Two Clocks in Riyadh, One Deficit

On 3 December 2025, on the eve of the 2026 budget release, Saudi Finance Minister Mohammed al-Jadaan told Bloomberg something Riyadh does not usually put on the record. "We have no ego — absolutely no ego," he said. "If we announce something and we need to adjust it, accelerate it and make it a priority more than others, or defer or cancel it, we will without blinking." He was talking about the megaprojects. The budget he delivered the next day, with SAR 1.313 trillion in planned outlay, a projected $44bn deficit, and no line-item name-check for NEOM or New Murabba, showed he meant it. What al-Jadaan did not spell out is that the same sentence applies to every line on the ledger, including the ones with children in classrooms and doctoral candidates at scholarship desks. The AI workforce build sits on that ledger too.

Two clocks, one purse

The macro problem you get when you read the Saudi and Emirati AI books together is that they are running on two different clocks against one balance sheet.

Clock A is the model-and-datacenter clock. The physical build — Aramco Digital's Groq deployment, the Humain–Nvidia campus, MGX's stakes across the American AI stack — has vendor contracts already in ink, cash-payback windows in the two-to-four-year range on the more sober use cases, and dedicated political sponsorship at Crown Prince level. Those workloads have a way of being defended in a fiscal squeeze because the vendor invoices do not defer politely.

Clock B is the human-capability clock. A six-year-old learning the SDAIA-aligned national AI curriculum rolled out for the 2025–2026 academic year does not enter the labour market until 2038. An MBZUAI PhD taking a scholarship desk in Fall 2025 is a research hire in 2029 at the earliest. SAMAI's million-plus workforce-literacy participants become useful in different roles at very different depths at very different moments. Human-capital returns compound; they do not print in a quarter.

Both clocks are being wound in Riyadh and Abu Dhabi at the same time, and the current fiscal position was not designed to run both to term without triage. That is the macro question worth sitting with today.

What is on the Saudi ledger?

Read the workforce commitments as line items rather than as a marketing story and the arithmetic gets specific.

The SDAIA Academy, per Saudi Press Agency reporting, has put more than 500 experts through advanced data-engineering training and 215 students from 28 universities into a live cohort in Riyadh, alongside a broader Ministry of Communications and Information Technology commitment to certify 100,000 citizens in AI and data skills. Sitting behind those numbers is the longer-term SDAIA target of roughly 20,000 deep AI specialists by 2030, first codified in the National Strategy for Data and AI, against a baseline where the Kingdom currently graduates around 2,500 to 3,000 specialists in that band per year.

Layer the education-system side. The nationwide AI curriculum, as set out by the Ministry of Education and SDAIA, is expected to reach more than six million students beginning the 2025–2026 academic year, with tens of thousands of teachers trained and a national assessment framework built to score it. The Microsoft skilling commitment announced on 12 February 2026 targets three million Saudi citizens by 2030 through partnerships with the Human Capability Development Program and Tuwaiq Academy. That is a vendor number, so apply the salt cellar, but even discounted by half it is a serious build.

Not all of that is education spend on the state's books. But a larger chunk of it than external readers assume is: certification programmes, teacher training, curriculum development, university partnerships, research chairs. Every one of those is a multi-year cash outflow with a benefit stream that materialises well after the current budget cycle ends.

Abu Dhabi's version

Abu Dhabi is running the same play at a smaller scale and a tighter pipeline. The Mohamed bin Zayed University of Artificial Intelligence graduated 104 students in July 2026, 13 PhDs and 91 master's, its largest cohort to date, including 20 Emiratis and the university's first Emirati PhD graduate, Salem AlMarri. The Fall 2025 intake added 403 new students, bringing the enrolled base above 700. Of MBZUAI's 316-strong alumni network, roughly 80% remain in the UAE inside a year of graduation, flowing into G42, Inception, M42, TII, and adjacent operators.

That is a real research pipeline. It is also, in absolute terms, small: a few hundred deep researchers a year on top of an alumni base of just over 300, feeding an ecosystem whose own headcount is measured in tens of thousands. The gap between the top of the pipeline and the working population it is supposed to underwrite is exactly the middle-tier training layer, and that layer is not being built inside MBZUAI. It is being contracted out to the same set of vendors carrying the load in Riyadh.

Where the fiscal squeeze will land first

The pressure point is not hidden. Saudi's Q1 2026 deficit alone came in at SR125.7 billion, about $33.5bn, nearly the full-year projection eaten in a single quarter as oil hovered under $65 and spending rose 20% year on year. The IMF's December 2025 Article IV commentary welcomed the recalibration in explicit terms; the Fund's staff were, in effect, telling the Ministry of Finance that the delay is prudent and the market will not punish it.

Given that setup, ask which lines get protected when the next round of triage comes.

The datacenter workloads, most of them, are locked in by contract, geopolitics, and Nvidia's queue. The physical megaprojects have already taken the visible hit. The AI-in-curriculum line for six million students is the most politically defensible — no one in Riyadh wants to be the minister who cancels schoolchildren's AI textbooks — but it is also the most technically deferrable: the framework can be paused, teacher training can slow, the assessment build can push out a year, and none of it prints on a market screen. SDAIA's specialist tracks and the Microsoft skilling programme sit in the middle tier, easiest to slow quietly, hardest to explain if pressed. MBZUAI is protected by both political sponsorship and modest absolute cost.

My reading is that the education-and-training clock is the one most likely to slip on the government side, offset partially by more foreign-vendor content — Microsoft, Nvidia, Coursera, and the Chinese partners quietly picking up market share — carrying the load. The observable signal will not be a headline cut. It will be extended timelines, unannounced deferrals, and a shift from Saudi-branded to vendor-branded programmes. Look for that in the second half of 2026, not the first.

The IMF's contrary read

Fairness demands the strongest version of the counter-case on the table. The IMF Staff Discussion Note on Gen-AI and the labour market argues, credibly, that productivity gains from AI diffusion can lift income levels enough to make the training investment pay itself back within a decade in economies that adopt aggressively. Oxford Economics's analysis of the Gulf AI value chain argues that the region's compute-plus-capital advantages create exactly the conditions where the workforce build compounds the datacenter build. Both are serious pieces of work, and their central point, that human capability is the return-multiplier on the physical spend, is one I would defend.

The place I would push back is the confidence in the pass-through. The IMF paper models diffusion in the abstract; the Gulf's actual pipeline is thin at the top and vendor-dependent in the middle. If the diffusion assumption slips by even a couple of years, the compounding case gets pushed out beyond the fiscal horizon most GCC finance ministries are planning to. That is the honest disagreement.

Note from a workforce planner in the Gulf

If I were running the workforce line at a Saudi ministry or an Abu Dhabi authority this quarter, my prescription would be narrow and specific.

I would protect the six-million-student curriculum line first, because it is the only line on the ledger whose full return arrives after every current politician has retired, and it is therefore the one most likely to be quietly starved. I would ring-fence teacher training at the current run rate for at least three years, on the argument that model imports can wait but a generation of teachers cannot be retrained twice. I would consolidate the middle-tier certification programmes, SAMAI, the MCIT 100,000-cohort, the Microsoft three-million ambition, into a single measured pipeline with published outcome data by 2027, on the argument that overlapping vendor programmes without shared assessment is the shape most likely to be cut when a spreadsheet gets opened. And I would double MBZUAI's PhD intake for the next five years, because a hundred-a-year research pipeline is not enough to underwrite the sovereign ambitions being announced at Crown Prince level, and if this build is real, the bottleneck is not compute or capital. It is the eighty people who will supervise the next generation of the guests coming through the door.

That is what I would defend the budget line for. The two clocks are not going to sync themselves, and if the ministries let them drift apart, they will find, sometime around 2030, that the compute came in on time and the people did not.

Tarry Singh is the founder and CEO of Real AI, an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan for Energy AI, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.