The Hospital Saudi Arabia Has Not Built Yet
Most health systems are spending this decade digitizing a care model shaped around the limits of human throughput. One country is still deciding a large part of its model. Why the conditions converge here, and how quietly the window closes.
Enterprise AI Series · CXO Intelligence Series · All writing
The argument in 30 seconds
- The crux. Almost every health system on earth has already decided what its hospitals are for, how its clinical roles are divided and how care gets paid for. Saudi Arabia has not finished deciding any of the three, and that is a condition, not an achievement.
- The optionality is real and it is perishable. It cannot be appropriated, banked or accelerated with capital. It is spent silently, and the spending looks like ordinary competent work.
- The scarce resource is not headcount. It is qualified human attention at the moment it matters, and almost every health system in the world spends that attention on moving information, because information does not move on its own.
- Saudi Arabia is unusual in timing, not in ambition. Capital, estate, workforce composition, purchasing reform, virtual care at national scale, a device regulator with a published AI position and a national AI company with healthcare among its named sectors are all in motion together. The simultaneity is the opportunity. No single one of them is.
- The workforce reframe. A meaningful share of the people who will operate the 2035 system have not been recruited, so the composition of that work is still open to decision, not inherited. This is a redesign argument and not a headcount-reduction argument.
- What is already real. More than 16 million virtual appointments and consultations in 2025 across a network of over 242 hospitals, and two domestically developed AI medical software products authorized by the SFDA on 10 September 2026. The care model and the regulatory route both exist.
- The clock. The window does not close on a date. A design decision becomes a staffing assumption, which becomes a workflow, which becomes a technology requirement, which becomes a contract, which becomes an installed base. It closes one commissioning decision at a time.
01Why here, and why now
The previous article in this series argued that enterprise AI has not moved earnings because the work was shaped around the constraint of human throughput, and almost nobody has changed the shape of the work now that the constraint has loosened. Accelerate a step inside a process built for people and the saved time pools in front of the next queue. The adoption data says the same thing from the other side. Near universal reported use of AI in large organizations has not produced a matching shift in enterprise level results, and agent deployment remains a small minority against a very large majority who say they are planning one.1
That argument has an uncomfortable corollary. If the binding problem is accumulated process debt, the organizations best placed to capture the value are the ones carrying the least of it. Membership of that list has nothing to do with ambition, budget or technical sophistication. It is a question of what has already hardened.
Saudi Arabia belongs in this discussion not because it is better at AI than anyone else, and not because its health system is in better shape than anyone else's. It belongs because an unusual number of structural variables are in motion at once, and each of them is fixed almost everywhere else.
Capital is being committed now, and the envelope is tightening. Health and social development is the largest single allocation in the 2026 budget at SAR 259 billion (about USD 69 billion), against total expenditure of SAR 1,313 billion (about USD 350 billion).2 Two things about that figure are usually misread. That line is a sector envelope. It covers social security and welfare alongside healthcare, and it came in below the 2025 estimate, not above it. A tightening envelope makes the cost of commissioning the wrong operating model larger, not smaller.
A meaningful share of the 2035 estate does not exist yet. In a mature system the buildings are standing and the only available question is how to modernize them. Here, facilities are still being planned, designed and commissioned, which means their bed counts, clinic footprints and staffing establishments are still numbers in a document, not concrete and contracts.
The composition of the future workforce is still open. The Kingdom registers around 800,000 health practitioners across all health professions, with Saudi nationals accounting for roughly 57 percent of them, and physician Saudization rising from 31 percent in 2018 to 40 percent in 2026 on that register.3 A narrower official count for 2024, covering five professions, records 129,772 physicians at 36.8 per 10,000 population and 243,336 nurses, with Saudi nationals at 42.5 percent of physicians.4 The two figures use different denominators and both are cited here because neither alone describes the workforce. These are large numbers growing quickly. Nobody should read this as a system standing empty. It is a system still filling, and what the new entrants will be hired to do has not been settled.
Purchasing has been separated from provision. The 2022 Cabinet decision established the Health Holding Company to run facilities and regulated the Center for National Health Insurance to purchase care for citizens, with the Ministry of Health moving toward a regulatory role.5 The Council of Health Insurance has selected AR-DRG as the national case-mix classification and is running a structured adoption program under its 2025 to 2027 strategy.6 Precision matters here, because this is where commentary overreaches. This is not a single payer system and no binding national timetable for DRG implementation has been published. What has happened is structural and it is enough. The entity that pays and the entity that delivers are no longer the same, and the classification that lets payment follow outcomes has been chosen.
The burden is chronic, concentrated and identifiable in advance. A nine year analysis of nearly 75,000 adults attending care in the Kingdom found obesity across all classes in 46.4 percent of the study population, with type 2 diabetes rising from 34 percent among those of normal body mass index to 56 percent in obesity class three.7 These describe adults attending care, not national prevalence, and read that way they remain the strongest argument for continuous modeling over episodic care. The trajectory of a chronic patient is knowable long before the admission that makes it expensive.
No one of these conditions is remarkable. Several countries have capital, several have a workforce gap, several have purchasing reform. What is unusual is that all of them are unresolved at once, in one decade, under a single national program.8 Design opportunities of that kind expire.
The question is not whether Saudi Arabia can build a modern health system. It plainly can, and it already is. The question is whether it builds the one that already exists elsewhere, five years newer.
02What has already been demonstrated
There is a lazy reading of the Gulf in which ambitious things are announced and quietly do not happen. In this case the reading is refuted by two facts that are worth stating precisely, because imprecision here is what makes a serious reader stop trusting an argument.
Virtual care is operating at national scale. Seha Virtual Hospital, operated by the Ministry of Health, recorded more than 16 million virtual appointments and medical consultations during 2025. Within that, more than 11.5 million virtual clinic appointments were completed, up 56 percent on the previous year, and more than 220,000 clinical cases were managed, up 49 percent.10 The published network figure is more than 242 connected hospitals across 48 main specialties and 68 subspecialties, with annual capacity above 597,000 beneficiaries.11 In October 2024 it was recognized by Guinness World Records as the world's largest virtual healthcare provider, as reported by the Saudi Press Agency.12
I am deliberately not claiming this makes the Kingdom unique. I do not have the comparative dataset that would support such a claim, and neither does anyone else writing about it. What the numbers support is narrower and sufficient. Saudi Arabia has demonstrated at population scale that specialist clinical work can be delivered without colocating the specialist and the patient. As premises go, that one is proven and not aspirational.
The regulatory route for domestic clinical AI has been walked end to end. On 10 September 2026 the Saudi Food and Drug Authority granted marketing authorization to two Saudi-developed AI-enabled medical software products, Dental IQ for analysis of dental radiographs and SAARIA for retinal image analysis in diabetic retinopathy screening. Both came through the authority's Innovative Medical Devices Pathway.13
The SFDA's framing of what it authorized matters more than the authorization. The authority stated that such outputs are intended to support clinical decision-making and do not replace the healthcare professional's clinical judgment
, with healthcare professionals responsible for reviewing and validating outputs.13
A regulator has drawn an autonomy boundary in public, with products on the market behind it. Two things follow. The care model and the regulatory pathway both exist. And the boundary drawn is a floor for one class of device, not a schedule covering everything a health system decides. Nobody has written that schedule, and it turns out to be the load-bearing gap in the entire proposition.
03The workforce question is a design decision
The most cited number in Saudi health planning is a projected requirement for around 175,000 additional healthcare professionals by 2030. It needs careful handling, because the argument here does not depend on it and would be weaker if it did. The figure comes from consultancy modeling published in 2023, and its stated base of roughly 232,000 is inconsistent with official counts, which record about 460,000 people across the five professions measured in 2024 and around 800,000 practitioners on the national register.1443 No Saudi government body has published an equivalent gap figure. Treat it as an order-of-magnitude planning signal, not a workforce forecast, and notice that the argument survives at half that number or twice it.
The planning condition it reveals is what matters. A large cohort of the people who will operate the health system of 2035 has not been recruited, trained into a role definition or placed in an establishment. Their work can still be described. It has not yet been inherited.
Consider why AI-native redesign fails in mature health systems. It is almost never technical. It fails because redesign that genuinely changes how work is divided also changes the roles of the people who must implement it. Rational people slow that down and they are right to, and the institution then absorbs the technology while preserving the structure. That resistance requires an incumbent. Where a meaningful share of the establishment is unfilled, the composition of the work stays a live decision instead of becoming a negotiation with people whose jobs it changes.
You cannot redesign the role of a workforce that has not been recruited.Which is a statement about establishment plans, curricula and job descriptions, and not about anyone's employment.
Two things get conflated here, and the conflation is fatal to the argument because it is the version an opponent will attack.
This is not a case for employing fewer clinicians. The Kingdom is short of clinicians on any reading, will remain short of them, and every policy instrument pointed at that problem is pointed in the right direction. The case is that the composition of a future workforce is a design variable. What proportion of new roles will exist to move information between people, and what proportion to exercise judgment, perform procedures and care for patients? That ratio is currently set by inheritance. It could be set by design. Three bands of clinical work make it concrete.
Band A is hands on the patient. Surgery, procedures, bedside care, acquisition, physical examination. Presence is the service. No technology in prospect substitutes for it and the Kingdom needs more of it, not less.
Band B is clinical judgment that does not require continuous physical presence. Review, interpretation, escalation, second opinion, care planning. Machines augment this band and do not replace it, as the regulator has said. What changes is reach, because one specialist judgment covers more patients when the information arrives assembled instead of having to be gathered.
Band C exists only because information does not move on its own. Coordination, scheduling, chasing results, reconciliation, transcription, handover assembly, documentation performed as a separate act after the clinical one. None of this is unimportant work. It is clinically essential, almost entirely mechanical, and it eats an enormous amount of qualified attention that was trained for something else.
Health systems count doctors and nurses. What they actually ration is qualified human attention at the moment it matters. A consultant reading a result three hours late, a nurse reconciling a medication list, a registrar assembling a handover are all spending the scarce thing on work that does not require it.
The AI Hospital does not eliminate human clinical judgment. It stops spending human clinical judgment on information movement that machines can perform, so that it is available for exceptions, irreversible decisions, uncertainty, communication and physical care.
There is an obvious objection here, and a good one. Incumbency is not only people. It lives in establishment plans already approved, curricula already accredited, recruitment contracts already signed and professional scope definitions that predate all of this. Resistance can form around a post before anyone occupies it.
That narrows the window. It does not close it. An establishment plan can be revised at the cost of a planning cycle. A curriculum can be amended between intakes. A signed recruitment contract is expensive but finite. A department of four hundred people with a budget line, a director and a professional identity cannot be revised at any price a health system will pay. The asymmetry is the point, and the asymmetry is large.
04How the window actually closes
Windows of this kind do not close on a date. They close through a chain of individually reasonable decisions, and the closing is invisible while it happens.
A hospital design decision becomes a staffing assumption. A staffing assumption becomes a workflow. A workflow becomes a technology requirement. A technology requirement becomes a contract. A contract becomes an installed base. By the time anyone sees the result, the operating model was decided years earlier by people who were never asked to decide it, in a document about bed numbers. The debt is incurred on the day the doors open, at full price, and paid down over roughly two decades.
If the operating model is designed first, the physical estate can be sized around the care model that results. If the buildings are designed first, the care model spends the next twenty years justifying the building.
This is the entire argument for acting on the next commissioning cycle instead of waiting for the next technology cycle.
The clusters make this concrete, and they are what makes the argument operational instead of theoretical. Twenty health clusters now operate across the Kingdom, each a corporatized entity holding the facilities in a defined catchment and accountable for a defined population, together serving more than 20 million people.9 They are not uniform. Populations range from around 303,000 in Al-Baha to around 3.9 million in Riyadh First, which matters if anyone proposes a single template.
A cluster is a genuine unit of redesign, because it holds a population, an integrated estate and, as purchasing reform lands, a budget attached to outcomes for those people. It is also, if nobody uses it that way, the vehicle through which the conventional model is replicated in every region simultaneously.
05The case against
An argument that does not state its strongest objections is advocacy. Here are six I would raise if I were being paid to take the other side, and what I think each is worth.
Hospitals cannot be replaced by software
Correct, and nothing here proposes replacing them. Surgery, resuscitation, childbirth, procedures and acute deterioration all require a body in a room and always will. Nobody is proposing to eliminate hospitals. The proposition is to change what requires one, and therefore how many beds and clinic rooms a cluster commissions for 2035.
Saudi Arabia already has hospitals, clinicians and systems
It does, at considerable scale, and none of this argument applies to them. The distinction is between installed base and future capacity. The existing estate will be modernized like everyone else's and should be. The claim concerns facilities not yet designed and roles not yet filled, a smaller set than enthusiasts imply and a larger one than skeptics allow.
AI is not reliable enough for this
Reliability is not one property, and treating it as one is how this conversation goes wrong in both directions. Information movement is solved. Bounded operational decisions are largely solved. Reversible clinical decisions with a human reviewer are a governance question more than a capability one. Irreversible and high-consequence acts are not close, and I would not design as though they were. The objection is decisive against uniform autonomy and has almost no force against graded autonomy, which is why the schedule matters more than the models.
Regulation will not allow it
The regulator has published its position on human oversight, adaptive algorithms, lifecycle monitoring and declared autonomous functions, and has authorized domestic products under it. What does not exist is a schedule allocating autonomy across decision classes, and the absence of a rule is not a prohibition. Treat it as an invitation to propose one and build the evidence behind it. The risk is not that regulation blocks this. The risk is that someone deploys without proposing anything and sets the agenda back a decade.
Patients want human doctors
They do, and they are right to. Patients want trust, clinical judgment, communication and human care. It does not follow that a human must personally perform every information-processing task required to deliver them. Sixteen million virtual appointments and consultations in a year suggests acceptance across a considerable range of care is further along than assumed. The objection holds for acute presentations and weakens for chronic management, which is one reason chronic cohorts belong first in any roadmap.
An expatriate workforce has no stake in a twenty year reform
This is the strongest objection in the list and I do not have a clean answer. A clinician on a three year contract has little reason to invest in an operating model whose benefits accrue after they leave. The partial answer is that the composition decision precedes recruitment, so the model is set before the workforce that has to sustain it arrives. The residual risk is real, it argues for putting Saudi clinical leadership in charge of the redesign instead of a consultancy or a vendor, and it is what I would watch most closely if I were accountable for delivery.
One more, which is less an objection than the practical way these programs die. Interoperability may not be ready in time. No argument answers that. The only answer is to start in one cluster and refuse to make national interoperability a precondition, which is the default failure mode of every large health IT program on record.
06Three decisions, before the next commissioning cycle
An argument is worth something only if it changes a decision. Three follow from this one, and all three can be taken before anybody agrees on what an AI Hospital is.
Find out which design assumptions are still changeable
List the facilities due for commissioning in the next thirty-six months and establish, for each, whether the bed count and clinic footprint can still be amended. Those numbers encode an operating model whether or not anyone said so. If nobody can name the assumption that produced them, the assumption is the conventional one and it has already been made on your behalf.
State the intended composition of the future workforce, not its size
For roles planned but not yet recruited, how many exist because information does not move on its own? Answerable in a planning cycle, unanswerable once the posts are filled. Nothing here requires reducing headcount. It requires deciding what the headcount will be doing.
Name one cluster as the unit of redesign
Not a national program and not a departmental pilot. A cluster holds a defined population, an integrated estate and an accountable leadership, which makes it the smallest unit at which an operating model can be observed. Run it alongside conventional clusters on identical measures, so the fallback is the status quo continuing untouched and the political cost of failure stays bounded.
Saudi Arabia does not need to catch up with anyone on health system digitization. It has virtual care operating at national scale, a regulator that has published its position on AI and authorized domestic products under it, a national AI company with healthcare among its four named sectors, purchasing separated from provision, and capital already committed.
What it additionally has is a health system still partly being designed, to be staffed by people not yet recruited, paid for by a mechanism still being built. That is the only circumstance in which a care model can be designed around machine capacity instead of having machine capacity bolted on afterward.
It is also the most perishable asset in the entire program. Nobody will decide to spend it. It goes quietly, one commissioned building and one approved establishment plan at a time.
What this piece does not answer. It has argued that the operating model is the decision and that the decision is still open. It has not said what the operating model actually is. The phrase AI Hospital has been used here as a placeholder for something that needs defining in operational terms, because a term that cannot be specified cannot be commissioned.
No. 03 does that work. It sets the conventional sequence beside the exception-first one, walks the five systems that make the second buildable, states which economic variables move and which are only claimed to move, and separates the part of the Kingdom's AI capability that already exists from the part that has no owner.
Sources and method
- McKinsey and Company, The State of AI, 2026 global survey, and Gartner, Hype Cycle for Agentic AI, April 2026. Cited in the preceding article in this series for the gap between reported AI adoption and enterprise level results, and between agent deployment and agent intent. mckinsey.com
- Ministry of Finance, Kingdom of Saudi Arabia. Budget Statement FY2026, announced 2 December 2025. Total expenditure SAR 1,313 billion (about USD 350 billion), health and social development SAR 259 billion (about USD 69 billion), the largest single sector allocation. Riyal conversions throughout use the fixed peg of SAR 3.75 to the US dollar. The sector definition covers healthcare provision alongside social security, welfare and related services, and is not a health-only figure. The allocation is lower than the 2025 estimate. Budget statement · announcement
- Saudi Press Agency, reporting the National Center for Health Workforce Planning under the Saudi Commission for Health Specialties, 23 April 2026. Approximately 800,000 registered practitioners, approximately 460,000 Saudi nationals. Saudi share of physicians 31 percent in 2018 rising to 40 percent in 2026, nursing and midwifery 30 percent to 38 percent. Registered practitioners across all health professions, a wider denominator than the GASTAT count. SPA
- General Authority for Statistics. Healthcare Establishments and Workforce Statistics Publication 2024, reference year 2024. Physicians 129,772, 36.8 per 10,000 population. Nurses 243,336. Saudi share of physicians 42.5 percent, of nurses 43.6 percent, of dentists 58.2 percent, of pharmacists 47.5 percent. GASTAT
- Saudi Press Agency, 2 June 2022. Cabinet approval establishing the Health Holding Company and regulating the Center for National Health Insurance, with the Ministry of Health moving toward a regulatory and supervisory role. SPA · Ministry of Health, health financing
- Council of Health Insurance. AR-DRG program page, describing Australian Refined Diagnosis Related Groups as the selected case-mix classification in support of value-based healthcare. CHI strategy 2025 to 2027, launched 20 July 2025, five pillars and 47 initiatives, including value-based payment models. The Council of Health Insurance describes itself as a regulatory body. It is distinct from the Center for National Health Insurance, which purchases care for citizens. This is not a single payer system, and no binding national timetable for DRG implementation has been published. AR-DRG · strategy
- Rising obesity and shifting disease patterns in Saudi Arabia, a nine year population-based analysis of chronic disease burden and multimorbidity. BMC Public Health. Study period January 2017 to April 2025, 74,881 adult patients, 956,547 visit entries. Obesity across all classes 46.4 percent of the study population. Type 2 diabetes 34 percent among those of normal body mass index rising to 56 percent in obesity class three. Figures describe adults attending care, not national prevalence. Springer Nature Link
- Saudi Vision 2030, Health Sector Transformation Program, and the Health Sector Transformation Report 2024. Program scope covering access, quality and efficiency, prevention and traffic safety, with twenty clusters established and three transitioning in the second half of 2024. Vision 2030 · 2024 report
- Health Holding Company. Twenty health clusters serving all regions, together providing services to over 20 million people. Cluster populations range from approximately 303,000 in Al-Baha to approximately 3.9 million in Riyadh First. About · cluster register
- Ministry of Health. Seha Virtual Hospital performance for calendar year 2025, published 27 January 2026. More than 16 million virtual appointments and medical consultations. More than 11.5 million completed virtual clinic appointments, growth of 56 percent. More than 220,000 clinical cases managed, growth of 49 percent. Ministry of Health · SPA
- Ministry of Health, Seha Virtual Hospital project page. More than 242 connected hospitals, 48 main specialties, 68 subspecialties, annual capacity exceeding 597,000 beneficiaries, seven service pathways. The page carries no measurement date for these figures. Page last updated April 2026. Ministry of Health
- Saudi Press Agency, 21 October 2024. Seha Virtual Hospital recognized by Guinness World Records as the largest virtual healthcare provider in the world. The record title is largest virtual healthcare provider, not largest virtual hospital. The award is attested by Saudi official sources. A corresponding record page could not be located on the Guinness World Records website at the time of writing. SPA
- Saudi Food and Drug Authority, 10 September 2026. Marketing authorization granted to two Saudi-developed AI-enabled medical software products, Dental IQ for analysis of dental radiographs and SAARIA for retinal image analysis in diabetic retinopathy, both through the Innovative Medical Devices Pathway. The authority states that such outputs are intended to support clinical decision-making and do not replace the healthcare professional's clinical judgment, with healthcare professionals responsible for reviewing and validating outputs. SFDA · Arab News
- Colliers International assessment of the Middle East healthcare landscape, reported June 2023. An estimated 175,000 additional healthcare professionals required by 2030, comprising 69,000 doctors, 64,000 nurses and 42,000 allied health professionals, against a stated base of roughly 232,000. Consultancy modeling from 2023, not an official Saudi figure. The stated base is inconsistent with the official counts in sources 2 and 3. Used here as an order-of-magnitude planning signal only. No Saudi government body has published an equivalent gap figure. report summary
What is sourced. Figures, dates, regulatory positions and organizational facts are cited above, taken from primary Saudi government sources wherever one exists. Where the only available source is consultancy modeling, trade press or a secondary report, the entry says so.
What is the author's analysis. The AI Hospital operating model, the three-band framing of clinical work, human attention as the scarce resource, the design window and the chain that hardens it, the five systems, the graded autonomy schedule and the sovereignty stack are the author's own framework. None is a published policy, a regulatory classification or an official plan, and none should be represented as one.
What is first-hand. The account of a detection product line in live radiology workflow for breast cancer, intracranial hemorrhage, chest pathology and fractures, and the current work on patient-level domain-specific models, is the author's own experience. No employer, client or organization is named in connection with it.
What is not a claim. Nothing here describes the internal plans, intentions or roadmaps of any organization, including the Ministry of Health, the Health Holding Company, the Saudi Food and Drug Authority, HUMAIN or any of their partners. Statements about those organizations are confined to what they have published.
About the author
Prashant Akhawat has spent more than two decades building technology and AI inside regulated industries. He has built an AI capability from a blank sheet twice, and both times it reached production.
He is Chief Technology and AI Officer at Ninestars Information Technologies, where he conceived and architected the AOTM platform and took it from ideation to live. Earlier, as chief operating officer inside a healthcare group, he carried technology accountability across care delivery, teleradiology and clinical software, and shipped a clinical AI product line into live emergency reporting workflow. His research on AI detection of intracranial hemorrhage has been presented at RSNA and ESER, and his present work is on per-patient domain-specific models for personalized clinical systems.
He developed AI-SAFE, an enterprise AI framework, and publishes the CXO Intelligence Series at akhawat.com.
The AI Hospital
Numbers 02 to 05 of The Decision Layer, published under the CXO Intelligence Series, read as one sequence. No. 02 makes the argument, No. 03 defines the operating model, No. 04 sets out governance and accountability, No. 05 specifies the architecture. No. 01 opened the series on a different subject and is not required reading for any of them.
Next. No. 03, The Building Becomes the Exception, on what an AI Hospital actually is and what it changes.
Suggested citation: Akhawat, P. (2026). The Hospital Saudi Arabia Has Not Built Yet. CXO Intelligence Series, Edition 10.