Technology

At LEAP 2026: Saudi Arabia’s AI Spending Meets the ROI Problem

Bakhtawar Majid

By: Bakhtawar Majid

7 min read

Billions are flowing into AI, infrastructure and startups in the Kingdom, but companies at LEAP 2026 are increasingly focused on the cost of getting those investments into production and making them work. 

[For more news, click here]

Day three of LEAP 2026, held in Riyadh on Wednesday 2 September, was billed as venture funding, regional headquarters and cybersecurity. It delivered all three, plus something nobody scheduled, a sustained argument conducted almost entirely by vendors about how little of the money spent on AI has come back. 

Saudi Arabia captured 45 percent of MENA venture capital in 2025, up from 32 percent, funding up twenty-seven-fold since 2018. The Kingdom led the region for a third consecutive year and now hosts nine unicorns, half of MENA’s total. Globally, AI companies absorbed $258 billion in 2025, or 61 percent of all venture capital. 

Venture activity was equally busy, with COFE Tech leading the rounds at $43 million, followed by FlyAkeed at $25.15 million and Nayla Finance at $18 million. Impact46 appeared four times, with $12 million in automotive, $8 million into Hazen.ai, $6 million into ZenHR and $1.5 million into Taka House. A fintech raised $8.5 million from angels, while a $4 million AI-and-SME-finance round marked Speedinvest’s first regional check. Ula Capital and South Korea’s FuturePlay announced a $100 million fund, and Impact46 with Hala grew SME financing from $26.7 million to $66.7 million. In total, the announcements amounted to $166.7 million in funds and $126.15 million in startups. 

Saudi Arabia builds the infrastructure around AI 

Alfanar Engineering Services announced a $150 million factory in Riyadh, SAR 550 million, building data center power modules with capacity to serve 2GW. Its pitch was ownership rather than assembly, R&D in Riyadh, Saudi engineers, IP held locally. 

CTS Group and Desert Technologies formed CTSDT, with two factories operating by year end, making e-pods, switchboards, fabricated skids and prefabricated mechanical modules. Phase one covers roughly 300MW of components annually for about SAR 170 million, with a plan to reach 1GW and exports into the Middle East, Africa and Turkey. 

HP’s Agent One, an agentic IT service tool built entirely by Saudi talent, carries Arabic capability and on-premises deployment, alongside FlowMaster for document automation and Najdi dialect research. Thousands of Saudi-made AI PCs have shipped with SAMI Advanced Electronics, and HP Garage 2.0, a startup collaborator, was announced alongside them. 

Snowflake’s Michel Nader, regional vice president and general manager for the Middle East, Turkey and Africa, committed $300 million through 2030 and two cloud regions by year end, alongside a national training programme run with MCIT. The company’s position is no AI strategy without a data strategy. Logarithm and Exabeam are moving their regional headquarters to Saudi Arabia, with a Cyber Security Center of Excellence, more than 35 staff already in the Kingdom, and a Google Cloud partnership. stc launched colab, its venture-building arm. 

HUMAIN Chat Enterprise, presented by deputy CEO and president Dr Yasser Al-Onaizan, launched mid-July and is fully sovereign under PDPL. It has passed 600 users in six weeks, with 30-plus organizations in live deployment, and runs a multi-model catalog with Arabic-native options, page-level citations, deck building and connectors into Slack, Teams and Notion. A new AI marketplace targets the failure everyone else spent the day describing, where “a lot of POCs and pilots never reach production”. The focus on what is being built and where it runs also fed into a wider discussion at LEAP about what sovereign AI actually requires. 

The cost of getting AI into production 

Most talk of sovereign AI fixes on GPUs and gigawatts secured, “that’s just the headline, that’s just the beginning of our conversation.” 

“From a hardware perspective, sovereign AI is nothing more than platform where NVIDIA and a non-NVIDIA coexist or United States product and the non-United States product coexist, the diversified supply chain.” stated Sunghyun Park, Co-founder and Chief Executive Officer, Rebellions 

At production scale with SK Telecom and KT, the constraint turned out to be cost rather than performance. Once running at scale, the cost compounds very fast, and total cost of ownership becomes the metric. The alternative to building alone is to become an ecosystem player. On chips or watts first, the watts come first, because securing power is what takes the time. 

Every data center and chip also needs copper, aluminum and critical minerals, which brought Ma’aden to LEAP for the first time. AI is not the whole story, cloud storage, scalable compute, sensors and low-latency communications in remote sites all came first. Exploration timelines have fallen from five years to two in some cases, the “zero entry mine” moves 20 percent more material. Its first autonomous truck deployment in 2013 took eighteen months to run, a complete system now takes fourteen weeks, at a fifth of the cost. 

The return on that investment was harder to pin down. Research puts 95 percent of AI applications below meaningful ROI, with deployment into support functions rather than core operations the problem. Most deployments are assistants supporting a human rather than agents owning an outcome. A task is too small to show impact and an enterprise too big to rebuild, which leaves the department as the unit that works. Intelmatix restructured a city mobility center around that principle, mixing a smaller human team with agents. 

“Rather than starting to look at which agents to build, I would just challenge and say actually look at which department to restructure around AI.” said Dr Anas Alfaris, Co-founder and Chief Executive Officer, Intelmatix 

Most enterprise AI, in Lucidya’s comparison, is a Ferrari driven to the grocery store. On the cloud panel, the diagnosis from Zoho and Databricks was enterprises fixing AI before their data, in a market moving from token maxing to value maxing. At Savola, working with Oracle, invoice processing fell from five days to one, the advice to business leaders was not to delegate AI to IT. 

“If you’re still using human speed, cybersecurity to defend against what’s happening today, what you’re doing is you’re bringing your knife to a drone fight.” said Wael Ahmed Fattouh, Chief Advisory Officer, Saudi Information Technology Company (SITE) 

Volume, speed and sophistication are all rising while the cost of mounting an attack has fallen. Ransom profitability has fallen too, so the target is data. Aramco already runs agents that have improved investigative response by around 85 percent, which opens a governance gap underneath. 

“For a human, an employee has a role. They’re given a privilege and access according to the duration in which they have that role. For an agent, that looks different.” said Majed Al-Salamah, Head of Cybersecurity Strategy and Oversight Division, Aramco 

Sovereignty, on that stage, was framed as the opposite of restriction. 

“For me, the word that comes to me when I think of sovereignty is freedom. It’s freedom to innovate. It’s freedom to create. It’s freedom to collaborate, because that’s what sovereignty gives you. Sovereignty is actually the removal of barriers.” said Wael Ahmed Fattouh, Chief Advisory Officer, Saudi Information Technology Company (SITE) 

Skills remain another unresolved problem. The sector is not ready, on TETCO’s blunt assessment, and the trap Misk identified is that skills-first hiring is arriving exactly as AI automates the roles people used to be promoted from. ZNotes founder Zubair Junjunia refused the framing. 

“They don’t need the system to catch up. They will build the system themselves.” stated Zubair Junjunia, Founder, ZNotes 

The discussion extended beyond enterprise systems and infrastructure into applications already being put to work. Other sessions included WeRide on 13 countries and 60-plus cities, Archer Aviation on Gulf regulators and Huawei on the token economy. Elsewhere, a localized diabetic retinopathy tool now runs across sixteen charity centers, returning results in under a minute. It was built and validated in the Kingdom because imported models lacked the sensitivity for a different patient population. 

“I kept asking myself, why does a VIP get good quality care where a janitor who’s just as important cannot?” said Prof Selwa Al-Hazzaa, Chairman of the Board and Founder, SDM (Smart Diagnostic Medicine) 

By the end of the third day, the investment was visible across venture funding, factories, cloud regions, enterprise systems and cybersecurity. The harder part is what follows those announcements, getting pilots into production, controlling the cost of running them, protecting the data they handle and having people capable of working with the systems. 


Related Articles 

Saudi Arabia opens LEAP’s fifth edition by putting a price on intelligence 

Saudi Arabia Is Building the Infrastructure Behind Its AI Ambitions at LEAP 2026 

At LEAP 2026, Commvault Puts Recovery at the Heart of Saudi Arabia’s Cloud and AI Push 

Share this article

Related Articles