NVIDIA Corporation (NVDA.US), in a hurry!

NVIDIA Corporation's (NVDA.US) annual GTC conference is about to open. However, on the eve of the conference, some unusual signals are appearing in the AI computing power industry chain. In the past two years, NVIDIA Corporation has been the most resilient company in the entire AI industry chain. The demand for computing power has exploded, and GPUs have become the most scarce resource in the AI era. Cloud providers such as OpenAI, Microsoft Corporation, Meta, Amazon.com, Inc., and Alphabet Inc. Class C are all lining up to purchase NVIDIA Corporation GPUs. Orders are scheduled for over a year, and "hard to come by" has become the industry norm. In this supply-demand structure, NVIDIA Corporation once had nearly perfect bargaining power: customers had to wait if they wanted computing power. However, in the past few months, a subtle change is emerging. On one hand, OpenAI's Stargate project has stalled and adjusted, while on the other hand, political risks from GEO Group Inc in the Middle East are impacting the pace of data center construction. As the expansion pace of computing power demand intersects with infrastructure uncertainty, the structure of the AI industry chain is quietly changing. Stargate Stalled: AI Computing Power's Largest Orders are Shifting In early 2025, companies such as OpenAI, SoftBank, and Oracle announced a project that shook the entire industryStargate. The scale of this project is unprecedented: a total investment of $500 billion planned to build a 10GW AI computing power infrastructure aimed at supporting OpenAI's model training and inference system. If successful, this would be one of the largest AI infrastructure construction plans in human history. However, reality quickly became complicated. In March 2026, CNBC reported that OpenAI, eager to deploy NVIDIA Corporation's next-generation chips at a new site, decided to no longer expand its flagship Stargate data center project in collaboration with Oracle Corporation. Instead, they sought to establish clusters with the latest generation NVIDIA Corporation GPUs (Rubin) elsewhere. The project, which originally planned to expand the data center from 1.2GW to nearly 2GW, ultimately did not move forward. The current Abilene site is expected to use NVIDIA Corporation's Blackwell processor, but its power supply is not expected to be in place within a year. While Oracle Corporation posted on X on Sunday claiming the reports were "false and inaccurate," the post only stated that the existing project was progressing as planned and did not mention any expansion plans. The original text from Oracle Corporation's X: Source: X "The recent media reports about the Abilene base are false and inaccurate. First, Crusoe and Oracle Corporation (Oracle) are working together in step to deliver one of the world's largest AI data centers in Abilene at record speed. Currently, two buildings are operational, and the rest of the park is progressing according to plan. Secondly, Oracle Corporation has completed leases for an additional 4.5GW of power to fulfill our commitment to OpenAI. We always collaborate with excellent partners and customers, consistently evaluate sites around the world to meet the increasing demands of Oracle Corporation's cloud infrastructure (OCI)." The upgrade speed of AI chips is far faster than the completion speed of data centers. NVIDIA Corporation used to release a new generation of data center processors every two years, but now CEO Jensen Huang demands that the company deliver a new generation every year, with each generation's performance making a significant leap. The Vera Rubin, unveiled and in production at CES this year, has an inference performance five times that of Blackwell. This market reality not only exposes the critical risks of AI trading but also finds Oracle Corporation deeply in crisis due to debt-driven expansion. Reports indicate that Oracle Corporation's book debt has exceeded $100 billion, and free cash flow has turned negative. Among the many cloud technology giants, Oracle Corporation is the only one mainly supported by debt for the construction of ultra-large-scale cloud providers. In contrast, Alphabet Inc. Class C, Amazon.com, Inc., and Microsoft Corporation rely on their large cash flow businesses. Meanwhile, Oracle Corporation's partner Blue Owl has refused to provide funding for additional facilities and plans to lay off 30,000 employees. For the past few years, the AI industry has always believed that GPUs were the biggest bottleneck. However, reality is proving a new fact: the real bottleneck is shifting from "chips" to "infrastructure." The core elements of a typical AI data center include: GPUs, power, cooling, networking, and land. In training-level AI clusters, a data center often requires hundreds of megawatts of power. For reference: the power scale of a 1GW data center is already close to the power consumption of a small city. This means that the expansion of AI computing power is not only about chips but also about energy and infrastructure. And this is exactly one of the difficulties the Stargate project is facing. Financing, power supply, construction cyclesany delay in any link could slow down AI infrastructure. Interestingly, this stalled project immediately attracted new buyers. According to sources, Meta is considering taking over this data center resource. More importantly, NVIDIA Corporation actively participated in brokering this potential deal. In other words, NVIDIA Corporation is no longer just selling GPUs to customers. It has started helping customers grab data centers, power resources, and computing power positions, which was almost unthinkable in the past. Middle East: The New Battlefield for AI Computing Power Just as fluctuations in AI infrastructure are appearing in the US, another crucial infrastructure region is also facing challengesthe Middle East. According to a report from Research And Markets in June 2025, the Middle East currently has approximately 170 data centers, with about 111 projects in the planning or construction stages. The region's existing computing power capacity is approximately 1.2GW, and future planned capacity is close to 4.5GW. It is estimated that by 2027, about $12 billion in new investments will flow into data centers under construction in the Middle East. According to a rough estimate by Data Center Map, Israel has 66 data centers, Saudi Arabia has 61, the UAE has 57, and Qatar has 11 data centers in the Middle East. This means that the Middle East has become an important new battleground for global AI infrastructure. Structurally, the data center market in the Middle East exhibits a clear dual-center pattern: the UAE is the most concentrated in terms of existing stock. The UAE is one of the countries with the highest data center density in the Middle East, with approximately 32 in the capital Abu Dhabi and approximately 23 in Dubai. Saudi Arabia, on the other hand, has become one of the most active markets in the Middle East in terms of new data center projects. Data shows that Saudi Arabia accounts for nearly 60% of the total power capacity in the emerging data center market in the Middle East and is expected to add about 350MW of data center power capacity by the end of 2025. In the past two years, global tech giants have almost simultaneously turned their attention to the Middle East. Why is this the case? In one sentence, because there is money, land, power, and policy windowsall four things are present at the same time. For super-large-scale AI data centers, scarcity is not just about GPUs but also capital, land, power, and policy channels. And the Middle East, indeed, offers a rare combination of conditions in these four dimensions: on one hand, sovereign funds are willing to provide long-term capital for major projects, and on the other hand, countries like the UAE and Saudi Arabia have relatively abundant land and energy resources, while also hoping to use AI and cloud infrastructure to transform their economic structures and transform themselves from traditional energy centers into new global digital hubs. This is why, starting in 2023, global cloud providers and AI companies began systematically laying out computing power infrastructure in the Middle East. Oracle was the first to clearly step up its game. On February 6, 2023, Oracle announced a $1.5 billion investment in Saudi Arabia to expand its cloud infrastructure capability and drive the development of the Riyadh public cloud region. This investment, along with its cloud layout in Jeddah, NEOM, and other places, forms Oracle's long-term strategy in Saudi Arabia. Subsequently, in 2024, the bets started to pick up speed. On March 4, 2024, AWS announced the construction of a new cloud region in Saudi Arabia, planning to invest over $5.3 billion in the country with the goal of enabling it in 2026. This is one of AWS's most significant infrastructure commitments in Saudi Arabia, reflecting Saudi Arabia's desire to take on more government, business, and AI-related workloads through local cloud regions, while AWS hopes to further bind itself to the future digital infrastructure growth in the Middle East. Next, on March 16, 2024, Microsoft announced a $1.5 billion investment in G42, an AI company in Abu Dhabi. This transaction, on the surface a stake investment, represents a deeper integration of Microsoft into the AI and cloud ecosystem in the UAE. By March 2026, Reuters reported that Microsoft's total committed investment in the UAE from 2023 to 2029 had reached $15.2 billion, with $7.3 billion already invested; this includes the $1.5 billion investment in G42 as well as over $4.6 billion in AI and cloud data center capacity construction. In the second half of 2024, Google Cloud also made a significant move. On October 30, 2024, the Saudi Public Investment Fund (PIF) and Google Cloud announced a collaboration to build a new global AI hub near Dammam in the eastern province of Saudi Arabia. Subsequently, on May 13, 2025, Google Cloud and PIF further announced progress on the project, specifying that the AI hub would be jointly invested in $10 billion by both parties and operated with the participation of local Saudi tech company Humain. Another representative project of sovereign AI is Stargate UAE. According to Reuters' report on May 22, 2025, this project in Abu Dhabi is being driven by G42 in collaboration with OpenAI, Oracle, NVIDIA Corporation, Cisco Systems, Inc., SoftBank, and others. The entire park is planned to ultimately reach a scale of 5GW, with an initial phase of 1GW, with the first batch of 200MW expected to go online in 2026. Reuters, citing TrendForce, estimates that the initial phase roughly corresponds to the deployment of 100,000 NVIDIA Corporation advanced AI chips. Whether from a power or chip perspective, this is not a typical data center expansion; it is pushing the Middle East directly to the forefront of the global AI super park competition. Therefore, the significance of the Middle East to NVIDIA Corporation is not just about selling more GPUs. In the US, AI data centers are constrained by power, construction cycles, and project progress; in the Middle East, massive capital and policy will provide new carrying space. So, NVIDIA Corporation's situation becomes: on one hand, it has to watch whether the ultra-large projects in the US are slowing down, and on the other hand, it must ensure that these new parks in the Middle East can smoothly absorb its chips and systems. The issue in the Middle East lies in wars. Another aspect of the computing power story in the Middle East is the political risk of GEO Group Inc. On March 2, 2026, AWS disclosed that some of its data centers in the UAE and Bahrain were damaged in drone attacks. Two facilities in the UAE were directly hit, and one facility in Bahrain was physically impacted by nearby attacks; AWS clearly stated that these attacks caused structural damage, power transmission interruptions, and secondary water damage from firefighting measures, with a lengthy recovery process expected. Reuters also noted that this was the first disruption of a large US tech company's data center due to military action and has already affected Financial Institutions, Inc. and core cloud services that rely on AWS. The significance of this event is substantial: it indicates that the risk facing AI infrastructure development in the Middle East has shifted from political premiums of the GEO Group Inc to actual facility damage and business disruption. The second layer of risk lies in the rising costs of investment and financing. Data centers are long-cycle, capital-intensive projects, and if regional conflicts persist, developers and cloud providers will face not only increased security spending but also rising insurance costs, more expensive debt financing, and extended project return periods. Reuters, citing JPMorgan's assessment on March 6, 2026, states that the escalation of Gulf conflict will increase risks to domestic investment, foreign direct investment, and talent attraction; projects and institutions relying on bond financing will also face higher financing costs. For countries like Saudi Arabia, which depend heavily on sovereign funds to promote their "2030 Vision," these funds are not just financial investors but also the main sources of funding for large-scale transformation projects, so if the macro environment deteriorates, their "financial and operational constraints" will increase. For NVIDIA Corporation, the most realistic aspect is that demand expectations will be repriced. The market was willing to give the Middle East a high imagination space because there was more than $5.3 billion in AWS investments in a Saudi cloud region, $10 billion AI hub collaborations between Google Cloud and PIF, $1.5 billion Oracle cloud infrastructure expansion in Saudi Arabia, and a huge project like Stargate UAE reaching 5GW. Reuters on March 2, 2026, has already put these projects side by side and bluntly stated that the escalation of the regional situation is bringing external focus back to the AI investment risks of major tech companies in the Middle East. For NVIDIA Corporation, the issue is not just whether these projects will proceed but rather: can these projects progress as planned, go online at the original pace, and digest high-end GPUs in the original scale. If any link is delayed, the optimistic expectations of the capital market for high-performance chip shipments in 2026-2027 will be discounted. For the large cloud giants, daring to stake critical computing power, data, and business continuity in that region is also a question. Conclusion It must be said that the computing power war in the AI era is escalating. If the AI wars of 2023-2024 were about GPUs, then the AI wars of 2025-2027 will become a battle of computing power infrastructure. The focus of competition will shift to data centers, power, networking, cooling, and political risks. And in this war, NVIDIA Corporation is both the biggest winner and bears the greatest risks. Because all expansions of the AI industry chain rest on its GPUs. When the industry chain is in the wind, it is a king, but when demand fluctuates, infrastructure is obstructed, and political intervention from GEO Group Inc occursNVIDIA Corporation must personally step in. Not to sell more GPUs but to ensure that these GPUs have a place to be used. This article was reproduced from Semiconductor Industry Observation, edited by Chen Wenfang for GMTEight.