HUMAIN Was Born From Vision 2030. Its Files Show a Kingdom Rewiring Itself.

The PIF-backed AI company went from launch to a 6.6 gigawatt promise in 18 months, against a 467 megawatt operating base, 2025 to 2026

The last time Saudi Arabia announced a national technology champion with this much capital behind it, the year was 2000 and the asset was oil, not compute. The document that matters today is dated differently. In February 2026, HUMAIN, the artificial intelligence company owned by the Public Investment Fund, signed a contract worth SR 1.88 billion, about 501 million dollars, with the systems integrator MIS to design and build a private AI data center. One contract, twelve months of work, one line in a supplier’s quarterly report. It is the most ordinary document in the HUMAIN file, and for that reason the most informative.

HUMAIN is normally covered through its headline numbers: partner announcements exceeding 23 billion dollars at launch, a stated ambition of 6 to 6.6 gigawatts of AI capacity by 2034, a chairman who is also the Crown Prince. But a company is legible through its contracts, its land parcels and its construction starts, because those cannot be re-announced. This article reads HUMAIN through that lens, in sequence, from May 2025 to late 2026, and asks what structure the sequence reveals.

The paper entity, May 2025

HUMAIN was launched on 12 May 2025 at the PIF Private Sector Forum in Riyadh, under the chairmanship of Crown Prince Mohammed bin Salman, with Tareq Amin as chief executive. Amin’s file matters to the reading. He came from Rakuten in Japan, where he served as group CTO and deputy CEO and built a cloud-native mobile network. He is a telecom infrastructure executive, not an AI researcher. PIF’s own annual report describes the mandate in plain terms: build the entire AI stack, data centers, cloud, models and applications, and reduce the economy’s reliance on oil.

Two days after launch, the first partnership wave arrived. On 13 May 2025, AMD committed to a collaboration worth up to 10 billion dollars over five years, targeting 500 megawatts of AI compute. The same day, Amazon Web Services announced a joint investment of more than 5 billion dollars for a dedicated AI Zone in the kingdom. On 29 May 2025, Humain Ventures appeared, a venture fund of 10 billion dollars aimed at AI startups across Europe, America and Asia. Three announcements in seventeen days, all capital, none concrete.

The concrete arrived in August 2025, when HUMAIN started construction on its first two campuses in Riyadh and Dammam, each initially rated at least 100 megawatts, with a network built on 18,000 NVIDIA Blackwell chips. August is when a paper company starts pouring foundations. Everything before it was announcements; everything after it can be audited.

Table 1: HUMAIN behavior nodes, May 2025 to September 2026

Date Action Verifiable artifact
12 May 2025 Company launched under PIF, chaired by MBS PIF annual report entry
13 May 2025 AMD deal, up to 10 billion dollars, 500 MW AMD announcement
13 May 2025 AWS AI Zone, over 5 billion dollars AWS announcement
29 May 2025 Humain Ventures fund, 10 billion dollars Company statement
August 2025 Riyadh and Dammam construction starts Site starts, 18,000 Blackwell chips
November 2025 xAI framework for 500 MW GPU network; AMD and Cisco JV targeting 1 GW by 2030 Joint statements
November 2025 AWS to deploy 150,000 AI accelerators in Riyadh AI Zone AWS announcement
February 2026 MIS contract, SR 1.88 billion, private AI data center MIS disclosure, Q2 2026 financials
February 2026 xAI investment of 3 billion dollars Company statement
September 2026 HUMAIN M3, Arabic large language model Product release

Sources: PIF annual report 🔗 and operator disclosures, 2025 and 2026.

What the sequence reveals

Read the table as a historian rather than a fan. The first pattern is that capital announcements preceded physical assets by three months, and physical assets preceded products by more than a year. HUMAIN M3, the Arabic-first model released in September 2026, came sixteen months after launch and four months after the first private data center contract. The stack is being built bottom-up: power and shells, then chips, then a model on top. This is the order a country with cheap electricity and deep pockets would choose, and the reverse of how American labs grew.

The second pattern is JV-by-JV control. HUMAIN’s joint venture with center3, the STC subsidiary, is split 51 to 49 in HUMAIN’s favor, starting at 250 megawatts and targeting 1 gigawatt. The land inventory, 211 plots secured, sits in Riyadh and Dammam. Every structure keeps the sovereign fund in the majority seat. The 2016 lineage shows here: Vision 2030 was announced in April 2016, and the digital buildout that followed, from STC’s data centers to SDAIA’s Hexagon facility at roughly 480 megawatts, always kept state entities holding the asset while global vendors supplied the parts.

The third pattern is the vendor map. NVIDIA, AMD, Cisco, Qualcomm, AWS, Google Cloud and Groq, all American or American-listed, appear in the file. No Chinese partner appears anywhere in the public record. The GPUs are designed in California and fabricated in Taiwan. A historian of the 1980s would recognize the shape: a national champion whose critical inputs depend on a foreign government’s export approvals. HUMAIN’s structure does not just enable American technology. It requires continued American permission.

The numbers the ambition sits on

The installed base is the honest denominator. Per the Saudi Ministry of Communications and Information Technology, cited by S&P Global and Forbes Middle East in 2026, national data center capacity rose from 68 megawatts in 2021 to about 440 megawatts in 2025, reaching 467 megawatts in the first quarter of 2026, spread across more than 60 facilities run by over 20 companies. Alvarez and Marsal’s August 2025 report, circulated by the Chinese commerce ministry’s Riyadh office, counted roughly 410 megawatts of IT load across about 40 operating facilities and projected about 1 gigawatt by 2030. The two figures differ by measurement basis, gross capacity versus IT load, and by facility count, and both belong in the record rather than one replacing the other.

Against that base sit the targets: HUMAIN projects about 1.9 gigawatts of associated capacity by 2030 and 6 to 6.6 gigawatts by 2034, per S&P and Forbes Middle East reporting. The MCIT’s own planning figure, carried by Asharq Business in 2026, is 3,400 megawatts of planned national capacity. DataVolt, the local operator, has committed 5 billion dollars for a 1.5-gigawatt campus in NEOM’s Oxagon, beginning with a 300-megawatt phase and designed renewables-first. Per-capita, Saudi Arabia stands at roughly 12 watts of data center capacity per person, against nearly 50 watts in both the United States and the United Arab Emirates, per the Alvarez and Marsal arithmetic on a 34-million population.

Table 2: Saudi data center capacity, measured and announced, 2021 to 2034

Measure Value Basis Source and date
Installed capacity, 2021 68 MW actual MCIT via SPA, 2026
Installed capacity, 2025 440 MW actual MCIT via S&P, 2026
Installed capacity, Q1 2026 467 MW actual MCIT via Forbes ME, 2026
IT load in operation about 410 MW, about 40 sites estimate Alvarez and Marsal, August 2025
Planned national capacity 3,400 MW target MCIT via Asharq, 2026
HUMAIN-associated pipeline, 2030 about 1.9 GW projection S&P Global, 2026
HUMAIN-associated pipeline, 2034 6 to 6.6 GW projection S&P Global, 2026
DataVolt Oxagon campus 1.5 GW, 300 MW first phase contracted NEOM and DataVolt, 2025 to 2026

Source: Forbes Middle East 🔗, 2026; MOFCOM 🔗 summary of Alvarez and Marsal, August 2025.

The investment ledger adds up on the Saudi side. Data center and digital infrastructure investment reached SAR 56.2 billion, about 14.8 billion dollars, per S&P Global, a more than sixfold expansion since Vision 2030 began, per SPA’s coverage in August 2026. Across the wider Gulf, S&P counts 112 operational data centers and 74 under development, with Saudi Arabia carrying about 65 percent of the region’s future power capacity and the Gulf’s total capacity expected to triple past 3 gigawatts within five years. The kingdom is not the only bidder in the race, but it is the largest single column in the ledger.

The neighborhood sets the pace

Saudi Arabia’s numbers read differently against the Gulf’s scoreboard, and the 2026 file has one. S&P Global counts 112 operational data centers across the GCC with 74 under development, and expects regional capacity to more than triple past 3 gigawatts within five years, with Saudi Arabia and the UAE absorbing more than 80 percent of the additions through 2030. Abu Dhabi is building Stargate UAE, the first international deployment of OpenAI’s Stargate platform, a 5-gigawatt campus with a 200-megawatt first phase due to go live in 2026, backed by a partnership with G42, Oracle, NVIDIA and SoftBank. In September 2026, Microsoft announced more than 10 billion dollars of investment across the UAE, Saudi Arabia, Qatar and Kuwait through 2030. Qatar’s Qai, a QIA subsidiary, formed a 20-billion-dollar joint venture with Brookfield for AI infrastructure, and KKR put 5 billion dollars behind Gulf Data Hub’s regional expansion.

Two positions separate the rivals. The UAE leads on operational capacity today, more than 400 megawatts, about 45 percent of the Gulf’s running IT capacity, and leans on global partners, Stargate, Microsoft, a 5-billion-dollar KKR line. Saudi Arabia leads on the future column, about 65 percent of the region’s planned power capacity, carried by a sovereign vehicle rather than a consortium of outsiders. The UAE strategy rents the frontier. The Saudi strategy is attempting to own the substrate under it. HUMAIN is the instrument of that distinction, and its 51 percent JV stakes and 211 land plots are what ownership looks like on paper.

The cloud layer underneath follows the same pattern. AWS is separately investing 5.3 billion dollars in a Saudi cloud region targeted for 2026, Google Cloud anchors a hub in Dammam alongside its HUMAIN collaboration, Oracle’s Riyadh region is already operational, and Microsoft’s footprint routes partly through its 1.5-billion-dollar stake in G42. The hyperscalers arrived before the sovereign champion did, which is precisely why HUMAIN exists. A government that watched foreign clouds win its enterprise workloads in 2023 and 2024 had a structural reason to build a domestic full-stack competitor in 2025.

The lineage: a third generation of the same bet

The historian’s reflex is to check whether any of this is new. The answer is: the asset class is new, the behavior is not. April 2016 produced Vision 2030 and the PIF’s mandate to build non-oil industry. The first generation of digital assets that followed was connectivity: STC’s cable-landing stations and colocation campuses in Riyadh, Jeddah and the Eastern Province, later spun into center3, which still doubles as the kingdom’s peering infrastructure. The second generation was research and state compute: SDAIA, founded in 2019, and the Hexagon data center program, which grew out of the Aramco perimeter and is reported at roughly 480 megawatts, now increasingly aligned to HUMAIN’s offtake. Saudi Arabia’s national supercomputer, Shaheen III, an HPE Cray EX system at KAUST, delivered about 120 petaflops of double-precision performance at commissioning, an HPC asset placed deliberately inside the same research corridor the new HUMAIN campuses now abut.

HUMAIN is the third generation, and the difference is the buyer. The first generation bought bandwidth. The second bought scientific compute for state researchers. The third buys merchant compute capacity at gigawatt scale, with the intent to sell it. Each generation kept the same three-part structure: a state entity owns the asset, a global vendor supplies the technology, and the revenue model borrows from whatever the current decade considers strategic. In 2005 that was fiber. In 2019 it was research HPC. In 2026 it is AI factories. The continuity is the point. HUMAIN is not a break with Saudi industrial history. It is the largest installment of a pattern that has run for twenty years.

How the structure constrains the company

Three constraints in the file have no announcements attached.

The first is the financing geometry. Building a data center shell with power and cooling, excluding IT hardware, runs 11 to 14 million dollars per megawatt of IT load on global benchmarks, per the Alvarez and Marsal report, with Saudi costs likely below the range but not beneath it. At the 6.6-gigawatt decade-end ambition, the implied bill sits in the tens of billions of dollars, against a PIF that also funds NEOM, football clubs and a giga-project portfolio. The development shift described in the report, from end users to third-party operators like DataVolt, center3, Khazna and ezditek, moves the burden from user balance sheets to sponsor balance sheets, and from equity toward debt. Debt has covenants. Covenants have schedules.

The second is the NEOM discount. The Oxagon campus inherits the credibility history of NEOM’s scope reductions and delays, fairly or not. The 2028 target for a 1.5-gigawatt renewables-backed facility compresses a typical three-to-five-year hyperscale cycle, and the undisclosed storage requirement behind the 24/7 renewable claim is the load-bearing unknown. Saudi Compute’s 2026 assessment put the test plainly: the 2026 and 2027 phase deliveries, not the 2028 ribbon-cutting, will reveal whether the largest announced net-zero AI factory is a construction schedule or a press release.

The third is the permission structure. Every GPU in the file is American silicon subject to American export controls, and the 18,000 Blackwell units installed in 2025 arrived under an approval regime that can change with an administration. A company structured to buy the full American stack has bought, along with the hardware, a standing dependency on Washington’s licensing mood. The absence of Chinese alternatives in the public record is not an oversight. It is the visible edge of that dependency.

What a historian would flag for the next file

The open questions are dated, which makes them checkable. By the second half of 2027, the MIS contract should have delivered its private AI data center within twelve months of the February 2026 signing. By 2028, Oxagon’s first 300 megawatts either energize or slip. By 2030, the 1.9-gigawatt HUMAIN projection either appears in MCIT’s installed-base statistics or joins the list of announced capacity that a future revision quietly reclassifies. The 2021 baseline of 68 megawatts, six years ago, makes even 1 gigawatt a fifteen-fold climb, and the gap between 467 megawatts of installed capacity today and 6.6 gigawatts of ambition is the widest in the file.

Vision documents from 2016 promised economic diversification by 2030, and the decade since has produced exactly one generation of real assets: airports, metro lines and now data center shells. HUMAIN is the latest chapter of that pattern, a state champion converting hydrocarbon wealth into an industrial asset class, one contract at a time. Whether the chapter closes with a functioning AI stack or a pile of commitments is not a question of ambition. It is a question of whether the 2027 and 2028 delivery dates hold, and the files from those years have not been written yet.

Data basis

Data as of November 2026. HUMAIN’s corporate timeline is reconstructed from PIF’s annual report entry, operator announcements and MIS’s February 2026 contract disclosure, each node in Table 1 carrying its own verifiable artifact. Capacity figures come from two official sources used side by side: the MCIT series of 68 megawatts in 2021 and 467 megawatts in Q1 2026, carried by SPA and Forbes Middle East, and the Alvarez and Marsal August 2025 report of about 410 megawatts of IT load across about 40 facilities, circulated by MOFCOM’s Riyadh office. The two measure different things, gross installed capacity versus IT load, and this article presents both without reconciling them. Pipeline projections, 1.9 gigawatts by 2030 and 6 to 6.6 gigawatts by 2034, are marked as projections, not contracts, wherever cited. Cost benchmarks of 11 to 14 million dollars per megawatt are global figures from the Alvarez and Marsal report, with Saudi costs noted as likely below the range but not measured here. No site visits or primary interviews were conducted; every node cited is a public disclosure or a published secondary report of one.

Limitations

  • The vendor-announcement values, such as the AMD collaboration of up to 10 billion dollars, are ceilings reported at signing, not disbursed capital, and later revisions are not tracked here.
  • Partnership totals exceeding 23 billion dollars at launch are third-party estimates, not an audited ledger.
  • Contract details behind the MIS and center3 structures come from supplier disclosures, so scope terms beyond the disclosed values are unknown.

Sources

About Global Infinity Labs

Global Infinity Labs is an AI innovation company specializing in MCP data services. This article was prepared by our expert team using insights and data provided through our MCP platforms.

Contact: support@globalinfinitylabs.com