orfloat

ORF-N-2026-010 · Commentary

Oman's AI program has started without its enterprises

Claim

Oman's national AI push is funded, legislated, and shipping: eight institutional moves in six years and OMR 79 million invested, while enterprise adoption has no published measure. With frontier capability doubling every three months since 2024 and the price of intelligence collapsing, the early mover in each Omani sector collects the spread the state has already paid to open.

Between August 2020 and this spring, the Omani state made eight institutional moves on artificial intelligence. It created a ministry, issued an executive program, received a royal directive, ratified a national program through the Cabinet, bound its own agencies to an AI ethics policy, launched a sovereign language model, published a five-year roadmap, and carved out an economic zone for AI by royal decree. Eight moves in six years, each one public, dated, and on the record.

Over the same six years, as far as any published record shows, most operating businesses in Oman made none.

This note is the policy half of an argument we have been making since May. The capability overhang is the distance between what a frontier model can already do and what an operating business actually does with it, and it widens on a clock. What the state’s record adds is this: in Oman, waiting is no longer a neutral act with the government as a bystander. The government has moved, in public, with money. The half of the curve still missing is private, and it is the half that pays.

Eight moves in six years

The spine of the record is short enough to print whole.

Six years of state-side execution on AI, gaps drawn to scaleaug 2020oct 2022nov 2023sep 2024apr 2025sep 2025mar 2026apr 202626 months13 months10 months7 months5 months6 monthsroyal decree 90/2020 creates mtcit: transport,communications and it under one ministry,carrying the space and advanced-technology mandateexecutive program for ai and advanceddigital technologies issuedroyal directive: prepare a national programfor the implementation and localizationof ai technologiescabinet approves the national program for aiand advanced digital technologies (2024-2026)general policy for the safe and ethical useof ai systems: binding on government units,reaching private sectors via their regulatorsma’een, the national llm, launches with theai studio and the national open data portaldigital economy harvest report: omr 79m in aito date; the 2026-2030 package announced,with digital centres in all 11 governoratesroyal decree 50/2026 establishes an ai specialeconomic zone in muscat
aug 2020   royal decree 90/2020 creates mtcit: transport,
           communications and it under one ministry, carrying
           the space and advanced-technology mandate
oct 2022   executive program for ai and advanced digital
           technologies issued
nov 2023   royal directive: prepare a national program for the
           implementation and localization of ai technologies
sep 2024   cabinet approves the national program for ai and
           advanced digital technologies (2024-2026)
apr 2025   general policy for the safe and ethical use of ai
           systems issued: binding on government units, reaching
           regulated private sectors through their regulators
sep 2025   ma'een, the national large language model, launches
           alongside the ai studio and the national open data
           portal at comex
mar 2026   digital economy harvest report published: omr 79m
           invested in ai to date; the 2026-2030 package is
           announced, with digital transformation centres in
           all 11 governorates and a sovereign cloud
apr 2026   royal decree 50/2026 establishes an ai special
           economic zone in muscat
the gaps between moves, in order: 26, 13, 10, 7, 5, 6, and 1 month.
Figure 1. six years of state-side execution on AI, from the ministry’s creation to a dedicated economic zone, with the vertical gaps drawn to scale. every node is a public, dated instrument or announcement; sources are in the references. read the spacing as well as the labels: the cadence compresses from 26 months between the first two moves to a single month between the last two. the question the rest of this note asks is what moved on the private side over the same six years.

The November 2023 directive is worth quoting, because it names the intent: His Majesty directed the preparation of “a national program for the implementation and localization of artificial intelligence technologies”. The Cabinet approved that program on 19 September 2024. It stands on three pillars: adopting AI in the economic and development sectors, localizing AI technologies, and governing AI applications with a human-centred vision. The April 2025 General Policy puts real obligations under the third pillar: human oversight on consequential decisions, explainability in applications touching employment, health, and financial services, bias testing of datasets and algorithms, and marked generative content. It binds government units today and reaches regulated private firms through their sector regulators.

The money is real

The program is not a press release with no budget line. Cumulative investment in AI initiatives stood at OMR 55 million for 2021 through 2023, and by March 2026 it had passed OMR 79 million, covering the Ma’een model (also transliterated Mu’een), the National Open Data Portal, the AI Studio, and a dedicated AI startup zone, with a stated ambition to grow AI investment by roughly 20% a year. The digital economy it sits inside contributed about OMR 800 million to GDP in 2023, around the 2% baseline the program started from. The legislated targets: 3% of GDP by 2025, 5% by 2030, and 10% by 2040.

Digital economy share of Oman’s GDP: the legislated ramp202120252030204005%10%2% baselineomr 800m (2023)3% target5% target10% by 2040
digital economy share of oman's gdp (source: mtcit / ndep)
  2021   2%    actual baseline
  2023   ~2%   omr 800m, core and direct (published mar 2026)
  2025   3%    target
  2030   5%    target
  2040   10%   target
the solid points are published actuals; the dashed leg is the legislated ramp.
Figure 2. the ramp the state has signed its name to: from a 2% baseline to 10% of GDP by 2040. the solid points are published figures; the dashed leg is target. a five-fold expansion of the digital economy’s share is the stated destination, and AI is named as a core mechanism for getting there. the 2025 point stays hollow because the march 2026 harvest report publishes no 2025 share figure; its latest published absolute is the 2023 omr 800m. sources: MTCIT, national digital economy programme and the march 2026 digital economy harvest report.

The state also lets itself be graded in public. The National Program’s stated objective is a place among the top 50 countries in Oxford Insights’ Government AI Readiness Index. The trajectory so far: 50th in 2023, 45th in 2024, then 64th of 195 in the 2025 edition, which rebuilt its methodology that year, so the drop is not a like-for-like fall. On the old yardstick the top-50 objective was already met; on the new one it is open again. The instructive part is the 2025 pillar detail: Oman scores 88.5 on policy capacity and 71 on public-sector adoption, against 43 on AI infrastructure and 39.9 on development and diffusion. Read plainly: the government’s intent and its own adoption grade well. What drags the country’s readiness is the ecosystem around it, and the ecosystem is where private enterprise lives.

What an operator can actually collect

The push is not rhetoric aimed at the private sector; there is a counter you can walk up to. Five mechanisms, today:

  1. A free smart-manufacturing assessment. MOCIIP, with the Gulf Organization for Industrial Consulting, runs Smart Industry Readiness Index assessments for factories: a 16-dimension audit by certified evaluators, with 20 factories selected in the pilot round and 5 funded through full digital transformation. The stated national goal is 30% of factories running smart by 2026.
  2. Future Fund Oman. USD 5.2 billion over five years across priority sectors including ICT. The SME and venture window alone has approved 132 projects with USD 56.7 million committed, from pre-seed to SME debt.
  3. Riyada’s grants for high-tech SMEs. The SME Development Authority runs a grants system for innovative and technology projects, with 67 supported so far and over OMR 38 million deployed across its financing programs since 2022, per the IMF’s January 2025 staff report.
  4. The AI Special Economic Zone. Royal Decree 50/2026 establishes a dedicated AI zone in Muscat under OPAZ, with the incentive and exemption toolkit of a special economic zone pointed at AI and technology firms.
  5. A subsidised talent pipeline and free data. Makeen has trained more than 11,000 Omanis in digital and AI skills at no cost to the trainee, with placement tracks into private institutions; technical IT roles are now 69% Omanised, by the March 2026 harvest report’s count. The National Open Data Portal is live and free, a clean input for any firm that wants to build.

None of these will wire a model into your reservations system, your supplier thread, or your books. What they do is lower the price of starting. The state has, in effect, subsidised the first step and published the dataset. That is what a government push looks like from the inside of a business: not a mandate, but a discount.

Meanwhile, the frontier did not wait

While those eight moves landed, the capability of the systems the program exists to adopt kept compounding. METR, whose time-horizon work we have written about before, measures the length of task a model can complete reliably. In 2020 the best model managed software tasks of about eight seconds. By May 2024, seven minutes. By February 2026, around twelve hours. The newest point METR has measured, an April preview of the model that shipped this week as Claude Fable 5, lands near seventeen hours, far enough up the scale that METR flags it as past the benchmark’s own sixteen-hour reliability ceiling. The doubling time across the whole series is about six months; measured from 2024 onward it is about three.

METR 50% time horizon by model release date, log scale2020202120222023202420252026model release date1 min10 min1 h12 hdavinci-002, ~8 secgpt-4o, 7 minopus 4.6, ~12 hmythos preview, ~17 hdoubling every ~3 monthssince 2024
metr 50%-success time horizon, by model release (log scale; source: metr th1.1)
  may 2020   davinci-002        0.14 min (~8 sec)
  may 2024   gpt-4o             7 min
  oct 2024   claude 3.5 sonnet  20.5 min
  feb 2025   claude 3.7 sonnet  60 min
  apr 2025   o3                 120 min
  aug 2025   gpt-5              203 min
  nov 2025   claude opus 4.5    293 min (~4.9 h)
  feb 2026   claude opus 4.6    719 min (~12 h)
  apr 2026   claude mythos preview (early fable 5)  1,045 min (~17.4 h)
doubling time (metr's published fit): ~188 days across the whole series,
~89 days measured from 2024.
the apr 2026 point sits above metr's stated 16 h reliability ceiling; opus 4.8
and the released fable 5 have no published metr measurement yet.
Figure 3. the length of software task a frontier model completes at 50% reliability, by release date, on a log scale. eight seconds in 2020; twelve hours in early 2026; near seventeen hours for the april preview of the model that shipped as claude fable 5 on 09 jun 2026, a point metr itself flags as past the benchmark’s 16-hour reliability ceiling. opus 4.8 and the released fable 5 have no published metr measurement yet, so the curve ends at the last point metr has actually measured. the doubling time itself has shortened from about six months to about three. this is the curve an adoption program is racing. source: METR, time horizon 1.1, retrieved 11 jun 2026.

The price of intelligence collapsed

The other half of the frontier story is what it costs. Stanford’s AI Index puts a number on it: querying a model at GPT-3.5-class performance fell from USD 20 per million tokens in November 2022 to USD 0.07 per million tokens by October 2024, a 280-fold fall in under two years. Epoch AI, measuring across benchmarks, finds the price of holding capability constant falls between 9-fold and 900-fold per year, with the median around 50-fold. Headline prices for the newest flagship models wobble upward with each release; the price of any fixed level of capability only falls.

Price of GPT-3.5-class intelligence, USD per million tokens, log scale2022202320242025$100$10$1$0.10$20, nov 2022$0.07, oct 2024280x cheaper in 23 months
price of gpt-3.5-class intelligence, usd per million tokens (log scale)
  nov 2022   $20.00   (gpt-3.5 at launch)
  oct 2024   $0.07    (equivalent benchmark performance)
a 280-fold fall in under two years (stanford ai index 2025). epoch ai puts the
median decline for constant capability at about 50x per year across benchmarks,
and faster since january 2024.
Figure 4. what a fixed level of intelligence costs, on a log dollar scale: the two points are the stanford AI index’s published endpoints. every workload that was too expensive to automate in 2023 should be re-priced, because the price it failed at no longer exists. sources: stanford HAI AI index 2025; epoch AI, llm inference price trends.

For a business this collapses the oldest objection. The workload you costed out in 2023 and shelved was priced against a number that has since fallen by orders of magnitude, while the model behind it went from a seven-minute attention span to a twelve-hour one.

The curve nobody measured

Here is the asymmetry that motivates this note. The state side of Oman’s adoption story is measured to the decimal: investment totals, GDP shares, index ranks, trained graduates, all published and dated. The enterprise side has no number at all. There is, today, no published figure for how many Omani businesses use AI in their operations. The first instrument that will produce one, the National Centre for Statistics and Information’s Digital Economy Survey, is in the field now and reports by mid-August 2026.

The global baseline makes the silence loud. McKinsey’s latest State of AI survey finds 88% of organizations worldwide now use AI in at least one function, up from 78% a year earlier. Whatever the Omani enterprise number turns out to be when it is finally measured, every operator we have sat with suggests it is nowhere near that, and the measurement gap itself is the tell: adoption has an owner in government and no owner inside most firms.

The state’s half of the curve is published, funded, and dated. The frontier’s half now doubles every three months. The only undrawn line in the picture is the one through your operation.

The early-mover arithmetic

Put the pieces side by side and the arithmetic for an Omani operator is unusually clean.

Your sector is named. The National Program’s target list reads like the registry of the local economy: logistics and transport, manufacturing, finance and banking, energy and renewables, culture and tourism, mining, fisheries and agriculture, alongside health, education, and government services. The field is thin. The whole country counts roughly 22 specialised AI companies, by the harvest report’s own census; the queue for being the reference implementation in your sector is short, and reference implementations are what regulators, banks, and ministries point to when they decide what good looks like. The compliance bar is visible early. The April 2025 policy tells you today what regulated AI work will be measured against tomorrow: human oversight, explainability, bias testing, marked generative output. Building to that bar from day one is cheap; retrofitting it is not. And the inputs are subsidised: the assessment is free, the data portal is open, the talent is trained at state expense, and the capital windows are funded.

Against that stands the clock from Figures 3 and 4. A capability that doubles every three months does not wait for a budgeting cycle, and a competitor who starts a year earlier is not a year ahead; they are four doublings ahead on a curve whose price floor keeps dropping. We made the general case in May, that the overhang is a planning question. The Omani case is sharper: the question has a sponsor now, and the sponsor has already paid.

What to do with this

Treat the next 90 days as the window it is. Pick one workload that touches money or hours: the supplier thread, the reconciliation, the booking flow. Wire a frontier model into it properly, with the oversight and explainability the 2025 policy will eventually expect of everyone. Measure the before and after. Then walk to the counters above with evidence instead of a deck.

That wiring is the whole of what we do. Orfloat is a small lab in Muscat that embeds and deeply integrates AI into a handful of operations at a time, and the national program just made every part of that work cheaper, better documented, and easier to justify to a board. If your sector is on the list above and your operation is still off the curve, start a conversation with us about a Discovery Phase.

References

  1. MTCIT. National Program for Artificial Intelligence and Advanced Digital Technologies, public version. September 2024. mtcit.gov.om (pdf)
  2. MTCIT. Launch of the National Program for AI and Advanced Digital Technologies. 22 September 2024. mtcit.gov.om
  3. MTCIT. Digital Economy Harvest in Oman 2021-2025: From Foundation to Enablement. March 2026. mtcit.gov.om
  4. Muscat Daily. MTCIT unveils new drive to expand digital economy. 10 March 2026. muscatdaily.com
  5. MTCIT. General Policy for the Safe and Ethical Use of Artificial Intelligence Systems. 1 April 2025. mtcit.gov.om
  6. Sultanate of Oman. Royal Decree 90/2020. August 2020. decree.om
  7. Oman Observer. Royal decree establishes AI economic zone in Muscat. April 2026. omanobserver.om
  8. Oman Observer. First large language model Mu’een launched. 8 September 2025. omanobserver.om
  9. Oxford Insights. Government AI Readiness Index. 2023, 2024, and 2025 editions. 2023 (pdf), 2024 (pdf), 2025 (pdf)
  10. METR. Time horizon 1.1. 29 January 2026, continuously updated; data retrieved 11 June 2026. metr.org
  11. Stanford HAI. AI Index Report 2025. hai.stanford.edu
  12. Epoch AI. LLM inference price trends. 12 March 2025. epoch.ai
  13. McKinsey. The state of AI. November 2025. mckinsey.com
  14. Oman Observer. Oman expands smart factories initiative. 2025. omanobserver.om
  15. Oman Investment Authority. Future Fund Oman. oia.gov.om
  16. IMF. Oman: Staff Report for the 2024 Article IV Consultation. January 2025. elibrary.imf.org
  17. Sultanate of Oman. Makeen programme. oman.om
  18. Times of Oman. NCSI to launch Digital Economy Survey 2025 in Oman. timesofoman.com