Oracle's strategy of acting as a middleman for AI resources appears to be faltering. Earlier this month, rumors of 30,000 job cuts laid bare the company's financial struggles. A specialized analysis report from TD Cowen indicated that Oracle is facing severe funding pressure due to difficulties financing its AI data center expansion, prompting considerations to lay off 20,000 to 30,000 employees to free up $8 to $10 billion in cash flow. This operational strain has been clearly reflected in the company's stock performance over the past year. The veteran tech giant, with over four decades in the industry, has experienced a dramatic stock price rollercoaster: a steady climb early in the year, followed by a staggering 35% single-day surge on September 10 to $328.33, catapulting its market capitalization to a peak of $922.224 billion. However, this was swiftly followed by a cliff-like descent. As of this writing, the stock price has fallen more than 50% from its peak, with its market value similarly halved.

Oracle, having firmly anchored its transformation towards cloud and AI infrastructure, precisely targeted the core demand for AI computing power. Why, then, did its rapidly ascending stock price suddenly brake hard? Behind this reversal, why has this established company, which has gone all-in on its AI gamble, failed to become the dominant player holding the chips in the AGI era?

**Part 1: All-In on AI, Oracle's Stock Takes a Wild Ride** "Oracle's bet on AI can be described as a last-ditch effort. Having missed the cloud computing wave, it simply cannot afford to miss the AI train," stated Shen Yue, an analyst at a certain institution. As the absolute dominator in the global enterprise database market, Oracle's core business has long held nearly a 45% global market share, with high customer barriers in critical sectors like finance and government, while enterprise software such as ERP and CRM continue to provide stable revenue. However, with the full arrival of the cloud computing era, its foundational database business became a key battleground for cloud providers. Leading players like AWS and Azure have persistently eroded Oracle's traditional base with their cloud-based database products.

This veteran giant's cloud transition has consequently been painful. Although its OCI (Oracle Cloud Infrastructure) business was launched early, its market share has long lingered in the single digits, failing to break into the top tier. A turning point occurred in 2024 when a strategy presentation by Oracle Chairman Larry Ellison fundamentally altered the company's trajectory.

1. **The AI Alliance: Oracle Joins the Feast** At the 2024 AI cooperation announcement involving OpenAI, Microsoft, and Oracle, the 80-year-old Larry Ellison重磅宣布: Oracle would fully transform into an AI infrastructure service provider, focusing on high-compute scenarios for large model training and inference, aiming to build world-leading AI supercomputing clusters. He stated bluntly, "The AI revolution requires massive computing power, and Oracle will be the infrastructure provider for this revolution." This collaboration marked Oracle's formal entry into the AI compute race. Shen Yue noted, "Oracle also has a significant background often overlooked in media reports. Ellison has a very good relationship with Trump; he was among the first tech leaders to support Trump in the last term. With this relationship providing cover, even if it requires substantial debt financing, he is determined to go all-in on AI."

To rapidly implement its strategy and secure a unique niche in the AI infrastructure space, Oracle signed its first exclusive data center agreement with OpenAI in July of the same year. The two parties finalized plans to build a 1-gigawatt AI data center in Texas, USA, utilizing NVIDIA H200 GPU chips paired with OCI's Zettascale architecture, specifically designed to meet OpenAI's high-compute needs for model training. In September, at the Oracle CloudWorld event, the collaboration with NVIDIA deepened further. Oracle was announced as a core purchaser of NVIDIA's H200 GPUs and launched the OCI AI Supercluster service, scalable to 131,072 GPUs. This service, open for industry-wide ordering, explicitly prioritized meeting OpenAI's compute demands, becoming a flagship product for Oracle's AI infrastructure.

On January 22, 2025, OpenAI, SoftBank, and Oracle jointly announced the "Stargate" large-scale AI infrastructure project, planning a cumulative investment of $500 billion over four years to build an AI infrastructure network, with the first batch of 16,000 NVIDIA GB200 chips scheduled for deployment and调试 by summer 2025. Thus, the "ONO" (OpenAI+Nvidia+Oracle) triangular alliance began to take shape. Oracle, leveraging its strategic alliance advantages with NVIDIA and OpenAI, secured massive合作订单 successively.

Reflected in its financials, Oracle's total revenue for fiscal year 2025 (June 2024 to May 2025) reached $57.4 billion, an 8% year-over-year increase. Cloud IaaS revenue, fueled by AI compute布局, surged 50% to $10.3 billion. Remaining Performance Obligations (RPO) soared to $138 billion, a 41% increase, with approximately 90% stemming from AI infrastructure contracts. This represented the initial results of aligning early with OpenAI and NVIDIA and entering the "Stargate" project, marking the preliminary success of Oracle's transition from a traditional software vendor to an AI compute service provider. During the fiscal year (June 2024-May 2025), Oracle's stock price climbed steadily alongside the continuous落地 of AI partnerships, rising from $120 to $160, a cumulative gain of about 33%,初步彰显ing market recognition of the long-term value of its transformation strategy.

2. **Orders Ignite, Stock Peaks** Shen Yue told Leiphone, "The core reason the stock price broke through $320 and peaked on September 10 was that Oracle's compute expansion was deeply tied to NVIDIA, forming a strong驱动 chain that directly ignited extreme market optimism about Oracle's AI transformation. A combination of hype and expectations blasted Oracle's stock price upwards." Compared to the industry's overall pace, Oracle's speed in staking claims in the AI infrastructure arena had already exceeded market expectations. If the previous $138 billion RPO was considered a '投名状' for its transformation, then the series of actions by the "ONO" alliance starting in the second half of 2025 directly pushed this transformation towards full爆发.

In July 2025, cooperation on the OpenAI "Stargate" project deepened further. The two parties jointly announced plans to develop 4.5 gigawatts of新增 AI data center compute capacity in the US. Combined with existing deployments, this would push the project's total capacity past 5 gigawatts, clearly pointing towards the落地 of large-scale compute collaboration. On September 9, Oracle's Q1 FY2026 (June-August 2025) earnings report revealed that its RPO had skyrocketed to $455 billion, a staggering 359% increase year-over-year. The single-quarter addition of $317 billion in contract revenue set a peak for order growth speed in the tech industry over the past decade. Oracle CEO Safra Catz confidently stated in the subsequent earnings call, "Market demand for AI infrastructure significantly exceeds current supply capacity."

The next day, the "ONO Alliance" revealed the explosive news: Oracle and OpenAI formally announced a five-year, $300 billion compute agreement. This rare super-order in the cloud computing sphere is expected to contribute an average of $60 billion in annual revenue starting in 2027 upon正式履约. Concurrently, Oracle also signed significant cloud合作 agreements with xAI and Meta. The落地 of this agreement with OpenAI led the market to consensus: Oracle's RPO breaking $500 billion was only a matter of time. J.P. Morgan stated in a report that this growth intensity was "unprecedented," essentially representing a collective endorsement by the global market of its AI infrastructure capabilities.

On September 10, 2025, Oracle (ORCL.N) experienced an epic surge, with its intraday price reaching a historic high of $345.69. The closing price that day soared 36% from the previous session, adding approximately $244.2 billion to its market capitalization in a single day (estimated based on the gain and pre-surge market cap), lifting its total market cap to around $923 billion. This surge not only set a record for Oracle's largest single-day gain since 1992 but also propelled founder Larry Ellison's net worth to $393 billion, briefly surpassing Elon Musk during trading to become the world's richest person, making it the focus of capital markets that day.

Wolfe Research was the first to react, significantly raising its price target on Oracle from $300 to $400, reflecting strong optimism about its AI compute business. Over a dozen mainstream institutions, including Bank of America and Jefferies, followed suit. Bank of America raised its target from $295 to $368. The core institutions' target price range ultimately focused between $330 and $368.

3. **The Party Cools, The Stock Continues to Fall** However, the euphoria in capital markets often cannot withstand cooling sentiment. After hitting a closing high of $328.62 on September 10, 2025, Oracle entered a cliff-falling模式. Its stock price and market cap have now both halved from their peaks, as the optimistic泡沫 built on super-orders was rapidly punctured.

Shen Yue believes, "Actually, Oracle, OpenAI, and NVIDIA have formed a complementary triangular structure of 'compute chips - cloud delivery - large models.' This structure has essentially tied up the US economy. So, when OpenAI publicly disclosed financial issues and sought government backup measures, it was a foolish and unnecessary move that indirectly harmed Oracle's stock price." He added, "But the specific reasons go back even earlier; the fuse for this cold spell was lit beforehand. Last July, the Wall Street Journal reported that the $500 billion 'Stargate' project, jointly led by OpenAI and SoftBank, was severely behind schedule, with expansion plans significantly scaled back. The $100 billion OpenAI promised to invest 'immediately' had been迟迟没落地."

At the October AI World conference, co-CEO Clay Maguire revealed that the gross margin for the AI data center business was expected to reach 30%-40%. He also announced plans, jointly with OpenAI, to build a massive data center campus in Michigan and proposed issuing $38 billion in bonds to support data center construction in multiple locations. The market, however, was not receptive to Oracle's repeated affirmations of commitment.

Shen Yue分析 that the post-September decline can be divided into two waves with distinct triggers. The first wave stemmed from concerns sparked by capital withdrawal from a core major client. The second wave's core trigger was earnings falling short of market expectations. After the sharp rise and fall, the market became less accommodating.

In December 2025, core investor Blue Owl Capital suddenly announced the termination of a $1 billion合作 for Oracle's data center in Saline Township, Michigan—a key component of the "Stargate" plan. Simultaneously, Blue Owl significantly reduced its stake in Oracle, cashing out over $5 billion. This directly triggered a wave of institutional selling, causing Oracle's stock to plunge nearly 10% in a single day, wiping out about $69 billion in market value and planting the seeds for subsequent continued correction.

Adding insult to injury, the release of the Q2 FY2026 earnings (September-November 2025) also backfired on Oracle. The report showed Q2 total revenue of $16.058 billion, at the lower end of the guidance range. Cloud revenue was approximately $8 billion, also at the lower end. Non-GAAP gross margin was 41.90%, far below the market consensus of 68.7%. Free cash flow dropped to -$10 billion. To cope with order growth, Oracle raised its capital expenditure outlook for FY2026 to $50 billion, representing about 75% of FY2026's expected revenue. The day after the earnings release, Oracle's stock plummeted over 16% intraday. The combination of weaker-than-expected cloud revenue and massive capital expenditure plans pushed market panic over the return on investment for Oracle's data centers and its debt risk to a peak.

**Part 2: What's the Logic Behind the Rapid Cooling of the 'Oracle Fever'?** In just a few months, Oracle's meteoric rise and cliff-like decline arrived just as abruptly. Li Zhe, an analyst at a certain institution, told Leiphone that this rapid cooling of capital enthusiasm was not a sudden 'black swan' event from a single risk, but rather the inevitable feedback from the market after multiple underlying concerns continued to ferment. "The recent frequent fluctuations in Oracle's stock price and CDS are mainly due to two reasons: First, people believe OpenAI's actual demand may not be as large as the order volume suggests, and Oracle's stance has also been wavering. Second, Oracle has too many data center orders on hand but faces tight funding; people doubt these data centers can be delivered on time."

1. **Two-Way Game: The Uncertain Path of Cooperation** A UBS report released in January pointed out that Oracle's stock decline in Q4 2025 was not an isolated company issue. The 41% correction from the mid-September high reflected a significant decline in investor confidence in its prospects and OpenAI's development. Looking back, it's clear that the cooperation between Oracle and OpenAI has always been tinged with a 'two-way game' dynamic.

Superficially, it's a win-win scenario of mutual binding: Oracle provides large-scale compute power to support OpenAI's model training and commercial落地, while OpenAI, with its $300 billion long-term compute order and 60% share of RPO, becomes Oracle's largest customer, endorsing its AI transformation and serving as the core支撑 for its阶段性 stock price increase. Although Ellison has repeatedly stated in public forums like earnings calls and Oracle AI World that the gigawatt-scale data centers custom-built for OpenAI lead global peers in AI model training efficiency and cost control—which was a key reason for OpenAI's initial deep partnership—disagreements beneath the surface of cooperation had already been quietly growing due to mismatches in compute supply and demand.

OpenAI's compute consumption for model training far exceeded initial plans. Meanwhile, Oracle, affected by factors like NVIDIA's chip allocation and delays in supporting data center infrastructure, saw delivery timelines for most "Stargate"-related sites slip, except for the flagship Abilene, Texas site, which progressed on schedule. The overall pace of compute supply consistently failed to fully match OpenAI's demand growth rate. In October 2025, The Information reported that while securing the agreement with Oracle, OpenAI had also entered into compute supply合作 with Microsoft, CoreWeave, and Google. Although OpenAI significantly increased its commitment to Oracle, this 'multi-sourcing' strategy, while符合算力安全 needs, also indicated a tendency to maintain a 'flexibly adjustable' attitude towards the partnership with Oracle.

Faced with rumors, OpenAI only issued a brief statement saying its合作 with Oracle "remains stable" but avoided discussing details. This vague response directly weakened market trust in the stability of Oracle's core orders. Subtly, during Oracle's entire Q2 FY2026 earnings call, when addressing OCI business and RPO, management mentioned OpenAI only twice. Instead, they highlighted合作 agreements with other higher-quality clients like Meta and NVIDIA, emphasizing the extremely low cost of switching to other AI suppliers. This implied that if OpenAI ultimately failed to fulfill the agreement, Oracle could at least partially replace its orders.

But for Oracle, a decline in the share of core orders not only directly impacts short-term revenue expectations but also erodes the 'compute scarcity' advantage it built through the OpenAI partnership, significantly weakening investor confidence in the long-term competitiveness of its AI business.

2. **Fulfillment Doubts: 'Reassurance' Turns into 'Burden'** For tech giants, large RPO has always been a 'ballast' for market confidence. Yet, when Oracle announced in its Q2 FY2026 report that RPO had increased by $67.7 billion, reaching a new high of $523 billion by quarter-end, the market responded with a sharp stock price drop. Even with AI demand remaining strong, the catalyst of new orders failed to work its magic a second time.

Li Zhe分析, "This reaction suggests that investors may be increasingly losing confidence in Oracle's ability to convert this massive (and still growing) order backlog into a sustainable, profitable revenue stream." So, what is causing investor skepticism about order conversion and free cash flow? Li Zhe told Leiphone, "The primary reason is the disparity between the 'paper wealth' of Oracle's RPO and the low efficiency of order conversion. Of Oracle's $523 billion RPO, about 60% comes from OpenAI. Meanwhile, only about 10% of the RPO is expected to convert into recognized revenue within the next 12 months; the rest are long-term orders beyond 2027. This means the cash realization cycle for these huge orders spans 3-5 years."

According to J.P. Morgan calculations, influenced by the high-interest-rate environment, the present value discount rate for Oracle's long-term RPO has increased by 20 percentage points compared to 2024. The order储备 once seen as a 'gold mine' has seen its implied value significantly shrink. Li Zhe emphasized, "What alarms the market even more is that the 'cash-burning' investment in AI infrastructure has fundamentally altered Oracle's cost structure and profit logic. Based on Morgan Stanley's estimates, the construction cost for 1GW of AI computing infrastructure is as high as $35 billion. To fulfill its order commitments, Oracle plans to build over 10GW of AI computing capacity by fiscal year 2030. This implies cumulative capital expenditures of $301-$356 billion between FY2026 and FY2030, far exceeding the market consensus of $189 billion."

With free cash flow at -$10 billion this quarter, Oracle needs to cover its operational and investment shortfall through capital market financing. In the current credit environment, large-scale financing will face rising cost pressures, potentially forcing cuts to stock buybacks or dividend payments to ensure liquidity.

"Uncertainty in customer demand further exacerbates concerns about the fulfillment prospects of Oracle's RPO. The market is now starting to worry about the sustainability of demand from core clients like OpenAI. If the adoption of generative AI industry-wide falls short of expectations, these large RPOs could become 'invalid orders,' while the infrastructure costs already incurred and the resulting lease and debt commitments would become a heavy burden," Li Zhe补充ed. "Oracle now holds a huge number of orders it needs to fulfill to meet capital market expectations. But the stock price and orders are two different things. It's easy for traders to get carried away and project far into the future. Even if all orders are fulfilled, the stock price may not necessarily return to $300 unless there's another explosive stimulus."

Now, the market's pricing logic has shifted from 'chasing order size' to 'questioning fulfillment probability.' When every order落地 comes with huge costs and unknown risks, its transformation into a 'burden' becomes inevitable.

3. **Financial Health Exposed: Red Flags Appear** If fulfillment issues are Oracle's 'superficial wounds,' then the continuously deteriorating financial condition is the 'internal injury' threatening its foundation. Beyond the potential fulfillment risks associated with OpenAI, the low margins on AI services, coupled with significantly increased debt and leverage to build AI infrastructure, continue to erode market confidence in Oracle's valuation.

As of Q2 FY2026, Oracle's total debt reached $108 billion. Cash and equivalents stood at $20 billion, significantly increased quarter-over-quarter, primarily due to proceeds from the sale of Ampere equity and recent debt issuance. The net leverage ratio was approximately 3.32x. Although management outlined an increasingly complex, debt-dependent financing structure for AI infrastructure construction in the earnings report, it was吊诡 that they did not mention any equity-based risk-sharing methods, further加剧ing market concerns about balance sheet flexibility.

Against this backdrop, management承诺 that AI data center gross margins would reach 35%-40% in the future while 'minimizing' capital costs. They attempted to address credit investor concerns by mentioning methods like 'Bring Your Own Chip' and chip租赁 agreements to control upfront capital expenditures. However, given the current situation, these claims appear untenable.

Zhang Xu, a cloud business负责人 at a certain company, told Leiphone that for Oracle's data center business to achieve 35%-40% gross margins, one of the following conditions must be met: First, achieve极致 economies of scale. Even with the current $532 billion RPO, realizing over $250 billion in revenue might only push margins into the low 20%s, requiring further scale expansion. Second, NVIDIA offers concessions. NVIDIA's current gross margin guidance of 75% squeezes profit margins along Oracle's supply chain. Third, Oracle develops its own chips to reduce hardware costs or encourages customers to bring cheaper chips, thereby improving its own margins.

"Obviously, for Oracle, which follows a 'chip-neutral' strategy and holds little bargaining power against powerful partners like NVIDIA, achieving any of these three conditions is difficult," Zhang Xu stated. Therefore, although Oracle plans to alleviate concerns about future capital intensity by increasing financing and improving margins, the effectiveness of these measures remains subject to careful scrutiny due to uncertainties surrounding order conversion, associated margins, and financing needs.

**Part 3: Can Oracle Still Break Through?** After experiencing the暴涨暴跌, Oracle, all-in on AI, stands once again at a crossroads. For Oracle, this transformation is not a simple business upgrade but a 'life-or-death' situation concerning the company's very foundation.

UBS suggested in its latest report that market confidence might recover if investors gained more confidence in the capacity release of the Abilene data center, if various lenders maintained stable willingness to finance AI construction, if Oracle disclosed financing needs more transparently, and if it introduced off-balance-sheet, light-capital financing solutions to ease credit pressure. For Oracle today, financing needs and data center construction progress have become critical variables determining the outcome of this transformation battle.

1. **How to Finance?** Current market concern over Oracle's credit risk is high. S&P rates Oracle BBB (the lowest investment-grade level), and both S&P and Moody's have a negative outlook. This is a dangerous signal for Oracle, which has urgent financing needs: it is only two downgrades away from losing its investment-grade rating, which would substantially increase its cost of capital.

During the Q2 FY2026 earnings call, Oracle provided incomplete disclosure regarding financing needs, stating that the required financing for AI infrastructure construction would be less than $100 billion. UBS分析ed that financing needs for the next few years would be slightly over $80 billion before free cash flow turns positive in FY2029. Oracle is actively exploring various off-balance-sheet financing solutions to transfer the burden to third parties, including: Firstly, supplier financing (from NVIDIA and AMD). Secondly, private credit firms purchasing NVIDIA GPUs and entering into sale-leaseback agreements with Oracle. Thirdly, 'Bring Your Own Chip' deals—where large clients like OpenAI, through their own agreements with chip suppliers like NVIDIA and AMD, fund the GPU purchases themselves.

UBS analyst Karl Keirstead noted that such transactions are not pioneered by Oracle; companies like Meta and xAI have signed similar agreements. If 50% of financing needs are met through such deals over the next three years, direct financing needs could drop to $40 billion, but this might lead to lower margins for the cloud infrastructure business.

Beyond these three methods, Oracle has a last-resort option—issuing more equity. Given Larry Ellison's long history of share buybacks, he is极不希望 to dilute his own equity. Therefore, based on Oracle's public statements, it is more likely to prefer solving financing needs through the aforementioned three methods. Li Zhe believes that the second wave of decline has pushed the AI bubble issue to its peak. The fact that Oracle can still raise money afterward suggests that investors might see confidence points not visible from outside the industry. "Oracle's current business model is akin to long-term apartment rentals. As long as rental prices don't fall and customer demand remains high, the model works. Otherwise, it's very risky. Currently, AI supply and demand are imbalanced, with demand exceeding supply. So, fundamentally, this business model doesn't carry特别大的 risk. If the rental prices for compute centers increase, the profit margin could even be higher than imagined."

2. **Can Data Centers Be Delivered Smoothly?** Rumors of data center delivery delays have been a persistent cloud over Oracle. Among them, the "Stargate 1" project in Abilene, Texas, is the core engine driving the rapid growth of Oracle's OCI business. In December 2025, Bloomberg reported that the delivery timeline for some data centers Oracle is building for OpenAI had been pushed from 2027 to 2028, affecting sites including Abilene. During the Q2 FY2026 earnings call, Oracle responded: The supercluster construction in Abilene is progressing on schedule, with over 96,000 NVIDIA GB200 chips already delivered. This suggests the Abilene project may become fully operational in the summer or fall of 2026.

Wang Xu认为, "The current market分歧 on Oracle is essentially a分歧 in judging the investment value of AI infrastructure. From an industry perspective, the core competitive barriers in the AI infrastructure field for the next 1-3 years will still be the ability to secure financing, acquire land, and rapidly construct and network hardware facilities. Oracle currently faces no fundamental困境; the core issue is that data center construction requires time, leading to slow revenue generation."

UBS分析ed in its report that if OpenAI plans a major AI model upgrade in Q1 2026, it would highly likely rely on the larger-scale NVIDIA GB200/GB300 training compute provided by Oracle. The timeline OpenAI gave for a potential "GPT-6" launch in Q1 2026 was likely based on a significant near-term release of training capacity from the Abilene project. If the Abilene project faces major delays, OpenAI would unlikely provide such a timeline.

Beyond Abilene, the market is also concerned about potential delays at other campuses Oracle is leasing. In mid-December, Oracle's data center partner Vantage announced it would break ground on a 1.4GW Stargate "Frontier" campus in Texas, scheduled for operation in the second half of 2026. Recently, DTE Energy received regulatory approval to provide the necessary power for a 1.4GW data center Oracle is building for OpenAI near Ann Arbor, Michigan. While Blue Owl withdrew from financing this campus, Blackstone is reportedly in talks to take over the financing. Additionally, Oracle recently announced breaking ground on an AI data center ("Lighthouse") in Wisconsin, built by partner Vantage, expected for completion in 2028. In summary, there is currently no confirmed news indicating delays in the delivery of Oracle's data center-related合作.

**Part 4: The AGI Era - Why Oracle Struggles to Be the House** "The reason for Oracle's extreme stock price volatility is simple: it lacks a core, foundational advantage. In the industry chain division of labor and landscape of the AGI era, it remains merely a 'middleman' integrating resources," investor Han Shuo stated bluntly in a conversation with Leiphone. When the tide goes out, you see who's swimming naked. Oracle's暴涨暴跌 revealed an uncomfortable truth: those who can 'be the house' in the AGI era are inevitably players who control core segments of the industry chain and possess fundamental barriers. Oracle's multiple shortcomings, from its industry chain positioning to its strategic choices, determine its inability to break out of the 'player' role.

1. **Passivity Up and Down the Chain: Reduced to a 'Supporting Role'** A harsh reality for Oracle is that even by going all-in on AI, it has never truly occupied a core position in the AI industry chain. It survives in the缝隙 between upstream hardware vendors and midstream large model companies, able only to earn thin margins through resource integration, becoming a 'supporting service provider' to the house players.

From the upstream hardware perspective, Oracle has abandoned the path of self-developing chips, possesses no core R&D capability, and is deeply trapped in a 'bottleneck' situation. The core bottleneck for AGI development is compute power, and the key to compute power is high-end GPU chips. In this field, NVIDIA has become the undisputed 'chain master,' its 75% gross margins stemming from irreplaceable technology architecture and ecosystem barriers—this dictates that its合作 with Oracle is about ecosystem synergy, not profit sharing. Han Shuo认为, "In the AI chain, Oracle is always a secondary player. Even if Oracle's data centers fail in the future, as long as NVIDIA remains profitable, other branded manufacturers will step in to build these data centers. Oracle currently lacks not only money but also power. But these aren't problems for NVIDIA, which holds AI hegemony."

Facing the midstream model layer, Oracle is even less able to define compute demand. Large models are the technical foundation and key springboard for realizing AGI's core value. Currently, leading model developers like OpenAI and Anthropic,凭借 their advantageous positions, dictate the standards for compute demand. Oracle has never ventured into large model development, lacking both the relevant technical teams and data resource accumulation. Its core client OpenAI, representing 60% of RPO, views it merely as 'one of the compute suppliers.' Han Shuo认为, "The ultimate goal of the triangular格局 between Oracle, OpenAI, and NVIDIA is to tie more players onto the ship. Now that the market is rationally回归, the focus will ultimately return to NVIDIA and OpenAI. Once OpenAI goes public, Oracle, stuck in the middle, will see its bargaining power further weakened."

Lacking both chip R&D capability and large model technology积淀, Oracle is注定 to have no voice either upstream or downstream in the AI chain—it must heed NVIDIA's stance while being treated as a dispensable backup by OpenAI. Its so-called infrastructure布局 is essentially a supporting game where it remains replaceable at any time.

2. **Strategic Path Dependence: Difficulty Breaking the House's Barriers** "Even by going all-in on AI, Oracle's strategy从一开始就注定ed it can only dance in shackles," Han Shuo认为. The shackles refer not only to fierce external competition but also to the基因烙印 ingrained in Oracle's bones—as a giant that dominated for years through databases and enterprise software, its moat was never technological R&D and innovation, but rather formidable sales channels and decades of accumulated government and enterprise resources. This 'sales-oriented'基因 directly determined that its AI transformation would also take the easiest, most稳妥 'shortcut.'

Traditional software giants transforming into AI最容易 fall into the trap of 'using old methods to navigate new tracks,' and Oracle is a classic example. This path dependence is not unique to Oracle, but in the AGI era, it becomes a fatal flaw. Avoiding hard challenges like chips and large models to focus only on supporting services like AI infrastructure注定s it cannot become the rule-setting house.

In contrast, the players truly站稳ing their ground in the AGI era did not succeed by taking 'shortcuts.' NVIDIA has persevered for decades in chip technology, monopolizing the AI GPU market through its own efforts, firmly grasping the initiative in compute power, and establishing industry supply rules with its technological barriers. Google布局ed its AI transformation early, attacking on multiple fronts. It developed its own TPU chips to fortify compute infrastructure while deeply cultivating the Gemini large model, linking it with Google Cloud to build an AI ecosystem, constructing a solid barrier through a full-stack technological闭环.

For Oracle, perhaps its困境 is not bad luck. Its AI transformation本质上 involves trading short-term financial pressure for long-term influence in the赛道. For this all-in player, there is no middle ground in this gamble: either achieve a 'second rise'凭借 its AI infrastructure business, or succumb to passive adjustment under the pressure of debt and fulfillment obligations. The final outcome of this豪赌 will be written by every milestone of capacity落地, the cost of every financing round, and the fulfillment status of every order.