Three numbers from this week, all of them sourced, none of them in dispute.
At LEAP in Riyadh on Monday, Humain announced a joint venture with AMD and Cisco carrying an investment of up to $10 billion, to deploy up to a gigawatt of AI capacity by 2030. It announced more than $5 billion with AWS. It announced a joint venture with Applied Intuition, a company valued at $15 billion, to put driverless trucks on Saudi roads.
In the same week, the Kingdom's own budget document records education spending of SAR 204 billion actual in 2024, SAR 199 billion estimated actual in 2025 — below the SAR 201 billion approved — and SAR 202 billion allocated for 2026. Flat, and in the middle year, slightly down.
And across the opening coverage of a $15-billion week, the education content amounts to one clause: that LEAP will give students and entrepreneurs access to global innovations, international expertise, career opportunities and venture pathways.
These are not contradictory numbers
It is tempting to line those figures up and call it a scandal. It is not one, and reading it that way gets the situation wrong.
Infrastructure gets announced at events like LEAP because infrastructure arrives in units large enough to announce. A gigawatt is a headline. Ten billion dollars committed to compute is a headline. A ministry's recurring education budget is a line item that moves by single digits and produces no photograph.
Nor is a flat education budget evidence of neglect. Saudi Arabia's education spending is already among the largest line items in the national budget in absolute terms. A plateau at SAR 200 billion is not a retreat. It is a system that has finished its build-out phase and is now spending on operations.
The observation worth making is narrower, and it is about sequence. The compute is being committed now, at scale, on a 2030 horizon. The workforce that will use it is being funded at a rate that assumes the next five years look like the last five.
The one human number on the day
There was exactly one figure on LEAP's opening day that described people rather than hardware. AWS said it has trained 682,000 people across the region, against a target of two million by 2030.
That is a real number and a large one, and it deserves to be read for what it is. It is a cloud provider teaching its own stack. It is enablement — valuable, commercially motivated, and bounded by the platform it serves. A person who completes it can operate AWS. That is not the same as a workforce prepared for what a gigawatt of regional compute makes possible, and nobody at AWS has claimed otherwise.
Between a gigawatt of capacity and 682,000 people trained on one vendor's tools, there is a layer with no announcement of its own: the curricula, the assessment, the retraining pathways, and the institutions that turn national compute into national capability.
Vision 2030's Human Capability Development Program has a target in exactly that space — 70,000 citizens sent to the top 200 global educational and training institutions by 2030. It is a real commitment. It is also, set beside the week's compute numbers, a rounding error in budget terms.
The largest education intervention of the week was filed as something else
There is one exception to all of the above, and it is worth more attention than it received.
Adobe, the Ministry of Communications and Information Technology and Humain announced a partnership giving more than 27 million Saudi citizens and residents aged 13 and above twelve months of free access to Adobe Firefly Standard and Adobe Express Premium, including an image generation model tuned to Saudi culture. Rollout begins by the end of 2026.
The figure attached to it is $4 billion, and it is worth being precise about what that number is: it is the retail value of the software access being given away, not capital deployed. The minister's own framing is turning AI investment into opportunity at scale. Reporting it as $4 billion invested in anything would misstate it, and we did so in an earlier version of this piece.
The age threshold is Adobe's own, and it is the detail that matters here. A floor of 13 includes, by definition, the Kingdom's entire secondary-school population and every university student in it. That is Fikr's reading, not Adobe's claim — neither Adobe nor the Saudi Press Agency mentions students, schools or education anywhere in the announcement.
The deal does carry explicit education components, once you look past the headline: free access to Adobe Digital Academy courses, training delivered inside MCIT's Center of Digital Entrepreneurship and The Garage, and an "AI Essentials for Marketers" programme run jointly with LinkedIn.
So this is the largest single distribution of creative and AI tooling into a national student body that anyone announced this week — and it was framed, throughout, as a creative-economy and AI-adoption deal. Not as education.
We are not claiming it was mislabelled. Adobe is a creative software company and the deal is a creative software deal. But the effect on learning is real and larger than anything filed under education, and it illustrates the point better than the absence does: capability is arriving in the classroom sideways, through vendor distribution, rather than through anything built or bought for teaching.
Two things follow for anyone building here. Twelve months is a trial, not a policy — the question of what happens in month thirteen is an open one, and it will be answered by whoever has built the habit by then. And a national cohort handed professional creative tooling with no curriculum around it is a gap shaped exactly like a product.
The asterisk on the whole week
One detail belongs in any account of LEAP 2026, and it is easy to skip past. This edition did not run in February, where LEAP normally sits in the calendar. It was moved to the end of August, rescheduled amid the war with Iran.
That matters for anyone reading the numbers. A six-month delay changes who could attend, which fiscal year the commitments land in, and what the event is being compared against. When the 2027 coverage sets this year beside last year, it will be comparing a February event with an August one. We are recording the asterisk now so that we are not tempted to drop it later.
Update, 3 September
This piece was written on the opening day. On 2 September, Tuwaiq Academy — Saudi Arabia's national technology academy — held its annual ceremony on the main stage and announced more than 250 new camps and programmes, along with new global partnership. (Our sources disagree on the partnership count: SPA's Arabic headline says one, an English SPA dispatch says twelve, and we could not open either body. The camps figure is SPA's own.)
That does not overturn the argument above, but it does correct its framing. The skills layer was not absent from LEAP. It was present in a form we were not looking for: not a funded company and not a ministry line item, but a free state academy running on vendor certification. We have written that up separately in Nobody Puts Tuwaiq in the Deck, because it changes what an EdTech founder is actually competing with here.
Why nobody announced it
Because the money is not there to announce, and the reason is structural rather than attitudinal.
Global EdTech venture funding was $2.6 billion for all of 2025, with roughly 87% of deals at early stage. That is the entire world's private capital for education technology, and it is smaller than a single one of Monday's infrastructure commitments.
Regionally it is thinner still. MENA venture funding in the first half of 2026 came to $1.35 billion across 214 deals — deal count down 41%, early-stage deals down more than half. No publisher breaks education out as a sector, because the number would be too small to warrant a line.
So the picture is consistent, not contradictory. Compute is capital-intensive, announceable, and attracts sovereign-scale money. Education is labour-intensive, slow, hard to photograph, and attracts a fraction of a percent of the same pool. Both behave exactly as their economics predict.
What we take from it
We were at LEAP this week because the gap is the thesis.
If a region commits at this scale to the infrastructure layer, demand for the layer above it does not stay flat. Somebody has to build the products that teach people to use what is being installed, in the language they work in, against the curricula they actually sit. That work is not going to be funded out of a ministry's operating budget, and it is not going to arrive as a clause in a cloud partnership.
Three things follow, and they are what we look for in a founder.
Build for the compute that is coming, not the compute that exists. A learning product designed around today's inference costs is designed around a number that a gigawatt of regional capacity is about to move.
Sell into the gap, not into the budget. A flat ministry line means displacement, not expansion: you are replacing a spend, not adding one. That is a harder sale and a different pitch.
Measure what you claim. Where the capital is thin, evidence is what separates a fundable company from an interesting one. That is not a preference of ours; it is what a $2.6 billion global pool does to a market. When there is enough money for everyone, a good demo raises a round. When there is not, the companies that get funded are the ones that can show what changed for a learner, in numbers someone else could check.
What we would have wanted to see announced
Not a fund. Funds get announced at this scale and deploy at another, and the region already has capital sitting undeployed.
Something narrower would have been more useful: a procurement commitment. A ministry, or one of the entities signing gigawatt-scale infrastructure deals, saying it will buy a defined volume of Arabic-language learning capability from companies that can evidence an outcome — with the evidence standard published in advance.
That would do something no amount of venture funding does on its own. It would create a market with a known buyer, a known size and a known bar, which is the thing an early-stage EdTech founder in this region cannot currently point at. Distribution, not capital, is what kills these companies. Compute got a buyer this week. Learning did not.
We do not know whether such a commitment is being worked on, and we are not going to guess. But it is the announcement that would have changed the shape of the week, and it costs a fraction of a gigawatt.
The gigawatt will get built. What runs on it, and who is trained to run it, is still open.