Together AI will run its open-source models on 250 megawatts of Saudi electricity and 120,000 AI chips from Humain, a state-backed artificial intelligence company in the kingdom. The deal, announced Monday at the LEAP tech conference, is expected to generate more than $5 billion in gross annualized revenue in its first year, with Humain receiving a share of that revenue.
The arrangement gives Together AI, valued at $8.3 billion in July, a major new base for serving customers who want to deploy open-source AI models at scale. It also marks one of the largest cross-border infrastructure partnerships in the AI sector this year.
Why this deal, and why now
The timing is hard to ignore. In the first three months of 2026 alone, local opposition in the United States blocked or delayed roughly $130 billion worth of AI data center projects, according to tracking by Data Center Watch. More than 500 counties and municipalities across the U.S. now have bans, moratoria, or severe restrictions on new hyperscale facilities. A Gallup survey published in May found that 71 percent of Americans oppose building AI data centers near their homes, with 48 percent strongly opposed.
Against that backdrop, a partnership that moves 250 megawatts of compute capacity outside the U.S. starts to look less like a growth story and more like an escape route. Together AI said it expects the Saudi data center to bring in $5 billion annually. That number is a forward-looking company estimate, not audited revenue. It signals ambition more than proof that the facility is already serving workloads at full scale.
Who benefits most from this framing? The announcement casts Together AI as a company solving a supply problem by going where the power and chips are available. It also casts Humain as a gateway to the Middle East, Europe, and Africa, with infrastructure ready to serve training and inference workloads across those regions. For Saudi Arabia, the deal fits a broader push to become a hub for AI compute, backed by the Public Investment Fund and years of planned data center construction through 2034.
And yet, the story could easily run the other way. What if this is less about open access to models and more about concentrating critical AI infrastructure in jurisdictions with fewer local veto points? Open-source AI depends on broad, distributed access to compute. If the cheapest, most available megawatts live in a small set of state-aligned data centers, does that expand the ecosystem or narrow it?
Open-source AI, foreign soil
Together AI’s core business is serving open-source AI models to developers and enterprises. The company has built its reputation on making those models easier to deploy and run at scale. The Humain partnership significantly expands its capacity to do that, at least on paper.
But the geography matters. Running open-source models on Saudi infrastructure means that a growing share of the world’s access to those models will depend on power, cooling, and political stability in one part of the world. That is not inherently bad. It is just a different risk profile than a more distributed buildout across many regions.
The revenue-sharing piece adds another layer. Humain will receive a cut of the income generated from the data center. That aligns incentives between the two companies. It also means that a foreign, state-backed entity has a direct financial stake in the success of Together AI’s open-source model business. For investors and customers, that may be a feature. For policymakers watching where critical AI infrastructure sits, it may be a question mark.
There is also the matter of visibility. Data centers in the U.S. face increasing demands for transparency on energy use, water consumption, and local impacts. Some states now require developers to secure local approval and commit to binding standards before permits move forward. In Saudi Arabia, the rules are different. The partnership brings compute capacity online faster, but it also moves some of the hardest questions about environmental and social costs further from the communities that will feel the effects of AI deployment.
None of this means the deal is dangerous. It means the story is not just about capacity and revenue. It is about where the leverage sits when the next constraint hits, whether that is power, chips, or political pressure.
Together AI and Humain say the partnership will offer developers and enterprises another option for high-performance AI infrastructure. That is true as far as it goes. The unanswered part is what happens when that option becomes one of the few that can scale quickly.
By the end of the announcement, the picture is clear: a U.S. startup, facing mounting resistance at home, has secured a large slice of foreign compute to keep its open-source models growing. The optimism is in the numbers and the speed. The caution lives in the map.