drowns the signal in another wave of capital commitments.

Start with the official number, because it is the one everyone quotes. Saudi Arabia's 2025 Vision 2030 annual report claims that 93 per cent of its key performance indicators met or exceeded their annual targets. That figure is not a lie. It is, however, a statement about a system that sets and revises its own targets, and it tells you less than it appears to. A KPI that is met after being quietly lowered has still, technically, been met. The honest reading of Vision 2030 in 2026 is that the diversification thesis is working in the places that were always going to be easier, and stalling in the places that were always going to be hard. AI sits squarely in the hard category, which is precisely why it deserves scrutiny rather than applause.

The scoreboard, read honestly

The genuinely good news is real. Non-oil activity now accounts for 55.6 per cent of real GDP, up from 45.4 per cent when the plan launched. Tourism blew through its original target of 100 million annual visitors years early, prompting the headline ambition to be revised upward to 150 million by 2030. These are not PR confections. They reflect a country that has changed materially in a decade.

Then there is foreign direct investment, the metric that matters most for any technology ambition because it measures whether outsiders are willing to put their own money behind the story. Here the gap is awkward and worth stating plainly. The 2030 target is 100 billion dollars of annual inflows, against roughly 8 billion when the plan launched. FDI has grown, but it remains stuck near 2.8 per cent of GDP against a 3.4 per cent target for 2025 and a 5.7 per cent target for 2030. A great deal of the capital flowing into Saudi AI is the Kingdom's own, recycled through the Public Investment Fund. That is a legitimate strategy. It is not the same thing as the world voting with its wallet, and an analyst who conflates the two will misprice the risk.

NEOM, and the most expensive lesson on the board

No part of Vision 2030 has travelled further from its original promise than NEOM, and no story better illustrates the difference between ambition and delivery. The Line, the 170-kilometre mirrored linear city that became the project's global symbol, has been cut back to a pilot segment of roughly two kilometres, with the wider vision deferred until after 2030. The 2030 residency target, once spoken of as 1.5 million people, has been reduced to as few as 100,000. The restructuring goes deeper than the headline asset. Oxagon has been moved toward Aramco, Trojena to the Ministry of Sport and Sindalah to Red Sea Global, an unwinding that amounts to NEOM being taken apart and parcelled out rather than scaled.

The official framing, from PIF governor Yasir al-Rumayyan, is that no projects have been cancelled and spending priorities have merely been reassessed. That is technically defensible and politically necessary, but it should not anaesthetise anyone. At the FII PRIORITY summit in Miami, PIF set out a five-year strategy that cuts capital spending by around 15 per cent and redirects it toward AI infrastructure, pharmaceuticals, renewables and a pipeline of public offerings. The lesson here is the one that matters for the AI thesis: when a sovereign with this much capital meets physical and financial reality, it bends. It is reasonable to expect the same gravity to act on the AI build-out, however differently it is being marketed.

There is an irony worth sitting with. The thing rising on NEOM's old footprint is compute. Oxagon is being repositioned as a data-centre campus, anchored by a 5 billion dollar, 1.5-gigawatt partnership with DataVolt whose first 300-megawatt phase is targeted for 2028. A data centre needs fibre and power, not a utopian skyline, which is exactly why it is a more honest bet than a linear city. The pivot from real estate spectacle to compute infrastructure is the single most telling move in the 2026 reset, and on the whole it is the right one.

Humain: the capital is real, the returns are a promise

The clearest expression of the new strategy is Humain, the PIF-owned and crown-prince-chaired AI champion launched in May 2025. Its ambition is to make Saudi Arabia the third-largest AI provider on earth, behind only the United States and China. The capital and the silicon behind that ambition are not vapour. Nvidia has confirmed an initial tranche of 18,000 GB300 chips with several hundred thousand more in the pipeline over five years, made possible by the November 2025 US decision to green-light advanced chip exports to Gulf partners. First data centres in Riyadh and Dammam, each around 100 megawatts, were scheduled to go live in the second quarter of 2026, with a roadmap to 1.9 gigawatts by 2030 and 6 gigawatts by 2034.

This is where an executive needs to hold two thoughts at once. The infrastructure is being poured, racks are being installed, and Humain has begun deploying over a thousand Qualcomm AI accelerators alongside its Nvidia and AMD commitments. That is genuine progress, not a render. But CEO Tareq Amin's framing, that the company will build in one year the capacity Saudi Arabia built in twenty, is a statement of intent, not an outcome. Capacity is not the same as utilisation, and utilisation is not the same as profit. The harder questions are who the paying customers are, at what margin, and whether demand for sovereign Gulf compute materialises at the scale the capex assumes. Until those answers exist, Humain is best read as the largest and best-funded bet of its kind in the region, with the returns still firmly in the future tense.

SDAIA and Allam: the regulator that is also a builder

Beneath the headline projects sits the institution that may matter most, and whose design carries a tension the Kingdom has not fully resolved. The Saudi Data and AI Authority, SDAIA, runs a dual mandate through two arms: the National Data Management Office for governance and the National Center for AI for research. In November 2025 its AI Adoption Framework moved it from advisory body to de facto regulator for the public sector, and the Cabinet's designation of 2026 as the Year of AI gave that authority a national stage.

The concern is structural rather than partisan. SDAIA writes the rules and builds the products. ALLaM, the sovereign Arabic-first large language model, was developed inside SDAIA's research arm and first launched on IBM watsonx in 2024; its commercial life now runs through Humain as the distribution channel. A body that both regulates AI and ships the flagship national model has every incentive to grade its own homework leniently. Other jurisdictions separate these functions for exactly that reason. The Saudi answer, that speed demands integration, is coherent for a country trying to move quickly. It is also the kind of arrangement that looks efficient in the build phase and risky in the accountability phase, and outside investors should price that governance question rather than ignore it.

The one that is quietly working

For all the scepticism the megaprojects invite, there is a Vision 2030 AI story that has shipped, and it belongs to the company that least needs the publicity. Saudi Aramco has put artificial intelligence into the parts of its business where the value is measurable. Its industrial model, Aramco Metabrain, is trained on close to nine decades of operational data and is used to optimise drilling paths and well placement. Aramco reports AI-driven reductions in unplanned downtime, maintenance cost and gas flaring across its operations.

Those numbers come from the company itself and deserve the usual caution that self-reported figures invite, but the pattern is credible because it is unglamorous. Aramco is not trying to become a sovereign cloud or a foundation-model champion. It is using AI as a tool inside a business it already understands better than anyone, and that is why it works. The contrast is instructive. The further a Vision 2030 AI initiative sits from an existing revenue stream and a concrete operational problem, the wider the gap between announcement and outcome tends to be.

The constraints capital cannot buy away

Two limits will shape the next five years more than any funding round. The first is talent. Saudi Arabia leans heavily on foreign specialists who command high salaries and do not always stay, and analysts point to an AI hiring gap of roughly 50 per cent in unfilled roles. Money can import expertise; it struggles to root it. The second is physics. Running and cooling gigawatts of compute in one of the hottest climates on earth raises water and efficiency questions that remain unresolved, even granting the Kingdom's genuine advantage in cheap energy. These are not reasons the strategy fails. They are reasons it cannot be willed into existence by capital expenditure alone, and they are where the credible execution risk lives.

The verdict

So what has actually shipped, three years on? More than the cynics allow and less than the brochures promise. The diversification of the wider economy is real. The AI infrastructure is being physically built rather than merely announced, and the 2026 pivot from architectural spectacle toward compute is the most clear-eyed decision the programme has made. Aramco's deployments are working. Those are not small things.

But the headline ambitions, third-largest AI provider, a sovereign model ecosystem, gigawatts of profitable compute, remain promises whose proof lies beyond 2030, funded substantially by the Kingdom's own balance sheet rather than by a market that has been convinced. NEOM is the standing reminder that Saudi targets are revisable and that the gap between a render and a ribbon-cutting can be measured in years and tens of billions. My own read is that Vision 2030's AI chapter will be judged not by the capacity Humain installs but by whether anyone outside the Kingdom chooses to run their most important workloads on it, and whether the talent base deepens enough to operate what the capital has bought. On present evidence those questions are open. Anyone telling you they are settled, in either direction, is selling something.