AI Is Not Just a Bubble, But It Will Rewrite Work and Profit
The bubble may be in the price, not necessarily in the technology
The word bubble often forces a false binary. Either AI is a fraud, or every AI investment is justified.
Reality is more uncomfortable.
Oracle's operating numbers are not weak. Public reports say its fiscal fourth-quarter revenue reached about 19.2 billion, up 21% year over year. Oracle Cloud Infrastructure revenue reached about 5.8 billion, up 93%. Remaining performance obligations, or RPO, reached $638 billion.
RPO is useful, but it is not cash in the bank. It represents contracted obligations that have not yet fully turned into recognized revenue, profit, and free cash flow.
That is where the market's anxiety sits. AI cloud infrastructure is not the same business as classic enterprise software. Traditional software licenses can produce attractive incremental economics. AI infrastructure requires data centers, GPUs, power, networking, financing, depreciation, and execution discipline before the economic return is visible.
MarketWatch reported that Oracle's fiscal 2026 capital expenditure was about 55.66 billion, with negative free cash flow of about 23.69 billion. It also reported that fiscal 2027 capital expenditure could rise to 90 billion to 95 billion.
From a board perspective, this is not only a technology question. It is a balance sheet question.
You can believe that AI will transform industries and still ask: when does the investment pay back, how concentrated is the customer base, and what happens to pricing when more compute capacity comes online?
AI will change business models, but not evenly
My view is that AI will not simply replace the software industry. It will split the software profit pool into several layers.
The first layer is infrastructure. It requires enormous capital. Winners may sign very large contracts, but they also inherit heavy-asset risk. Oracle is under pressure because the market is increasingly viewing it not only as an enterprise software company, but also as an AI infrastructure company.
The second layer is models and platforms. The real question is not who can tell the best story about AI. It is who can embed model capability into stable enterprise workflows.
The third layer is industry applications. The value may not sit in general-purpose chat. It may sit in closing the books faster, detecting supply-chain exceptions earlier, improving customer-response time, reducing engineering rework, or making risk reviews less manual.
For most companies, the first-principles question is not "Can we use a large language model?" It is "Which business decision cycle becomes shorter?"
If AI only gives employees another chat window, it is not strategic. If it allows sales, inventory, delivery, and cash collection to be discussed with the same data language, it has entered the business model.
That also means software pricing will change. Customers will become less willing to pay only for seats, modules, and licenses. They will ask for measurable outcomes: fewer manual hours, shorter cycle time, lower error rates, faster conversion, better risk control.
Software companies will move from selling tools to selling capacity, automation outcomes, and risk reduction.
That path is real, but it is not easy. Once a software vendor promises outcomes, it has to touch the customer's process, data, permissions, and organizational accountability. That is exactly where many AI programs become difficult.
AI replaces tasks before it replaces job titles
I do not like the simplistic claim that AI will eliminate this or that job title.
A job is a container of tasks. AI usually attacks the task before it attacks the title.
A finance role may include voucher preparation, exception checking, report explanation, budget discussion, and business advice. The first two are easier to automate. The last three depend more on business context and cross-functional communication.
A software engineer does not only write code. The role includes understanding business intent, decomposing requirements, judging technical debt, designing boundaries, handling production incidents, and communicating with product and operations teams. AI changes coding deeply. It does not remove the need for people who can take responsibility for system outcomes.
IMF-related analysis and public comments suggest that AI may affect about 40% of jobs globally and about 60% in advanced economies. The key word is affect, not eliminate. Some jobs will be augmented. Some will be reorganized. Some entry-level tasks will shrink.
The most fragile area is the entry-level pathway.
Many people learn an industry through basic tasks: preparing materials, drafting first versions, testing, running reports, handling first-line customer questions. These tasks may look low value, but they are often how people acquire context.
If AI removes too many entry tasks without creating new learning pathways, companies may gain short-term efficiency and lose long-term capability.
A responsible AI transformation has to redesign tasks, redesign learning paths, and redesign how productivity gains are shared.
The work-free utopia will not arrive automatically
The idea that AI will let everyone stay home and rest is emotionally attractive. I do not think technology alone will deliver it.
Higher productivity only means society can produce more with less labor. It does not decide who receives the gain.
If most of the upside goes to capital owners, platforms, and a small group of highly skilled workers, many people will not experience utopia. They will experience insecurity, wage pressure, and self-funded retraining.
The outcome changes only if institutions and companies change with the technology: stronger retraining, more flexible working time, better social protection, clearer algorithmic accountability, and more reasonable sharing of productivity gains.
Inside companies, the same principle applies. AI-driven efficiency should not only become lower expense in a spreadsheet. It should also become better processes, better customer experience, stronger employees, and more resilient organizations.
If a company uses AI only to cut headcount without redesigning work, it may gain short-term margin but lose long-term competitiveness.
My executive judgment
I would not look at Oracle's stock pullback and conclude that AI is over.
I also would not look at strong AI demand and conclude that every AI investment is justified.
The better conclusion is more layered: the technology revolution is real, some valuations are fragile, business models are being rewritten, and labor institutions are not ready.
Executives should avoid swinging between hype and panic. They should return to a few practical questions.
Which business cycle does AI shorten?
Does it change revenue, cost, risk, or only presentation quality?
Which tasks are being replaced, and which human capabilities become more valuable?
Will part of the released time and profit be reinvested into workforce capability?
The future will not become a utopia by default. AI is a stress test: for capital markets, for enterprise operating models, and for society's ability to distribute productivity gains beyond a small group.
The future I hope for is not one where people passively withdraw from work. It is one where fewer people are trapped in repetitive, low-value, exhausting work.
That future will not be waited into existence. It has to be designed.