The emerging risks to Vision 2030, and who actually carries them

1. Where things stand

The budget, and the gap it is fighting. Per the Ministry of Finance, the official 2026 deficit is SAR 165.4bn, or 3.3% of GDP, or about $44.1bn. Debt reached 33.9% of GDP in the first half of 2026, and the borrowing plan for the year is 31% larger than the stated deficit, which tells you the government expects to raise more than the headline suggests it needs.

The breakeven disagreement is worth understanding properly, because it is usually reported as institutions failing to agree. They are not disagreeing. They are measuring different things. The IMF's $86.60 and Bloomberg's $94 cover the central government. The $111–113 figure adds PIF's domestic commitments. Both are correct for what they count, and quoting either without saying which perimeter it uses is how the number becomes misleading.

How sensitive the budget actually is. Per the register, Saudi Arabia exports around 6.33 million barrels a day, so every dollar on the oil price is worth roughly $1.8–2.8bn a year to the state. That is the multiplier behind every number in this article, and it explains why a $20 move in Brent is a different category of event here than in a diversified economy.

The shock absorber is nearly used up. Per the register, when oil disappoints the line that moves is capital spending. Salaries, subsidies and debt service do not flex. Capital expenditure is 98% of the deficit, and roughly 60% of that flexibility has already been spent. PIF contract issuance fell from about $71bn to under $30bn, a 58% cut. NEOM's residency target dropped from 1.5 million people to under 300,000, an 80% reduction.

Read that carefully, because it changes how every risk below should be weighted. The Kingdom absorbed the last shock by cutting the programme. It cannot absorb the next one the same way, because the cut has already happened.

The diversification is real. This should be said clearly, because the risks below are easier to list than the achievements. Per official national accounts, non-oil GDP has reached 55% in real terms. Oil's share of nominal GDP fell from 22.3% in 2024 to 17.1% in 2025. Female labour-force participation went from 17% in 2017 to 36.3% by early 2025, beating a target that was set at 30%. These are not presentational numbers. They are among the fastest structural shifts any large economy has achieved in a decade.

The difficulty is that success creates its own exposures. A fiscal base resting on the non-oil economy is a fiscal base resting on non-oil tax receipts, and those grew only 1–5% year on year in the first half of 2026.

2. The ten emerging risks

Ranked by how much damage the event would do, not by how confidently we can evidence it.

That distinction is worth defending, because it produces an uncomfortable-looking list. Four of the ten carry no defensible number, and three of those four sit in the top three places. We could have ranked by evidence quality instead and produced something that looked more rigorous. It would have put the measurable, modest risks at the top and buried the ones capable of ending the programme.

A register where every item carries a number is a register where some of the numbers are invented, and one invented number discredits the real ones. Where we are estimating rather than measuring, the entry says so in its own words.

2.1 A second corridor closes while the bypass is already full

Saudi Arabia absorbed the closure of Hormuz by running the East-West Pipeline at a record 7 million barrels a day. That used the entire capacity. If a second route closes, there is nothing left to reroute through, and the exposure stops being fiscal and becomes economy-wide.

Start with what happened, because it is genuinely impressive and it is the foundation of the risk. Iran shut the Strait of Hormuz on 28 February 2026. Crude shipping through Gulf ports fell 95% and LNG shipping fell 99%. For most exporters in the region that was the end of the conversation. Qatar has no overland option and cannot build one, because liquefied gas cannot travel by pipe.

Saudi Arabia had built for exactly this. The East-West Pipeline runs across the country to Red Sea terminals, and the Kingdom opened it to a level it had never previously needed. Exports continued. The insurance paid out.

Now look at what that leaves. A pipeline running at a record is a pipeline with no headroom. The redundancy did not partially absorb the shock and leave a reserve. It absorbed all of it, and is now fully committed. Analysts have already named the follow-on case: Bab al-Mandeb closing while Hormuz remains shut, which would put both of Saudi Arabia's export routes under pressure at once.

We are not going to attach a probability to that, because none exists and we would be inventing it. What we can say is that the standard reassurance, that Saudi Arabia proved it can route around a closure, becomes the opposite of reassuring once you notice the proof consumed the whole capacity.

The practical version of this risk for a finance ministry is simpler than the geopolitics. It is that the Kingdom currently has one working export route and no second one, and that condition is new.

2.2 Chinese demand falls, and diversification has made it worse

Per our own computation from JODI and Chinese customs reporting, China takes 28.4–31.6% of Saudi crude exports. A severe Chinese demand crisis would cost $46.0–51.2bn a year, or 104–116% of the entire annual deficit, from one customer. The diversification programme is deepening this exposure rather than reducing it.

The share is not published anywhere. We computed it from 1.8–2.0 million barrels a day against total exports of 6.33 million, using JODI and Chinese customs reporting. That an exposure of this size has no stated figure is itself worth noticing, and it is the reason most assessments of Saudi risk understate the concentration.

Here is why it is worse than a normal customer-concentration problem. If a large buyer reduces purchases, a seller usually loses volume but keeps the price, or discounts to hold volume. In a severe Chinese downturn, Saudi Arabia would lose both together, because China is large enough that its demand moves the global price. Volume and price fall in the same event, with nothing offsetting.

Then follow it through the budget, per the register. The revenue gap widens borrowing against debt already at 33.9% of GDP on a plan already 31% above the stated deficit. Borrowing absorbs some of it, capital spending absorbs the rest, except capital spending is 98% of the deficit and has already been cut 58%. Each step is fine on its own. Together they describe a system with less give than the headline ratios suggest.

The part almost nobody says out loud is the direction of travel. Diversification is supposed to reduce dependence on a single counterparty. In this case it is increasing it. The petrochemical joint ventures, the minerals offtake agreements and the downstream investments all point at China, the same customer that buys the oil. The Kingdom is not building a second customer base. It is building more ways to sell to the first one.

Two honest limits. The dollar range assumes roughly $70 a barrel; at a lower price it would be smaller. And a severe Chinese crisis is our scenario, not a forecast anyone has published. The exposure is arithmetic. The trigger is judgement.

2.3 A Taiwan contingency ends the AI strategy rather than delaying it

HUMAIN's named silicon partners are Nvidia, AMD and Qualcomm. All three fabricate at TSMC in Taiwan. There is no fabrication capacity in Saudi Arabia and alternative silicon takes 18 to 24 months to design. The same event would move the oil price funding the programme.

Most risk registers classify Taiwan as an indirect exposure for Saudi Arabia, and for oil that is correct, since the Kingdom sells crude rather than chips. For the diversification strategy it is direct, and the distinction matters enormously.

The AI programme is the centrepiece of the plan to build an economy that does not depend on hydrocarbons. Every accelerator in it is fabricated on one island by one company. Saudi Arabia cannot make them, cannot buy them from a second source at the required scale, and cannot redesign around them quickly. An 18-to-24-month design cycle is the floor, and it is a floor set by engineering time and foundry queues rather than by money, which means the Kingdom's greatest advantage, capital, does not shorten it.

Now add the second half, which is what makes this rank third rather than tenth. A cross-strait contingency would disrupt roughly $1.3tn of annual trade through the strait, a third more than Malacca carries. That is a global macro event, and it would move the oil price at the same moment it halted the chip supply.

So one event strikes both sides of the balance sheet. The diversification strategy loses its inputs while the hydrocarbon revenue funding that strategy loses its stability. Committed data-centre capital strands during the window. On an organisation chart these look like two unrelated risks sitting in two different portfolios. They are one risk.

No defensible probability exists for a Taiwan contingency and any analysis offering one is inventing it. We rank it here on consequence alone: this is not a programme that slows down. It is a programme that stops.

2.4 Oil stays below breakeven long enough to force a real choice

Every dollar below breakeven costs $1.8–2.8bn a year. Sustained, the gap runs to roughly $60–80bn, larger than the entire stated deficit, and the line that normally absorbs it has already been cut.

The mechanics are unusually simple. When oil disappoints, Saudi Arabia cuts capital spending, because salaries, subsidies and debt service cannot move quickly and capital projects can. That is a sound design and it has worked repeatedly.

It has now been used. Per the register, NEOM was cut by 80% on residency and PIF contract issuance fell 58%. Roughly 60% of the available flexibility is gone. The absorber for the next shock is materially thinner than it was for the last one, which raises the effective severity of every other negative item on this list.

What makes this an emerging risk rather than a description of the present is duration. One or two years below breakeven is a financing question and the Kingdom has the balance sheet for it. A sustained period forces a choice between three things that have so far all been protected: the remaining programme, the subsidy structure, and the debt trajectory. No published plan says which one gives way, and that silence is the risk.

One caution on the price data itself. The war made oil two-sided rather than simply cheap. Brent spiked to $114.97 on 4 May 2026 and fell 20.53% to $91.37 by 29 May, the worst monthly decline since the Covid demand collapse. Published 2026 figures for the same period do not agree well enough to build a single average from, so we quote dated prints rather than a year figure, and we would advise anyone modelling this to do the same.

2.5 The giga-project cut arrives at the banks before it arrives in the budget

NEOM's cut and the 58% collapse in contract issuance have already happened. What has not happened yet is the credit event that follows, when contractor distress turns into non-performing loans at Saudi banks.

This one is easy to miss because the cause looks like good news. A government reducing spending on a project it has decided to rescope is behaving responsibly, and the budget line gets smaller. On the government's own accounts, this reads as savings.

Follow the money outward and it looks different. An entire contractor ecosystem was built around a decade of state capital spending: engineering firms, equipment suppliers, labour contractors, logistics operators, and the banks that financed all of them. That ecosystem is now sized for a programme that no longer exists at the scale it was sized for.

The sequence from here is well understood from other markets. Receivables age. Contractors that geared up for the original volume default on facilities written on the original assumptions. Banks provision against a construction loan book, and provisioning reduces their capacity to lend into everything else.

So the fiscal cost of the decision does not appear in the ministry that made it. It appears in the banking system, one step removed and a year or two later, and it arrives as a credit question rather than a budget question. A giga-project that quietly halves is a solvency event for its suppliers long before it is a headline for anyone else.

This is the risk on the list with the earliest observable warning signal, which is why it is worth watching even though it ranks fifth on severity. Saudi bank provisioning against construction exposure would show it before anything else does.

2.6 Automation removes the jobs Saudization is counting on

The plan localises more than 340,000 private-sector jobs from 2026, in entry-level, administrative and service roles. Those are precisely the roles the AI programme is funded to automate, on the same timeline. The employment-to-population ratio has already peaked and reversed.

This is the finding we would most encourage a reader to sit with, because the two halves of it are usually assessed by different teams and never put side by side.

The demographic case for Saudi Arabia is genuinely strong. Per GASTAT, a young population, rising participation, and female labour-force participation that went from 17% to 36.3% in eight years, past a 30% target and within striking distance of the revised 40%. Against ageing peers in Europe and East Asia, this is a real and durable advantage.

The automation case is also strong, and it is being funded deliberately. The AI programme exists in part to raise productivity in administrative and service work, which is exactly what large-scale language and agent deployment does first.

Now put the timelines together. Saudization requires 340,000-plus private-sector jobs to be filled by Saudi nationals from 2026, concentrated in entry-level and service categories. The AI investment is scheduled to reduce the number of those roles over the same period. Both programmes are official. Both are funded. They are working on the same cohort in opposite directions, and no published document reconciles them.

The early evidence is not encouraging. Employment-to-population peaked in early 2025 and has since reversed, and the fall in unemployment stopped at 7%. That happened before automation was a meaningful factor, which suggests the labour market was already struggling to absorb the cohort.

For a finance ministry the consequence is a spending-side problem, and it compounds with the revenue-side risks above rather than offsetting them. Localisation that fails does not leave employment where it was. It pushes the state toward public-sector hiring and larger transfers, at a point when health and education already cost $74.9bn, 24.5% of revenue, and will not return tax receipts for roughly two decades.

Whether automation eats these specific roles is our inference rather than a sourced finding. The collision of the two timelines is not an inference. It is in both programmes' own documents.

2.7 A food price shock raises the subsidy base permanently

Per the register, Saudi Arabia imports around 80% of its food, at $20–25bn a year. A 31% price rise adds roughly $7.0bn to the subsidy bill, 15.8% of the deficit, and food subsidies are almost never withdrawn once granted. But the Kingdom is also a fertiliser exporter, which hedges part of this, and nobody is counting that.

The dependency is close to total in the categories that matter. Wheat is one third self-sufficient, corn is 89% imported and soybeans 95%.

The likely route to a shock is indirect, which is why it is under-watched. Up to 30% of internationally traded fertiliser normally crosses the Strait of Hormuz. Fertiliser does not reach food prices immediately. It reaches them through the following planting and harvest cycle, and that lag explains why the effect has not shown up yet, and it is not a reason to discount it.

The exposure is not the spike. It is the ratchet. A government facing a food price rise increases support, and food support is politically very difficult to withdraw once people have adjusted to it. The lasting cost is the new, higher baseline rather than the temporary peak.

Here is the part that is missing from every version of this analysis we have seen. Saudi Arabia is a net fertiliser exporter. Ma'aden's phosphate business is projected to grow from around $17bn to $70–80bn of GDP contribution by 2030. A fertiliser-driven food shock therefore raises the Kingdom's import bill and Ma'aden's revenue at the same time, from the same cause.

The net exposure is smaller than the gross exposure, and we could find no analysis that nets them. That does not make the risk go away. The subsidy ratchet lands on the budget while the Ma'aden upside lands on a state-owned company's income statement, and those are different balance sheets. But anyone sizing this risk from the import bill alone is overstating it.

2.8 The AI build-out costs the budget more than it returns

The programme targets 6 GW by 2034. At 80% utilisation that draws 42.0 TWh a year, twice the electricity consumed by every desalination plant in the country. The power is subsidised, the cooling water is desalinated, and the tax base is reduced by the exemptions that attracted the investment.

Begin with the scale, because it is easy to read 6 GW without feeling it. Forty-two terawatt-hours a year is double what Saudi Arabia currently spends producing all of its desalinated water, and desalination alone is already 6% of national electricity consumption.

Now trace the loop, because it closes on itself in a way that is specific to this country. The data centres need cooling. Cooling in Saudi Arabia needs water. Water in Saudi Arabia is desalinated. Desalination needs electricity. So the campuses draw power directly and then draw more power indirectly through the water they consume, in a system where water production is already one of the largest electricity loads.

Then look at who pays for each input. The electricity is supplied at subsidised rates. The water is produced with subsidised energy. The corporate tax that might recover some of the cost has been reduced by design, because the exemptions are what attracted the investment in the first place. And direct employment per dollar of capital spent is low, so the income-tax and consumption channels are thin too.

The conclusion is not that the programme is a bad idea, and we want to be careful here because this distinction is routinely collapsed. A sovereign AI capability can be strongly worth having for reasons that have nothing to do with its return to the treasury: strategic autonomy, industrial capability, talent retention. What should not happen is the strategic case being presented as evidence for the fiscal one. The Ministry of Finance can fund the enabling infrastructure and collect neither the full tariff nor the tax, while the Kingdom still comes out ahead. Both of those can be true simultaneously.

Delivery is the other half of the picture. Saudi Arabia has 467 MW of data-centre capacity actually energised, against a 6 GW target more than an order of magnitude larger. The awkward shape of this risk is that it only becomes acute if the programme succeeds.

2.9 The hydrogen plant has one customer, who is also the builder

All of NEOM's hydrogen output is committed to Air Products on a 30-year exclusive contract. Air Products is also the EPC contractor building the plant. There is no second route to market, because the structure never created one.

The facility is $8.4bn, producing 600 tonnes of hydrogen a day, equivalent to 1.2 million tonnes a year of ammonia, from 2027, with up to 4 GW of dedicated solar and wind behind it. It is genuinely one of the most ambitious green hydrogen projects anywhere.

It is presented as buyer-side validation, and that reading is fair. Securing a thirty-year offtake for the entire output of a first-of-its-kind plant, before it is built, is a real commercial achievement in a market where most hydrogen projects never reach final investment decision.

Read the identical fact from the other direction and it is total counterparty concentration. The project has exactly one commercial relationship. The price, the volume, the credit and the construction all sit inside it. Air Products building the plant it will be the sole customer of is efficient, and it also means there is no independent party in the structure whose interests differ.

Both readings are true of the same fact, and we carry both rather than choosing. What we would flag is what happens if that single counterparty renegotiates, defers or fails. Green hydrogen markets are still thin enough that replacing a buyer at this scale is not a matter of accepting a worse price. It is a question of whether an alternative buyer exists at all.

Two things soften it. The project is an ACWA/Air Products/NEOM joint venture rather than a Ministry of Finance obligation, so the direct fiscal exposure is indirect. And by the programme's own statement, it returns no revenue to the ministry before 2030 in any case. At 19.4% of the capital programme this is not the largest exposure on this list. It is the most concentrated, and concentration is what turns a manageable loss into an unrecoverable one.

2.10 Committed capital strands as the climate makes the work harder

MENA is a recognised climate hotspot and the giga-projects are outdoor construction. As heat and drought intensify, the working window narrows, but the capital commitments are contractual, so the spending holds while the delivery slips.

This is the least concrete risk on the list and we have ranked it last for that reason. It is included because the mechanism is real and because it is almost entirely absent from how these projects are assessed.

Large construction programmes in the Gulf already work around summer heat. The question is what happens as the window narrows further. Outdoor work becomes slower and more expensive per unit delivered, and the schedule extends. Meanwhile the capital commitments behind those schedules are contractually fixed. The money is committed on a timetable the physical work can no longer meet.

The result is a widening gap between spend and delivery. That is the same logic the Kingdom applies to event hosting, where the venue must exist before the event generates revenue. Applied to giga-projects it means capital sits in partially delivered assets for longer, earning nothing.

There is a second-order effect worth noting. Rising heat raises cooling demand across the whole economy, which raises electricity demand, in a system where desalination is already 6% of national consumption and the AI programme is proposing to add twice that again.

We cannot size this. Doing so would require the giga-project schedule book, which is not public and which we do not hold. We would rather say that plainly than produce a number that looks like analysis.

3. What this means

The exposure is off-budget. The most consistent finding across these ten is that the largest items sit at PIF rather than the Ministry of Finance. A finance ministry reading its own accounts will not see NEOM, HUMAIN or the giga-project book, and will not see the seventeen dollars a barrel separating the two breakevens.

The two biggest risks share a counterparty. China is the demand side of one and the actor in another. They are not independent draws, and a register treating them separately understates the correlated case.

The deficit is squeezed from both ends at once. China and the oil price work on revenue. The labour market and the food subsidy work on spending, and both are ratchets rather than cycles. Nothing in the standing position absorbs that, because the shock absorber, the capital programme, is the line that has already been cut.

What to watch. Three things would move this assessment: a published reconciliation of the two breakeven perimeters, quarterly MCIT energised-capacity data against the 6 GW target, and Saudi bank provisioning on construction exposure. The third is the earliest warning of the risk nobody is currently pricing.