The $100 Billion Bet That America Isn't Watching

On a scorching morning in August 2025, construction crews broke ground on two massive data center campuses in the Saudi Arabian desert — one in Riyadh, the other 400 kilometers east in the oil city of Dammam. Eleven buildings. Two hundred megawatts of initial capacity. Enough computing power to rival what most European nations can muster.[1] The man overseeing it all, a Jordanian-American engineer named Tareq Amin, had a line he kept repeating to anyone who'd listen: Saudi Arabia would become the world's largest exporter of AI tokens.[2]

If you've spent the last three years following the AI revolution from an American vantage point, that sentence probably sounds absurd. The AI story, as told in the United States, has four protagonists: OpenAI and its partnership with Microsoft, Anthropic and its alliance with Amazon, Google's Gemini, and Elon Musk's xAI. They are headquartered, respectively, in San Francisco, San Francisco, Mountain View, and Memphis. The talent is American. The chips are designed in Santa Clara. The narrative is domestic.

But while American tech media obsesses over the next GPT release and congressional hearings about AI safety, something enormous is happening 7,000 miles away — and almost nobody in the United States is paying attention. Saudi Arabia has launched a company called HUMAIN, backed it with more than $100 billion in committed capital and infrastructure spending, signed partnerships with nearly every major American chip and cloud company, and begun building what could become the third-largest AI infrastructure on Earth, behind only the United States and China.[3] And the argument for why this matters isn't just about geopolitics or petrodollars. It's about something more fundamental: the democratization of artificial intelligence itself.

HUMAIN didn't arrive quietly. On May 12, 2025, Crown Prince Mohammed bin Salman personally announced its creation — one day before President Donald Trump touched down in Riyadh for a state visit that would produce $600 billion in Saudi investment commitments.[4] The timing was not coincidental. HUMAIN was the centerpiece of a new US-Saudi technology relationship, a sovereign AI company wholly owned by the Public Investment Fund, the kingdom's nearly $940 billion sovereign wealth fund.[5] MBS didn't just chair PIF — he chaired HUMAIN's board directly, signaling that artificial intelligence had joined oil, defense, and foreign policy as a matter overseen personally by the crown prince.

The launch was orchestrated as a spectacle of tech diplomacy. Sam Altman, Jensen Huang, Elon Musk, and Mark Zuckerberg all attended the Riyadh investment forum. Within 48 hours, HUMAIN signed $23 billion in strategic technology partnerships — with NVIDIA, AMD, Amazon Web Services, and Qualcomm.[6] NVIDIA committed several hundred thousand of its most advanced GB300 Grace Blackwell chips over five years. AMD inked a $10 billion deal to deploy 500 megawatts of AI compute.[7] AWS pledged more than $5 billion for a first-of-its-kind AI Zone in Riyadh. Qualcomm agreed to build a 500-engineer chipset design center in the Saudi capital.[8]

These were not vague memoranda of understanding. By August 2025, HUMAIN's data center capacity was already sold out — and 99 percent of the demand was international.[9] By January 2026, crates of NVIDIA chips were physically arriving in the kingdom. Amin posted photos on LinkedIn, grinning beside pallets of processors worth billions.

To understand the velocity, consider the leadership. Amin is not a Saudi bureaucrat. He's a Portland State–educated electrical engineer who built Rakuten's mobile network in Japan, served as a senior vice president at Reliance Jio in India — the company that brought a billion Indians online — and then ran Aramco Digital, the oil giant's tech subsidiary.[10] His deputy, Dr. Yaser Al-Onaizan, previously led the National Center for AI under Saudi Arabia's data authority, but before that held senior positions at Amazon Web Services and IBM Research, where he accumulated more than 20 patents and 60 published papers. The CFO, Samih Elhage, came from McKinsey and Nokia. These are not provincial appointments. They are Silicon Valley–grade executives who happen to be building in the desert.[11]

The sheer financial scale of HUMAIN defies easy comparison. The company's estimated infrastructure buildout through 2034 — 6.6 gigawatts of data center capacity — carries a price tag of roughly $77 billion.[12] On top of that sits a $10 billion venture capital fund called Humain Ventures, targeting AI startups in the US, Europe, and Asia. In February 2026, HUMAIN invested $3 billion in Elon Musk's xAI during its $20 billion Series E round, becoming a significant minority shareholder.[13] When xAI was subsequently folded into SpaceX, HUMAIN's holdings converted to SpaceX equity — giving Saudi Arabia's sovereign wealth fund a stake in America's most strategically important rocket company.

In November 2025, HUMAIN led a $900 million Series C round for Luma AI, the generative video startup, at a $4 billion valuation.[14] Saudi Aramco, the national oil company, separately invested in HUMAIN itself — creating a striking loop in which oil wealth is recycled directly into AI infrastructure.[15]

What makes HUMAIN architecturally unusual is a strategy Amin calls “silicon diversity.” Rather than betting exclusively on NVIDIA — as most American hyperscalers have done — HUMAIN has spread its chip procurement across NVIDIA, AMD, Qualcomm, Groq, and AWS's custom Trainium processors.[16] The logic is both strategic and economic: no single supplier bottleneck, and the ability to match different chip architectures to different workloads. HUMAIN deployed open-source models on Groq's inference cluster in its sovereign data centers, achieving over 1,000 tokens per second on a 20-billion-parameter model.[17]

AMD and Cisco formed a joint venture with HUMAIN in November 2025 targeting one gigawatt of AI infrastructure by 2030.[18] xAI agreed to build a 500-plus-megawatt flagship data center in Saudi Arabia — its first facility outside the United States — with nationwide deployment of Grok models integrated into HUMAIN's platform.[19] Adobe became HUMAIN's first global data center customer, partnering on multimodal generative AI tuned for Arab culture.[20] The US Commerce Department formally approved the export of chips equivalent to 35,000 NVIDIA GB300 processors to HUMAIN — a decisive reversal of the Biden administration's restrictive AI Diffusion Rule, which had lumped Saudi Arabia into the same tier as countries deemed high-risk for technology diversion.[21]

AirTrunk, the Blackstone-backed data center developer, signed a $3 billion agreement to build facilities in the kingdom.[22] In January 2026, HUMAIN secured up to $1.2 billion in additional financing to accelerate infrastructure expansion.[23] HUMAIN's CEO has publicly stated a goal of a dual-listing IPO on the Saudi Tadawul exchange and Nasdaq within three to four years.[24]

If infrastructure is HUMAIN's skeleton, its Arabic-language large language model, ALLaM, is its beating heart — and the strongest argument for why this company matters beyond balance sheets.

Arabic is spoken by more than 450 million people across 27 countries. It is the world's fifth most-spoken language.[25] And it is catastrophically underrepresented in the datasets that power the AI systems Americans use every day. The language presents staggering computational challenges: a single Arabic root can generate dozens of morphological forms; there are 27-plus national dialects that differ so substantially that a single word — بس — can mean “only” in Egypt, “but” in Lebanon, and “enough” in the Gulf. Code-switching between Arabic, English, and French is common in everyday speech.[26]

ALLaM was built to address this gap directly. The model family spans 7 billion to 70 billion parameters, with the flagship ALLaM 34B trained on more than 500 billion Arabic tokens — described as the largest Arabic-language training dataset ever assembled. The development team — 120 AI specialists including 35 PhDs, with equal gender representation — drew on data contributed by 16 Saudi public entities, 300 Arabic-language books, and validation by over 600 domain experts.[27] Unlike most Arabic AI efforts, which fine-tune English-language models with translated data, ALLaM was designed to think natively in Arabic. Independent evaluation confirmed it as the most advanced Arabic LLM built in the Arab world.[28]

HUMAIN launched a consumer-facing application called HUMAIN Chat powered by ALLaM 34B — available on iOS, Android, and web, with speech input that handles multiple Arabic dialects.[29] On Arabic MMLU benchmarks, the 70-billion-parameter version achieved state-of-the-art results across five of eight benchmark categories, scoring 4.92 out of 5 on generation and code-switching tasks in independent evaluations.[30]

In October 2025, HUMAIN unveiled HUMAIN ONE, an agentic AI operating system for enterprises and government. Rather than launching discrete applications, users express intent in natural language, and more than 150 AI agents orchestrate tasks across connected systems — HR, finance, procurement, productivity — through a unified interface.[31] At the PIF Private Sector Forum in February 2026, Amin went further, unveiling HUMAIN OS, described as the first AI-native operating system to be commercialized outside the US and China. The US-based company Turing became HUMAIN's first American enterprise customer in March 2026, building the world's first enterprise AI Agent Marketplace on the platform.[32]

Here is the argument that American AI discourse systematically ignores: the concentration of AI capability in four or five San Francisco companies is not a feature of the technology. It is a bug. And the emergence of sovereign AI actors like HUMAIN is precisely the corrective the world needs.

Consider the arithmetic. The four largest American hyperscalers spent roughly $325 billion on infrastructure in 2025 alone. Their models are trained overwhelmingly on English-language data. Their safety frameworks, cultural assumptions, and content moderation policies reflect American and broadly Western norms. When a farmer in Bangladesh asks an AI system a question in Bengali, when a doctor in Cairo tries to use an AI diagnostic tool in Arabic, when a teacher in Lagos wants to generate curriculum materials in Yoruba — these users are, at best, an afterthought. At worst, they receive outputs filtered through what one researcher described as “a Western lens rather than an Arab one.”[33]

HUMAIN's CEO articulates the alternative plainly: “It is good for humanity to have knowledge — especially around AI — not to be all centralized in one location.”[2] This is not merely corporate messaging. Saudi Arabia can offer AI compute at roughly 30 percent less than equivalent US processing costs, thanks to abundant cheap energy — both solar and fossil — and a geographic position at the nexus of three continents, within 2,000 miles of 2.9 billion people.[34] If HUMAIN achieves even a fraction of its 6.6-gigawatt ambition, it could serve as a compute provider for emerging markets across the Middle East, Africa, and South Asia that currently have no viable path to large-scale AI infrastructure.

The broader landscape reinforces the trend. India's BharatGen initiative is building sovereign multilingual models for 22 Indian languages. The UAE's Falcon and Jais models are open-source and freely available. Egypt's Intella has spent 18 months building one of the most diverse Arabic dialectal datasets for its banking and telecom AI agents. Brazil is investing $320 million in an AI-capable supercomputer.[35]

Benjamin Rosman, a researcher at the University of the Witwatersrand who builds language models for African languages, captured the spirit of the moment: “How do we make sure that the entire planet benefits from AI? I want more and more voices to be in the conversation.”[36] An ACM Communications survey published in 2025 concluded that Arabic-language AI models “hold significant potential across multiple sectors, offering transformative applications in education, governance, healthcare, and cultural preservation.”[37] These are not speculative benefits. They are the practical consequences of building AI systems that actually work for the populations that need them most.

None of this happens in a vacuum, and any honest accounting of HUMAIN must reckon with the geopolitical complexities. Saudi Arabia is not a democracy. Critics argue that advanced AI in Saudi hands could enable more efficient surveillance and dissent-tracking.[38] These concerns deserve serious engagement, not dismissal.

But they also deserve proportion. The United States exports weapons systems, surveillance technology, and intelligence-sharing frameworks to Saudi Arabia and has done so for decades. The question of whether to sell AI chips to the kingdom is not qualitatively different from the question of whether to sell F-15s — it is merely newer. Carnegie Endowment fellows noted that the initial chip approval for HUMAIN represented a “manageable level of risk” — enough for commercial AI development, not enough to threaten America's frontier lead.[39]

CSIS analysts have drawn a provocative parallel, suggesting that compute could become to the 21st century what oil was to the 20th — and that the US-Saudi relationship could evolve into a “compute-dollar” arrangement analogous to the petrodollar system.[40] Saudi Arabia is positioning itself as a “geopolitical swing state” in AI, maintaining deep commercial ties with the United States while preserving economic relationships with China.[41] This dual positioning creates tension but also leverage — and in a world where AI infrastructure is genuinely essential, having more providers in more geographies may be stabilizing rather than destabilizing.

The talent question looms large. Saudi Arabia currently has roughly 11,000 trained AI specialists, compared to hundreds of thousands in the US and China. But the trajectory is striking: Stanford's AI Index ranks Saudi Arabia in the top 10 globally for net AI talent migration. The SAMAI Initiative trained over 1.1 million Saudi citizens with AI certifications in 2025, with 52 percent female participation. The 2024 Global AI Index ranked Saudi Arabia first in the world in government AI strategy, and the Public Sector AI Adoption Index 2026 ranked it first globally for public-sector AI implementation.[42]

In Riyadh, they have a phrase for what they're doing. Amin used it at the PIF Private Sector Forum in February 2026: Saudi Arabia intends to be “the first country outside the US and China that will commercialize its own operating system.” The ambition is to become the third pole of artificial intelligence — not a copycat, not a client state for American technology, but a genuinely independent AI power with its own models, its own infrastructure, and its own agenda.[43]

Whether HUMAIN can execute at this scale remains an open question. Many of its deals are framework agreements, not binding contracts. Construction timelines may slip. The geopolitical environment is volatile. Competition from the UAE's Stargate Campus — a joint venture with OpenAI and SoftBank — is formidable. Some technical analysts note that the four big US hyperscalers still outspend everyone, and that HUMAIN's commercial AI infrastructure remains in early stages.[44]

But the skeptics may be missing the forest for the trees. HUMAIN doesn't need to beat Google or Microsoft to matter. It needs to build Arabic-language AI that actually works for 450 million people. It needs to provide affordable compute infrastructure for countries that will never build their own. It needs to prove that artificial intelligence can be developed outside the American monoculture — with different cultural assumptions, different linguistic foundations, and different priorities. The company has already signed its first American enterprise customer, is serving users in more than 130 countries through its Groq-powered inference infrastructure, and is planning a dual-listing IPO within three to four years.[45]

Saudi Arabia designated 2026 as its official “Year of Artificial Intelligence.”[46] For a country that built its wealth on extracting carbon from the ground, the symbolism of pivoting to extracting intelligence from silicon is not subtle. But it may be the most consequential economic transformation any petrostate has ever attempted. And the rest of the world — including the Americans too busy watching OpenAI's next product launch to notice — stands to benefit whether they're paying attention or not.

The AI revolution will not be monopolized. That, more than any benchmark score or chip deal, may be HUMAIN's most important contribution. In a world where intelligence is becoming infrastructure, the question is not whether more nations will build it. The question is whether we'll recognize, before it's too late, that letting them is the point.