NVIDIA's annual GTC conference is approaching, but unusual signals are emerging from the AI computing power supply chain just before the event. Over the past two years, NVIDIA has been one of the most composed companies in the entire AI industry. With computing demand exploding, GPUs have become the most scarce resource in the AI era. Whether it's OpenAI, Microsoft, Meta, Amazon, or Google, all cloud providers are queuing up to purchase NVIDIA GPUs. Order backlogs often stretch beyond a year, with "GPU shortages" becoming an industry norm. Under this supply-demand structure, NVIDIA once held near-perfect pricing power: customers needing computing power had no choice but to wait.
However, a subtle shift has emerged in recent months. On one hand, OpenAI's Stargate project has stalled and undergone adjustments; on the other, Middle Eastern geopolitical risks are impacting data center construction timelines. As the pace of computing demand expansion intertwines with infrastructure uncertainties, the structure of the AI industry chain is quietly changing.
Stargate's stagnation represents uncertainty for the AI sector's largest computing order. In early 2025, OpenAI, SoftBank, Oracle and other companies announced the Stargate project, which sent shockwaves through the industry. The project's scale was unprecedented: a planned total investment of $500 billion aimed at building 10GW of AI computing infrastructure, primarily to support OpenAI's model training and inference systems. If successfully implemented, this would have been one of the largest AI infrastructure construction projects in human history.
But reality soon proved complicated. By March 2026, according to CNBC reports, OpenAI decided against expanding its flagship Stargate data center project with Oracle, opting instead to seek locations with newer-generation NVIDIA GPUs (Rubin). The project originally planned to expand the data center scale from 1.2GW to nearly 2GW, but this expansion ultimately failed to proceed. The current Abilene site is expected to use NVIDIA's Blackwell processors, but its power supply isn't expected to be available within a year.
Although Oracle posted on X that related reports were "false and inaccurate," the post only stated that existing projects were proceeding as scheduled without mentioning any expansion plans. Oracle's original post stated: "Recent media reports about the Abilene base are false and inaccurate. First, Crusoe and Oracle are working in lockstep to deliver one of the world's largest AI data centers in Abilene at record-breaking speed. Two buildings are already operational, and the rest of the campus is progressing as planned. Second, Oracle has completed leasing an additional 4.5GW of electricity to fulfill our commitment to OpenAI. We consistently collaborate with excellent partners and customers, continuously evaluating global sites to meet the growing demand of Oracle Cloud Infrastructure."
AI chip upgrade speeds are far outpacing data center completion rates. NVIDIA previously released new data center processors every two years, but now CEO Jensen Huang requires the company to deliver new generations annually, with each generation achieving leapfrog performance improvements. The Vera Rubin architecture, unveiled at CES in January and already in production, delivers five times the inference performance of Blackwell.
This market reality not only exposes key risks in AI commerce but also leaves Oracle's debt-driven expansion deeply mired in crisis. Reports indicate Oracle's book debt has exceeded $100 billion, with free cash flow turning negative. Among major cloud technology giants, Oracle stands alone as the only hyperscale cloud provider primarily relying on debt to fund construction, while Google, Amazon and Microsoft depend on their substantial cash flow businesses.
Meanwhile, Oracle's partner Blue Owl has refused to fund additional facilities and plans to cut 30,000 jobs. For years, the AI industry believed GPUs represented the biggest bottleneck. But reality is proving a new truth: the real bottleneck is shifting from "chips" to "infrastructure."
A typical AI data center's core requirements include: GPUs, electricity, cooling, networking, and land. In training-grade AI clusters, a single data center often requires hundreds of megawatts of power. For reference: a 1GW data center's power requirements approach the electricity consumption of a small city. This means AI computing expansion isn't just a chip problem—it's fundamentally an energy and infrastructure challenge. This precisely represents one reason the Stargate project encountered difficulties. Financing, electricity, construction cycles—delays in any环节 can slow AI infrastructure development.
Interestingly, this shelved project immediately attracted new potential buyers. According to sources, Meta is considering taking over these data center resources. More significantly, NVIDIA is actively facilitating this potential transaction. In other words, NVIDIA is no longer just selling GPUs to customers. It has begun helping clients secure data centers, electricity resources, and computing capacity—something nearly unimaginable in the past.
The Middle East has emerged as a new battleground for AI computing power. While US AI infrastructure faces fluctuations, another crucial infrastructure region—the Middle East—also faces concerning developments. According to a June 2025 Research And Markets report, the Middle East currently hosts approximately 170 data centers, with about 111 additional projects planned or under construction. Regional existing computing capacity stands at approximately 1.2GW, with future planned capacity nearing 4.5GW. By 2027, approximately $12 billion in new investments are expected to flow into Middle Eastern data center construction.
Rough statistics from Data Center Map indicate Israel possesses 66 data centers, Saudi Arabia has 61, the UAE has 57, and Qatar has 11. This means the Middle East has become a significant new battlefield for global AI infrastructure.
Structurally, the Middle Eastern data center market shows a clear dual-center pattern: The UAE has the most concentrated existing capacity, being one of the region's most data center-dense countries, with Abu Dhabi hosting about 32 facilities and Dubai about 23. Saudi Arabia has become one of the most active markets in new data center projects, accounting for nearly 60% of total power capacity in emerging Middle Eastern data center markets, with approximately 350MW of new data center power capacity expected by end-2025.
Over the past two years, global tech giants have simultaneously turned their attention to the Middle East. Why? In summary: money, land, electricity, and policy windows are all simultaneously available. For hyperscale AI data centers, scarcity extends beyond GPUs to include capital, land, electricity, and policy access. The Middle East uniquely offers rare combined conditions across all four dimensions: sovereign funds willing to provide long-term capital for major projects, while UAE and Saudi Arabia possess relatively abundant land and energy resources, simultaneously seeking to transform their economic structures through AI and cloud infrastructure—transitioning from traditional energy centers to new global digital hubs.
Consequently, since 2023, global cloud providers and AI companies have systematically布局 computing infrastructure in the Middle East. Oracle was the earliest to clearly intensify commitments, announcing on February 6, 2023, a $1.5 billion investment in Saudi Arabia to expand cloud infrastructure capabilities and promote Riyadh's public cloud region development. This investment, combined with cloud布局 in Jeddah and NEOM, constituted Oracle's long-term positioning in Saudi Arabia.
By 2024, betting clearly accelerated. On March 4, 2024, AWS announced plans to build new cloud regions in Saudi Arabia with over $5.3 billion investment, targeting 2026 activation. This represented one of AWS's most significant infrastructure commitments to Saudi Arabia, reflecting Saudi ambitions to host more government, enterprise and AI-related workloads through local cloud regions, while AWS sought to绑定 Middle Eastern digital infrastructure growth.
Subsequently, on April 16, 2024, Microsoft announced a $1.5 billion investment in Abu Dhabi AI company G42. While表面上 appearing as equity investment, this transaction实质上 represented Microsoft embedding itself deeper into UAE's AI and cloud ecosystem. By March 2026, Reuters disclosed Microsoft's total committed investment in the UAE for 2023-2029 reached $15.2 billion, with $7.3 billion already deployed—including the aforementioned $1.5 billion G42 investment plus over $4.6 billion for AI and cloud data center capacity construction.
By second-half 2024, Google Cloud formally entered the arena. On October 30, 2024, Saudi Public Investment Fund (PIF) and Google Cloud announced cooperation to build a new global AI hub near Dammam in Eastern Province. Subsequently on May 13, 2025, Google Cloud and PIF further announced project advancement, clarifying this AI hub would involve $10 billion co-investment with local Saudi tech company Humain participating in launch and operations.
Another highly sovereign AI representative project is Stargate UAE. According to May 22, 2025 Reuters reports, this project landed in Abu Dhabi, promoted by G42 together with OpenAI, Oracle, NVIDIA, Cisco and SoftBank. The entire campus's final planned scale reaches 5GW, with Phase 1 at 1GW—the first 200MW expected online by 2026. Reuters cited TrendForce estimates that Phase 1 alone roughly corresponds to 100,000 advanced NVIDIA AI chips. Whether measured by power or chip metrics, this isn't ordinary data center expansion but directly pushes the Middle East to the frontline of global AI super-campus competition.
Consequently, the Middle East's significance to NVIDIA extends beyond selling more GPUs. In the US, AI data centers face constraints from electricity, construction cycles and project推进节奏; whereas the Middle East offers new承载空间 through massive capital and policy determination. Thus, NVIDIA faces a dual situation: monitoring whether US megaprojects slow down while ensuring Middle Eastern new campuses can顺利 absorb its chips and systems.
The Middle East's problem: war represents the other side of the computing story—geopolitical risk. On March 2, 2026, AWS disclosed some UAE and Bahrain data centers suffered damage from drone attacks. Two UAE facilities were directly hit, while Bahrain facilities suffered physical impact from nearby attacks. AWS explicitly stated these attacks caused structural damage, power delivery interruptions, and secondary water damage from fire suppression measures, with recovery expected to take considerable time.
Reuters noted this marked the first time US major tech company data centers experienced disruption from military action, already affecting some AWS-dependent financial institutions and core cloud services. This incident carries significant implications: it demonstrates Middle Eastern AI infrastructure risks have escalated from geopolitical premiums to actual facility damage and business interruption.
The risk's second layer involves rising investment and financing costs. Data centers inherently represent long-cycle, capital-intensive projects. Once regional conflicts persist, developers and cloud providers face not only increased security expenditures but also rising insurance costs, more expensive debt financing, and extended project回报 cycles. March 6, 2026 Reuters cited JPMorgan判断 that Gulf conflict escalation would increase risks for local domestic investment, foreign direct investment and talent attraction; simultaneously, projects and institutions relying on bond issuance for financing would face higher funding costs.
For countries like Saudi Arabia高度依赖 sovereign funds推进 "Vision 2030," sovereign funds aren't just financial investors but primary funding sources for major transformation projects. Therefore, once macro environments deteriorate, their "financial and operational constraints" will increase.
For NVIDIA, the most practical layer involves demand expectations being repriced. Previously, markets willingly assigned extremely high imagination space to the Middle East because it hosted not only AWS's over $5.3 billion Saudi cloud region investment, Google Cloud and Saudi PIF's planned $10 billion AI hub, Oracle's $1.5 billion Saudi cloud infrastructure expansion, but also 5GW-scale megaprojects like Stargate UAE. March 2, 2026 Reuters reporting already并列呈现 these projects,直言 regional situation escalation is refocusing attention on major tech companies' Middle Eastern AI investment risks.
For NVIDIA, the question isn't just "whether these projects will eventually proceed" but rather: can these projects advance according to original plans,架设 according to original节奏, and消化 high-end GPUs according to original scale? Any环节 delays will discount capital market optimism about 2026-2027 high-performance computing chip shipments.
For major cloud giants, whether they dare to stake critical computing power, crucial data and essential business continuity there also remains questionable.
In conclusion, the computing war of the AI era is intensifying. If 2023-2024 represented the GPU war, then 2025-2027 will become the AI infrastructure war. Competition focus will shift to data centers, electricity, networking, cooling and geopolitics. In this war, NVIDIA stands as both the biggest winner and bears the greatest risks, since all AI industry chain expansion depends on its GPUs. When the industry chain enjoys tailwinds, NVIDIA reigns supreme. But when demand fluctuates, infrastructure encounters obstacles, and geopolitics intervenes—NVIDIA must personally enter the arena. Not to sell more GPUs, but to ensure these GPUs have places to be used.
Comments