Oracle cuts 20,000 jobs All in AI: AI competitions have turned into a money-burning race. Do small and medium players still have a chance?

Written by: Web4 Research Center

In the early morning of March 31, when Oracle employees in multiple countries around the world opened their inboxes, they found a blunt email: “After careful consideration of Oracle’s current business needs, we have decided to eliminate your position in our organizational restructuring. Today is your last day of work…”

No prior communication, no HR interview. After the email was sent, system access permissions were shut off instantly, and unassigned restricted stock was immediately voided. According to investment bank TDCowen’s estimates, this round of layoffs would affect 20,000 to 30,000 employees, about 18% of its total global workforce of 162,000.

This is the largest-scale layoff in Oracle’s history.

However, just days before this email was sent, the company had delivered a quarterly report that most public companies envy. In the second quarter for the period ended November 30, 2025, Oracle’s GAAP net profit was as high as $6.1 billion, and earnings per share grew 91% year over year. The performance was excellent—yet the company suddenly cut nearly one-fifth of its headcount. So what exactly happened?

The answer surfaced quickly: according to TD Cowen’s analysis, the total planned investment for the expansion could reach as much as $156 billion—close to one trillion RMB.

To raise this massive amount, Oracle has already raised $45 billion to $50 billion through debt and equity financing for building Oracle Cloud infrastructure. Within two months, it added $58 billion in new debt. And the $8 billion to $10 billion in cash flow that this round is expected to free up is nothing more than “spare change” in this huge bill.

A company earning several hundred billion a year still needs layoffs to “raise money”—the entry ticket for the AI race. How high has it already gotten?

1. Conflicting Signals:

Why is Oracle laying people off while it’s still making money?

These layoffs were not without signs. As reported by Bloomberg on March 5, Oracle at the time was planning multi-department layoffs involving “thousands of people,” and some roles were believed to be replaced by AI. Even so, when the large-scale layoffs were actually implemented, it still surprised the market.

The hardest-hit areas in this round of layoffs included the revenue and health sciences organization (about 30% layoffs), the SaaS and virtual operating services organization (also about 30% layoffs), and a substantial reduction at NetSuite’s India development center. Based on disclosures in Oracle’s 10-Q quarterly report filed in March 2026, the company set aside a $2.1 billion restructuring budget for this.

However, Oracle’s earnings report numbers present a different picture. Total revenue in the second quarter was $16.1 billion, up 14% year over year. Revenue from cloud infrastructure (IaaS) was up as much as 68% year over year. Remaining performance obligations—revenue already contracted but not yet recognized—jumped more than fivefold from a year earlier to reach $523 billion.

Revenue is rising, profits are rising, and its cloud business is growing at nearly a 70% annual pace—yet Oracle still needs to lay off nearly 20,000 people. That sounds like a paradox.

But there’s one detail in the earnings report that explains it all: free cash flow in the second quarter turned unusually negative, at -$10 billion. Oracle raised its full-year capital expenditure guidance significantly to $50 billion, about $15 billion more than the initial expectation.

That’s the answer. Profits are up, but cash flow is negative; orders are surging, but all the money is being poured into data centers.

In fact, the same script is playing out across Silicon Valley. Since 2025, Amazon has cut around 30,000 enterprise roles, Meta restarted layoffs in March 2026, and the entire tech industry is doing two things at the same time—making massive investments in AI while cutting back heavily on non-core businesses. The reason is simple: AI infrastructure is burning through money so fast that even the most profitable tech giants have no choice but to make trade-offs between human resources and compute power.

Oracle chose the latter.

2. The AI race has changed—moving from “competing on technology” to “competing on capital”

If the AI race from three years ago was still a contest of algorithms and model architectures, then the AI race in 2026 has turned into a full-blown capital game.

The training costs for large models have become shockingly high. Industry estimates suggest that training a GPT-5 level model has already surpassed $1 billion. Building data centers is also astronomical—building your own compute centers requires investing billions of dollars in land, electricity, liquid cooling systems, and network bandwidth. But the even bigger cost is operations: for every additional user an AI application gets, behind it are real electricity bills and inference expenses.

Given this cost structure, tech giants’ response has been strikingly consistent: spend money—then spend more—keep spending.

According to a report by market research firm Futurum Group, the five largest U.S. cloud and AI infrastructure providers—Microsoft, Google, Amazon, Meta, and Oracle—will have total capital expenditures of $66 billion to $69 billion in 2026. Compared with roughly $38 billion in 2025, that’s nearly double.

Looking at it in detail, the arms race among these giants has reached an eye-popping level.

Amazon leads the pack with a $200 billion capital expenditure plan. That figure even exceeds the most optimistic analyst expectations—previously, the market had estimated about $147 billion. Amazon CEO Andy Jassy argues that AI capacity is being absorbed by the market at a “install-and-monetize” pace, and AWS’s annualized revenue has accelerated to $142 billion. Even so, Amazon’s stock price still fell by roughly 8% to 10% after the announcement, and investors are concerned about the timeline for this investment return.

Google’s parent company Alphabet follows closely, with expected capital expenditures of $175 billion to $185 billion in 2026—nearly double the $91.4 billion in 2025. The number itself is enough to make the point: a company’s annual hardware spending has already surpassed the GDP of the vast majority of countries.

Meta plans to invest $115 billion to $135 billion in capital expenditures in 2026, mainly for building the “Meta Super Intelligences Laboratory” and expanding data center capacity. Meta’s capital expenditures in 2025 were $72.2 billion, and this increase is downright aggressive.

Microsoft is moving forward at a pace of $37.5 billion in capital expenditures per quarter, with its 2026 fiscal year expected to invest $120 billion or even more.

Oracle’s posture in this round of competition is equally aggressive. According to people familiar with the company, one major reason Oracle has launched this round of layoffs is insufficient returns on AI investment, with technology and financial services becoming the hardest-hit areas. Worth noting is that TD Cowen’s analysts, in a report published earlier this year, estimated that if Oracle laid off 20,000 to 30,000 people, it could generate an additional $8 billion to $10 billion in free cash flow.——This means Oracle, from the very beginning, treated layoffs as one component of a broader funding plan for AI infrastructure.

What do these five companies’ combined annual capital expenditures of nearly $700 billion mean? For comparison: it’s higher than the GDP of the entire country of Israel, and also exceeds the total revenue of all cloud infrastructure services worldwide.

The AI race has split into two completely different tracks.

Track A is the infrastructure and foundation model layer—requiring capital on the order of trillions, tens-of-thousands GPU clusters, and globally deployed data centers. This is the battlefield for the giants, a “nuclear deterrence-level” capital game. Track B is the application and use-case layer—lower barriers, and it tests understanding of vertical industries and insights into niche scenarios.

Can smaller and mid-sized players squeeze into Track A? Almost impossible.

3. Lessons for Web4 founders:

Don’t compete with the giants on who has more money, or who understands the scenario better

For entrepreneurs in the AI and Web4 space, the signal from Oracle’s layoffs couldn’t be clearer:

Don’t build “power plants” for the AI era. Build “appliance companies” for the AI era.

Power plants are the business of the giants—investments at the trillion-level, driven by economies of scale, and backed by global deployment. Appliance companies, on the other hand, leverage the already-built “power grid” to create great products that solve specific problems. History has repeatedly validated this pattern: after each revolutionary leap in technological infrastructure, the greatest value ultimately isn’t created by the infrastructure builders, but by entrepreneurs who build applications and scenarios on top of it.

The current AI infrastructure frenzy has an astonishingly similar underlying logic to the earlier internet bubble, the mobile internet wave, and the cloud computing revolution. In the internet wave, the giants and nation-level capital built the backbone networks and submarine fiber cables, but it was the companies that built e-commerce, social platforms, and search on top of that network that truly changed the world. In the mobile internet era, telecom operators and communications giants deployed base stations and 4G networks, but what created trillion-dollar market caps were the startups building application scenarios on top of smartphones.

The intersection of AI and Web4 is precisely the blind spot that giants currently can’t see, can’t understand, and for now can’t do.

The business logic of giants determines that they pursue “general AI capabilities”—a single model that solves all problems. By nature, this logic rejects highly customized, low-frequency, non-standard vertical scenarios. And in those ignored gaps, startup companies find their biggest opportunities.

So which specific directions are worth paying attention to?

The first direction is AI Agents and on-chain automation. The smart contract ecosystem naturally requires features like automated execution strategies, on-chain audits, liquidity management, and more. Most automation solutions in today’s Web3 ecosystem still stop at simple scheduled jobs or trigger-script layers, lacking true intelligent decision-making. Giants’ micro-level understanding of complex on-chain logic and their engineering capabilities are far behind startup teams that have deeply focused on this domain. Using AI agents to provide intelligent automation services for areas like DeFi strategies, DAO governance, and on-chain security is a classic “niche market that giants don’t care about.”

The second direction is combining private data with AI inference. Users own their data. AI models can provide services through zero-knowledge proofs or federated learning, completing inference tasks without touching raw data. This model has enormous application potential in fields with extremely high data privacy requirements, such as healthcare, finance, and law. While giants have strong model capabilities, their business model fundamentally depends on data collection and centralized processing, which creates a structural contradiction when it comes to truly respecting data ownership. Web4 entrepreneurs can leverage blockchain’s trust layer advantage to build a moat in this gap.

The third direction is AI Copilots for vertical industries. Web3 game asset valuation, NFT liquidity forecasting, cross-chain asset orchestration optimization, on-chain identity credit assessment… these scenarios are narrow and highly vertical enough that giants have no incentive to invest resources to build them. But each such niche scenario could support a small-but-beautiful startup. The key is to truly understand the core pain point of the scenario, rather than vaguely building a “general Web3 assistant.”

A simple but effective decision framework can help founders evaluate whether their project is safe: if your core competitive advantage is compute power or model parameters, you will eventually be crushed by the giants. If your core competitive advantage is deep understanding of industry knowledge, user relationships, or on-chain data, then when giants enter this domain, it will actually validate your direction—because what giants need isn’t your technology, but your understanding of the scenario.

At the end of the day, AI won’t eliminate you, but competitors using AI will. And in the world of Web4, that competitor is often not the giant—it’s another small team that understands the scenario even better than you do.

Conclusion

The moment Oracle’s layoff email was sent, the 20,000 to 30,000 employees didn’t just lose a job—they lost a metaphor for an era.

The ticket to the AI race has risen to a level ordinary people can hardly imagine. But beyond the ticket, there’s another path.

Power plants belong to the giants; appliance companies belong to startups. Some of the roles that were cut will indeed be replaced by AI forever, but more opportunities are growing out of those gaps.

AI is not an opportunity for everyone, but it always belongs to the people who find the right position.

( This article is for industry analysis only and does not constitute any investment advice. )