The email arrived at six in the morning, local time, which meant six in the morning in Redwood City and six in the morning in Bangalore and six in the morning in Mexico City and six in the morning in Montevideo. It was signed “Oracle Leadership”. Not by a person. Not a manager, not a vice president, not anyone with a name and a phone number and a reason. By a noun phrase.
That was 31 March 2026, and according to reporting by The Next Web, employees across the United States, India, Canada, Mexico and Uruguay opened their laptops to find their employment had ended while they slept. Multiple outlets reported no prior warning from HR and none from direct managers, because in many cases the managers had not been told either. Some were reading the same email at the same hour, discovering simultaneously that their teams had been cut by a third and that they themselves were still employed, or were not.
Five months later, on 1 September 2026, Oracle's fiscal second quarter opened. Business Insider had reported in mid-August, citing people familiar with the plans and an internal document, that another round of cuts had been drawn up to land before that date, with managers again asked to submit lists of names and some teams facing double-digit reductions. Oracle declined to comment. It has declined to comment on essentially all of it for months, which is its right and also, in a way, the story.
Somewhere in that silence sits a piece of arithmetic that almost nobody states plainly. A company reduces its headcount. The money that used to be salary becomes cash flow. The cash flow becomes capital expenditure. The capital expenditure becomes data centres, and the data centres run the models the company will subsequently market as the reason fewer people are needed. The worker's compensation is converted, through a series of entirely legal and unremarkable accounting steps, into the infrastructure that justifies their absence.
Nobody Told The Managers Either
Start with what was actually said, and by whom, because the sourcing matters enormously here.
The 20,000 to 30,000 figure did not come from Oracle. It came from TD Cowen. The Register's Lindsay Clark reported on 29 January 2026 that the investment bank had circulated a note estimating Oracle might cut between 20,000 and 30,000 roles and might sell Cerner, the health records business it bought for 28.3 billion dollars in 2022, to fund its data centre commitments. The same note, as reported by both The Register and CIO, calculated that a reduction on that scale would generate 8 to 10 billion dollars in incremental free cash flow, and that Oracle's contract with OpenAI alone implied roughly 156 billion dollars of capital spending.
The note described something less discussed and more revealing: American banks had pulled back from lending to Oracle-linked data centre projects, credit default swap spreads had tripled, and, as the analysts put it, both equity and debt investors had raised questions regarding the company's ability to finance the build-out.
So 30,000 is an analyst's upper bound on a hypothetical, published two months before any cuts began, by a bank modelling how a client under financing stress might close a funding gap. It is not a disclosure, and Oracle has never confirmed it. The widely repeated formulation that the company “eliminated approximately 30,000 positions, roughly 18 per cent of its global workforce” is a media construction built by taking the top of TD Cowen's range and dividing it by a headcount of 162,000. It hardened into fact through repetition, which is how most numbers in this industry do.
What The Filings Say When Nobody Is Quoting Analysts
There is a document that does not speculate, and it is worth more than every anonymous source in this story combined. Oracle's Form 10-K for the fiscal year ended 31 May 2026 states that the company employed approximately 141,000 full-time employees, approximately 49,000 in the United States and approximately 92,000 internationally. The previous year's 10-K puts the figure at approximately 162,000, with 58,000 in the United States and 104,000 internationally.
That is a net decline of 21,000 people, or roughly 13 per cent, in twelve months. Not 30,000. Not 18 per cent.
The distinction cuts both ways. Net headcount understates gross separations, because Oracle was hiring aggressively into its infrastructure organisations across the same period, so the number who actually lost jobs is certainly higher than 21,000. But it is the only audited figure anyone has, and it sits nine thousand below the headline number. Anyone telling you confidently that Oracle sacked thirty thousand people is telling you what an analyst thought might happen, not what a filing says did.
Note where the reduction fell. The United States lost approximately 9,000 roles, international operations approximately 12,000. This was not primarily an American event, which is one reason the American early warning system barely registered it.
The financial context in the same filing is unambiguous. Cash used for capital expenditure increased from 21.2 billion dollars in fiscal 2025 to 55.7 billion in fiscal 2026, an increase Oracle attributes, in its own words, primarily to the expansion of our data centers. The company expects the trend to continue. Restructuring and other expenses rose from 374 million dollars to 1.838 billion, up 391 per cent, consisting primarily of employee severance under what Oracle calls the 2026 Restructuring Plan.
Capital expenditure rose by 34.5 billion dollars. Severance rose by 1.46 billion. Those two lines sit on the same statement, in the same year, at the same company.
Then there is the sentence in the risk factors, which deserves quoting exactly because it is the closest thing to an admission in the entire affair. Oracle writes that “the adoption and deployment of AI technologies across our operations have resulted, and may continue to result, in reductions to our workforce.”
Have resulted. Past tense. In a document signed under penalty of law.
The same paragraph continues, and this is the part that will matter later: such restructurings “may also lead to shortages of sufficiently skilled employees in certain roles, loss of valuable institutional knowledge and damage to employee morale and retention.”
Oracle's lawyers, it turns out, have read the literature on capability erosion. They filed it under risk disclosure rather than acting on it.
Seven Hundred And Ten Notices For Twenty One Thousand Jobs
Here is a detail that should trouble anyone who believes labour market data captures reality. Under the American Worker Adjustment and Retraining Notification Act, employers of a certain size must give sixty days' notice of mass layoffs at a single site. Oracle filed WARN notices with California's Employment Development Department covering approximately 710 employees: around 300 in Redwood City, more than 180 in Santa Clara, over 150 in Pleasanton and 50 in Santa Monica, with separations scheduled for early June. WARNTracker's compilation of Oracle America filings shows 5,332 workers across 33 notices spanning almost a decade.
Seven hundred and ten, against a documented net decline of twenty one thousand.
The gap is not evidence of wrongdoing. WARN thresholds are site-based and headcount-based, most of Oracle's reduction happened outside the United States where the statute does not apply, and rolling reductions across many small offices can be lawful while remaining almost invisible. But the instrument designed to make mass displacement visible to the public captured roughly three per cent of it.
The instruments were built for factory closures. They are being asked to measure something that arrives at six in the morning, in five countries, one inbox at a time.
Twenty Two Per Cent Of Everything
The Challenger, Gray & Christmas job cut report is the closest thing the United States has to a real-time ledger of announced layoffs, and it has tracked artificial intelligence as a distinct stated reason since 2023. Its numbers are worth handling precisely, because they are widely misquoted.
In 2025, according to Challenger's year-end report, American employers announced 1,206,374 job cuts, a 58 per cent increase on the 761,358 announced in 2024. Of those, 54,836 were attributed to AI. Technology accounted for 154,445 cuts, up 15 per cent. Hiring plans fell to 507,647, down 34 per cent and the lowest since 2010.
The thirteen-fold figure comes from comparing the AI-attributed total in 2023, 4,247, with the 2025 total of 54,836. That is a factor of roughly 12.9, it is real, and it is sourced to Challenger's own tracking rather than any academic reconstruction.
But note what the 2025 figure is not. It is not 100,000 tech workers displaced by AI. Layoffs.fyi, the crowd-aggregated tracker, recorded around 122,500 tech layoffs across roughly 257 companies in 2025, down from 152,922 across 551 companies in 2024. Those are tech layoffs from all causes. The AI-attributed 54,836 covers the entire American economy, every sector, and is about four and a half per cent of all announced cuts that year. Conflating the two produces a number roughly twice as alarming as the evidence supports, and the case does not need the inflation.
Because the 2026 data is startling on its own terms. Challenger's July 2026 report, released on 6 August, recorded artificial intelligence cited in 112,713 job cut announcements for the year to date, approximately 24 per cent of all cuts, and the single leading stated reason for five consecutive months. Since tracking began in 2023, AI has been named in 184,538 announcements. Technology led all sectors again with 149,023 cuts through July, up 67 per cent year on year and 31 per cent of everything.
So the honest version of the thirteen-fold claim is this: AI-attributed layoffs went from 4,247 in 2023 to 54,836 in 2025, then more than doubled that annual total within the first seven months of 2026. The trajectory is steeper than the headline.
And then, in the month this piece was being written, the streak broke.
Challenger's August report, released on 4 September, records 52,881 announced cuts for the month, up 58 per cent on July and yet the lowest August total since 2022. Restructuring led all reasons with 16,173 cuts, the highest monthly restructuring figure since January. Artificial intelligence fell to the fourth most cited reason, with 3,462 cuts, its lowest monthly total since December 2025, when the figure was 142. In the report's own words, it ends a five-month run, beginning in March, in which AI was the leading monthly reason.
The annual picture is unchanged. AI has been cited in 116,175 announcements so far in 2026, approximately 22 per cent of all cuts, and it remains the leading reason year to date. Technology still leads every sector with 155,126 cuts through August, up 52 per cent year on year and 29 per cent of the total. Announced cuts for the year stand at 529,914, down 41 per cent on the same period in 2025. Hiring plans for August came to 12,325, the highest August figure since 2022.
One month is not a trend, and nobody should build an argument on it. But it is worth sitting with what a single month can do to a number this widely quoted. Nothing about the technology changed between July and August. What changed was which word employers reached for when asked to explain themselves.
Andy Challenger Says The Quiet Part In His Own Report
Now for the part that complicates everything, delivered by the person with the least incentive to complicate it.
Andy Challenger, chief revenue officer at Challenger, Gray & Christmas, is quoted in his own firm's July 2026 report saying something remarkable about the data his firm produces. “Naming AI in a layoff announcement can win over investors while pushing current and prospective employees away,” he said. “That's why the messaging has swung from hedging to aggressively citing it.”
Read that again. The man who runs the most cited AI layoff tracker in America is telling you the metric moves partly because companies discovered that saying “AI” moves the share price. He added a second warning: “As regulations start to take shape, companies will be even more careful in their announcements, which would make tracking the impact of AI on jobs more opaque.”
The same report includes a candid methodological confession. Challenger writes that “the ambiguous nature of what constitutes AI-attributed cuts was highlighted in July,” and walks through two cases. Visa announced a seven per cent reduction attributed to an efficiency drive in which AI would reshape work, so Challenger categorised those cuts as AI. Montefiore, a hospital system in the Bronx, eliminated twelve utilisation review nursing posts after adopting software from Datavant. The New York State Nurses Association issued a press release describing this as replacing human workers with AI software. Hospital leaders called that characterisation misleading. Datavant markets AI and natural language processing capabilities. Absent clarity, Challenger filed those twelve jobs under “Technological Update (possibly AI)”, a category that absorbed 20,219 cuts in 2025.
The market evidence behind that remark is not subtle. On 26 February 2026, Block cut roughly 4,000 people, taking its workforce from over 10,000 to just under 6,000, and co-founder Jack Dorsey attributed the decision to the intelligence tools the company was building. “Within the next year,” he wrote, “I believe the majority of companies will reach the same conclusion and make similar structural changes.” Block's shares rose as much as 24 per cent on the news.
Twelve nurses, one contested press release, one classification decision. Multiply that judgement call across a year and you understand how much interpretive latitude sits inside a number reported as though it were a thermometer reading.
Challenger's own summary for July is not the line that circulated. “Hiring has also increased over last year by 25%,” he said, “so while AI is shifting the labor market, it is not dismantling it.” That deserves to be taken seriously rather than swatted away.
A month later, with AI knocked off the top of his own list, his emphasis had moved further in the same direction. “Employers are making plans to add workers, with 46% of those plans coming from manufacturing industries,” he said of the August figures. “The questions are how long will it take employers to actually fill these roles and will they find workers with the requisite skills.”
Whether firms can find workers with the requisite skills is not a story about machines replacing people. It is a story about a training pipeline, and it will return.
The Case For Oracle, Made Properly
The strongest version of the corporate argument is better than its critics usually allow.
Oracle is not eliminating jobs because AI can do them. It is eliminating jobs because it needs cash, urgently, to fund an infrastructure commitment on which its entire future depends, and labour is the largest discretionary line item available. This is capital reallocation, not technological substitution. Calling it an AI layoff is a category error committed by journalists who like a tidy narrative.
The financial pressure is real and documented. Oracle's free cash flow deficit for fiscal 2026 ran to roughly 23.7 billion dollars. The company raised around 43 billion in debt plus five billion in stock sales, and anticipates roughly 40 billion more in fiscal 2027. On 9 July 2026, S&P downgraded Oracle to BBB-, one notch above junk, naming OpenAI as a key credit risk against a backlog of 638 billion dollars in remaining performance obligations, roughly half tied to that single unprofitable counterparty. Moody's attached a negative outlook. The shares fell close to half from their September 2025 peak.
The first quarter of fiscal 2027, reported on 10 September, relieved none of that pressure and illustrated a great deal of it. Revenue rose 30 per cent to 19.3 billion dollars, cloud infrastructure revenue 121 per cent to 7.4 billion, operating income 57 per cent and net income 63 per cent. Capital expenditure for the three months was 28.5 billion dollars against 8.5 billion a year earlier. Free cash flow was negative 5.4 billion. Remaining performance obligations reached 664 billion, up 209 billion year on year and 26 billion above the 638 billion figure S&P had in front of it in July, with more than 30 billion dollars of new AI contracts signed in the quarter. Chief financial officer Hilary Maxson reaffirmed capital expenditure guidance of 90 to 95 billion dollars for the year and told analysts the company now expects around half of that backlog to convert into sales over the next 36 months. During the quarter Oracle sold 20 billion dollars of its own stock through an at-the-market programme, exhausting the facility in full.
A company posting record profits and record backlog while burning cash at that rate is not economising on salaries because it has run out of work for people to do. It is economising on salaries because the arithmetic of the build-out leaves nothing on the table.
A company in that position that did not cut costs would be behaving irresponsibly towards its shareholders, its bondholders and, arguably, the employees who remain.
There is a second, better argument, and Sam Altman of all people has made it. Speaking around the India AI Impact Summit in early 2026, the OpenAI chief executive conceded that “there's some AI washing where people are blaming AI for layoffs that they would otherwise do”, while maintaining that genuine AI displacement is also occurring. When the man selling the technology says companies are over-attributing job cuts to it, that is not a talking point. It is testimony against interest.
The empirical support is stronger still, and it comes from an institution that asks the same firms the same questions every August. Between the 2024 and 2025 rounds of the Federal Reserve Bank of New York's regional business surveys, recapped on Liberty Street Economics in a post introduced by research director Kartik B. Athreya, AI use among service firms in the New York and northern New Jersey region rose from 25 to 40 per cent, and among manufacturers from 16 to 26 per cent. In the 2025 round, only one per cent of service firms reported letting workers go in response to AI over the preceding six months, and no manufacturers reported any at all.
The 2026 round was published on 1 September by Jaison R. Abel, Richard Deitz, Natalia Emanuel and Nick Montalbano under a title that gives away the finding: “Businesses Are Using AI to Transform Work, Not Cut Jobs”. Four per cent of service firms reported laying off workers in response to AI over the past six months, compared with just one per cent in the previous year's survey. No manufacturers reported layoffs this year or last. About thirteen per cent of service firms said they had hired more workers because of AI, and no manufacturers increased hiring because of it. Just over a third of service firms and more than twenty per cent of manufacturers report retraining staff in response to AI. The Fed's own summary is that existing workers are much more likely to be retrained than replaced.
A quadrupling is a quadrupling and should be written down as one. But four per cent of service firms is four in every hundred, and thirteen per cent hiring more people because of AI is three times that. Whatever is happening inside those firms, it is not a labour market being dismantled from the inside.
And Stanford's Digital Economy Lab, often cited as the smoking gun, is explicit that it is not one. In their August 2026 update, Erik Brynjolfsson, Bharat Chandar and Ruyu Chen state plainly: “We do not see widespread, economy-wide job displacement associated with AI.”
Any argument that AI is currently eating the labour market whole must get past the New York Fed, Challenger's own chief revenue officer, Stanford's leading empiricists and the chief executive of OpenAI. It does not.
Where That Case Runs Out Of Road
But notice what the steelman establishes, and what it quietly concedes. It establishes that AI is not currently causing mass unemployment. It does not establish that the transition is being paid for fairly. Those are different claims, and the second survives everything the first throws at it.
Take the Stanford finding in full. The same paper that finds no economy-wide displacement finds that employment among workers aged 22 to 25 in highly AI-exposed occupations now stands about 19 per cent below where it would be if it had kept pace with similarly aged workers in less-exposed occupations. That gap widened from 15 per cent in July 2025 to 19 per cent by June 2026, measured on ADP payroll data. The declines concentrate precisely where AI usage substitutes for human tasks, and are flat or rising where it complements them. Experienced workers show no comparable gap.
And crucially, the mechanism is not dismissal. Brynjolfsson, Chandar and Chen find the adjustment “appears to operate primarily through reduced hiring of young workers rather than increased separations.”
This is why the New York Fed can find ninety six per cent of service firms reporting no AI layoffs whatsoever while something large is nevertheless happening. Nobody has to be fired for a generation to be excluded. You simply do not open the door. There is no announcement, no WARN notice, no Challenger category, no six in the morning email, because there is no employee. The cost lands on people who never appear in any layoff statistic because they were never hired.
The Fed's own surveys support this reading rather than contradicting it, and the two rounds read together are more instructive than either alone. In 2025, alongside the one per cent reporting past layoffs, thirteen per cent of service firms anticipated AI-driven layoffs in the coming six months, and firms expected to reduce hiring plans, particularly for college-educated workers. A year later the realised figure was four per cent, not thirteen. The dismissals the firms expected to make largely did not arrive. The hiring restraint they expected to exercise is not something the survey was built to count, and it requires no announcement, no severance package and no separation date.
So the AI washing critique is correct and insufficient. Some firms are certainly attaching an AI label to ordinary cost cutting. But the reverse error is larger and less examined: much genuine AI-driven substitution is occurring through hiring restraint, and it produces no data at all.
The Trap That Foresight Cannot Escape
This is where the theory becomes indispensable, because it explains what the empirical data cannot: why firms would keep doing this even if they could see exactly where it leads.
In “The AI Layoff Trap”, Brett Hemenway Falk of the University of Pennsylvania and Gerry Tsoukalas of Boston University build a task-based model of automation and then do something unusual with it. Most such models live in the labour market. Theirs lives in the product market.
The mechanism is elegant and slightly horrifying. Displaced workers are also customers. When a firm automates, it captures the entire cost saving, but the resulting loss of consumer spending spreads across every firm in the sector. Each firm bears only a fraction of the demand destruction it personally causes. The rest lands on its rivals.
Falk and Tsoukalas call this a demand externality, and show that it makes over-automation a strictly dominant strategy. Every firm's profit-maximising automation rate exceeds the collectively optimal level, and this holds under perfect foresight. As the authors frame it, if the cliff ahead is visible to all, why would rational firms race towards it? Their answer is that visibility is irrelevant. The incentive structure does the work regardless of what anyone believes.
The distortion worsens with competition rather than easing. A monopolist internalises the externality because it owns the whole demand pool. Fragmented markets exhibit the widest gap. In the frictionless limit, the game collapses into a Prisoner's Dilemma in which every firm displaces its entire human workforce even though collective restraint would raise all profits. The resulting loss, they specify carefully, is not a transfer from workers to shareholders. It is deadweight. It harms both.
Then comes the finding that should reorganise the policy conversation. Falk and Tsoukalas test six instruments. Upskilling and worker equity participation narrow the wedge but cannot close it. Coasean bargaining fails outright, because automation is a dominant strategy and no voluntary agreement between firms is self-enforcing. Capital income taxes do not alter the equilibrium automation rate at all, because they operate on profit levels rather than the per-task margin where the externality lives. Universal basic income raises the floor on living standards and leaves the automation incentive precisely unchanged.
Only a Pigouvian automation tax, set equal to the uninternalised demand loss per task, implements the cooperative optimum.
There is a further twist they call the Red Queen effect. Higher AI productivity widens the wedge rather than resolving it, because each firm perceives a market-share gain from automating faster than its rivals, and at the symmetric equilibrium those gains cancel, leaving only the added distortion. Better AI makes the trap worse.
Set that model against Oracle and it fits with unpleasant precision. Banks were retreating, competitors circling, and CIO reported OpenAI redirecting near-term capacity needs towards Microsoft and Amazon. The combined 2026 capital expenditure of the four largest hyperscalers is projected between 630 and 725 billion dollars, against roughly 388 to 410 billion in 2025. Oracle is not choosing to cut. It is running the dominant strategy in a game where restraint is unilateral disarmament.
Which is why appeals to corporate conscience are wasted breath, and why the six in the morning email is not a failure of manners. It is the compressed physical form of a structural incentive.
Speed As A Solvent Of Memory
Wolfgang Rohde's “Short-Term Gain, Long-Term Fragility” attacks the problem from a different direction, and supplies the mechanism Oracle's own risk factors gesture at without naming.
Rohde separates two things that get conflated. Capability masking is perceptual and accounting-level: visible AI output creates the appearance that organisational capability has been replaced. Capability erosion is the slower structural consequence: the human systems that actually sustain capability weaken over time.
He sets out five linked steps. AI produces visible, plausible output that appears to substitute for human work. Managers and policymakers infer from that output that the underlying capability has been replaced. Organisations respond through hiring restraint, leaner staffing and deeper dependence on external platforms. Hidden costs accumulate through verification burdens, loss of tacit knowledge and contraction of the apprenticeship pipeline. Finally these firm-level decisions scale into systemic fragility, narrower entry paths and more concentrated power.
His evidence comes mainly from software engineering, chosen not because the mechanism is unique there but because it is unusually measurable. AI-generated code remains uneven in correctness, maintainability and security, and still requires substantial human verification. Repository-level research suggests current systems struggle to use broader codebase context reliably, and long-context work shows performance degrading when relevant information sits in less salient parts of the window. The verification burden does not disappear. It moves onto the smaller number of experienced people still present to carry it.
Rohde reframes it as a balance sheet of deferred obligations: technical debt, capability debt, institutional debt. Capability debt arises when visible output is preserved by weakening the human skill, review capacity and apprenticeship structures needed to sustain reliable work.
He puts it in one line that is worth the price of admission. “When institutions mistake visible output for durable capability, speed becomes a solvent of memory.”
He is scrupulous about what the argument does not require. The mechanism, he writes, needs no bad faith or coordinated intent. It emerges from ordinary incentives, optimistic readings of AI output, and the convenience of accepting visible productivity as proof of real substitution.
Now read Oracle's risk factor again: restructurings may lead to “shortages of sufficiently skilled employees in certain roles, loss of valuable institutional knowledge and damage to employee morale and retention.” That is capability erosion, described accurately, disclosed to investors, and proceeded with anyway. And Stanford's 19 per cent gap for 22 to 25 year olds is the apprenticeship pipeline contracting in real time, on payroll data, exactly as step four predicts.
Two independent literatures, one theoretical and one empirical, arriving at the same place from opposite directions.
Thirty Two Frameworks That Cannot See It
The third paper explains why none of this is anybody's job to prevent. Sivasathivel Kandasamy's “The Human Utility Factor” proposes a computable welfare metric combining Agency, Wellbeing and Economic Stability, operated through three levers: automation depth, redistribution intensity and employment coverage. The construction is technical; the diagnosis underneath is simple and damning.
Kandasamy surveys thirty two AI governance frameworks and observes that every one is designed to govern individual AI systems rather than aggregate macroeconomic effects. The EU AI Act, the leading instrument in the field, contains no quantitative constraint on macro-socioeconomic stability whatsoever. You can be fully compliant with every AI regulation on earth while participating in an economy-wide displacement event, because compliance is assessed at the level of the model and the harm occurs at the level of the labour market.
His empirical section supplies the thirteen-fold figure that has travelled furthest: AI-attributed layoffs rising from 4,247 in 2023 to 54,836 in 2025, with AI overtaking every other stated cause by March 2026. He sets this against firms reporting record revenues while cutting, noting Microsoft shedding more than 15,000 roles in 2025 while posting 70.1 billion dollars in quarterly revenue, and UPS eliminating 48,000 positions under an explicit automation dividend rationale while raising the share of volume through automated facilities from 63 to 66 per cent.
His formulation of the core problem answers the question this article started with. Firms internalise the productivity gains of automation while externalising its displacement costs onto workers, households and public transfer systems.
To his credit, Kandasamy declines to overstate. He acknowledges that AI-attributed layoffs remain a fraction of total job losses, and frames his claim narrowly: society is navigating an accelerating automation transition with no instrument to detect when a stability boundary has been crossed, and no mechanism to enforce course correction before the crossing becomes irreversible.
The most useful thing in the paper is an accident of its own methodology. When Kandasamy ran reinforcement learning agents against his welfare metric, they found an alternate optimum. Without a separately enforced redistribution floor, the agents learned to satisfy the metric by converging on high-automation, low-redistribution equilibria that met the target while violating everything it was designed to protect. A machine optimising a welfare score found the loophole, and the loophole was to automate heavily and redistribute nothing. That is not a modelling artefact. It is a description of the present.
Two Point Zero Five Per Cent
There is a comparison in Kandasamy's data that reframes the debate, and it has nothing to do with how fast AI is advancing.
Denmark and Sweden sustain the lowest inequality in the OECD despite some of the highest robot densities in the world. Their active labour market spending runs at 2.05 and 1.27 per cent of GDP respectively. The equivalent American figure is 0.10 per cent. An order of magnitude. Two orders, against Denmark.
The variable separating these outcomes is not the pace of automation. Denmark automates more than the United States and produces less social damage. What separates them is whether absorption capacity is pre-positioned before displacement occurs rather than improvised afterwards.
Set that against Oracle's income statement. Capital expenditure of 55.7 billion dollars. Severance of 1.838 billion. Reported severance terms of four weeks of base pay plus one week per year of tenure, capped at 26 weeks. Against an American active labour market budget of one tenth of one per cent of GDP, and no obligation on any firm to contribute to reabsorbing the workers it releases.
The transition is not unaffordable. It is unfunded, which is a different thing, and a choice.
Falk and Tsoukalas explain why it will stay unfunded: no firm can unilaterally bear a cost its competitors avoid, and no voluntary agreement among firms is self-enforcing. Kandasamy explains why no regulator will impose it: the instruments govern models, not economies. Rohde explains why the bill arrives late enough that nobody connects it to the decision: masking precedes erosion by years.
Three papers, three disciplines, one conclusion. There is no actor in the current system whose job it is to notice.
The Question The Email Does Not Answer
So what does it mean when a company can generate 8 to 10 billion dollars for AI investment by dismissing tens of thousands of people through a message sent before dawn?
First, that the number is smaller and more contested than the headlines suggest. Oracle's audited filings show 21,000 net, not 30,000, and the company has never confirmed a figure. TD Cowen's estimate was a projection, and the 8 to 10 billion dollars a hypothetical yield on a hypothetical cut.
Second, that the mechanism is real regardless. Capital expenditure rose 34.5 billion dollars. Severance rose 1.46 billion. Headcount fell 21,000. Oracle's own 10-K states that AI adoption has resulted in reductions to its workforce. Those facts sit in one audited document and require not a single anonymous source.
Third, that the thirteen-fold increase tells us less about AI's capabilities than about who has been handed the bill. Andy Challenger says outright that naming AI wins over investors, and in August his own tracker knocked AI off the top of its list of reasons for the first time since February. The New York Fed finds four per cent of service firms reporting AI layoffs and thirteen per cent hiring more people because of it. Stanford finds no economy-wide displacement, and also a 19 per cent employment gap for 22 to 25 year olds in exposed occupations, widening, operating through doors that never opened rather than doors that closed.
So the answer to who is being asked to pay is not primarily the people who received the six in the morning email, distressing as that was. They at least got severance. The people paying are the ones who will never receive an email at all, because the entry-level post that would have trained them into the expertise Oracle now lists as a risk factor was quietly not created. They do not appear in Challenger's data. They trigger no WARN notices. They do not show up in the Fed's survey, because a firm that never posts a role has not laid anyone off.
That is what externalising displacement costs looks like in practice. Not a dramatic severing, but a slow closing of entrances, invisible to every instrument built to detect the other thing.
Oracle's fiscal second quarter opened a fortnight ago. Business Insider reported that managers had been asked to submit lists before it did. Oracle declined to comment then and has confirmed nothing since: no round, no number, no date.
It did file a 10-Q. In the note on restructuring, the company records that total estimated costs under the 2026 Restructuring Plan stood at up to 2.1 billion dollars as at 31 August, of which 1.971 billion had already been accrued, leaving very little of the original budget unspent. Then one further sentence. Subsequent to 31 August, management supplemented the plan by approximately 700 million dollars to reflect additional actions that we expect to take.
Additional actions that we expect to take. No number of people. No country. No date. Seven hundred million dollars of severance and exit costs, budgeted after the quarter closed and disclosed in the same filing week as a 28.5 billion dollar quarter of capital expenditure and a 664 billion dollar backlog. The arithmetic this article opened with has acquired another instalment, and it is still being recorded in the only place it is ever recorded, which is the notes.
The email, if it came, arrived at six.
Sources and References
- Oracle Corporation, “Form 10-K, fiscal year ended 31 May 2026”, filed 2026. https://www.sec.gov/Archives/edgar/data/0001341439/000119312526277521/orcl-20260531.htm
- Oracle Corporation, “Form 10-Q, quarterly period ended 31 August 2026”, filed September 2026. https://www.sec.gov/Archives/edgar/data/0001341439/000119312526389274/orcl-20260831.htm
- Oracle Corporation, “Q1 fiscal 2027 earnings call transcript”, 10 September 2026. https://www.fool.com/earnings/call-transcripts/2026/09/11/oracle-orcl-q1-2027-earnings-call-transcript/
- Oracle Corporation, “Form 10-K, fiscal year ended 31 May 2025”, filed 2025. https://www.sec.gov/Archives/edgar/data/1341439/000095017025087926/orcl-20250531.htm
- Challenger, Gray & Christmas, “Job Cut Announcement Report, August 2026”, 4 September 2026. https://www.challengergray.com/wp-content/uploads/2026/09/Challenger-Report-August-2026.pdf
- Challenger, Gray & Christmas, “Job Cut Announcement Report, July 2026”, 6 August 2026. https://www.challengergray.com/wp-content/uploads/2026/08/Challenger-Report-July-2026.pdf
- Challenger, Gray & Christmas, “2025 Year-End Report: Highest Q4 Layoffs Since 2008; Lowest YTD Hiring Since 2010”, January 2026. https://www.challengergray.com/blog/2025-year-end-challenger-report-highest-q4-layoffs-since-2008-lowest-ytd-hiring-since-2010/
- Brett Hemenway Falk and Gerry Tsoukalas, “The AI Layoff Trap”, arXiv:2603.20617v3, 3 June 2026. https://arxiv.org/abs/2603.20617
- Sivasathivel Kandasamy, “The Human Utility Factor: A Computable Welfare Metric That Reframes AI Governance as a Constrained Optimisation Problem”, arXiv:2607.26068, 23 June 2026. https://arxiv.org/abs/2607.26068
- Wolfgang Rohde, “Short-Term Gain, Long-Term Fragility: AI Labor Substitution and the Erosion of Sustainable Capability”, arXiv:2605.27399, 23 April 2026. https://arxiv.org/abs/2605.27399
- Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, “No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19%”, Stanford Digital Economy Lab, 12 August 2026. https://digitaleconomy.stanford.edu/news/canariesaug26/
- Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence”, Stanford Digital Economy Lab, August 2026. https://digitaleconomy.stanford.edu/app/uploads/2026/08/Canaries_August2026.pdf
- Jaison R. Abel, Richard Deitz, Natalia Emanuel and Nick Montalbano, “Businesses Are Using AI to Transform Work, Not Cut Jobs”, Liberty Street Economics, Federal Reserve Bank of New York, 1 September 2026. https://libertystreeteconomics.newyorkfed.org/2026/09/businesses-are-using-ai-to-transform-work-not-cut-jobs/
- Kartik B. Athreya, “AI's Impact on Labor and Hiring”, Liberty Street Economics, Federal Reserve Bank of New York, 5 August 2026. https://libertystreeteconomics.newyorkfed.org/2026/08/ais-impact-on-labor-and-hiring/
- Federal Reserve Bank of New York, “Are Businesses Scaling Back Hiring Due to AI?”, Liberty Street Economics, 4 September 2025. https://libertystreeteconomics.newyorkfed.org/2025/09/are-businesses-scaling-back-hiring-due-to-ai/
- Lindsay Clark, “Banker claims Oracle may slash up to 30,000 jobs, sell health unit to pay for AI build-out”, The Register, 29 January 2026. https://www.theregister.com/2026/01/29/oracle_td_cowen_note/
- Gyana Swain, “Oracle may slash up to 30,000 jobs to fund AI data-center expansion as US banks retreat”, CIO, 30 January 2026. https://www.cio.com/article/4125103/oracle-may-slash-up-to-30000-jobs-to-fund-ai-data-center-expansion-as-us-banks-retreat.html
- Ana-Maria Stanciuc, “Oracle is cutting up to 30,000 employees to pay for AI data centres”, The Next Web, 31 March 2026. https://thenextweb.com/news/oracle-layoffs-march-2026
- Cris Tolomia, “Oracle planning new round of layoffs in August 2026”, Quartz, 12 August 2026. https://finance.yahoo.com/technology/ai/articles/oracle-planning-round-layoffs-august-134527039.html
- The Next Web, “Oracle is one notch above junk after S&P downgrade as AI data-centre spending burns through cash”, July 2026. https://thenextweb.com/news/oracle-sp-downgrade-ai-spending-junk-risk
- CNBC, “Block shares soar as much as 24% as company slashes workforce by nearly half”, 26 February 2026. https://www.cnbc.com/2026/02/26/block-laying-off-about-4000-employees-nearly-half-of-its-workforce.html
- Fortune, “Block CEO Jack Dorsey cuts 4,000 jobs and says most companies will reach the same conclusion”, 27 February 2026. https://fortune.com/2026/02/27/block-jack-dorsey-ceo-xyz-stock-square-4000-ai-layoffs/
- International Business Times UK, “Sam Altman: AI Isn't the Real Reason Behind Layoffs”, 2026. https://www.ibtimes.co.uk/ai-role-tech-layoffs-complex-reality-1781174
- Layoffs.fyi, “2025 Tech Layoffs”, 2026. https://layoffs.fyi/2025-layoffs/
- WARNTracker, “Oracle America layoffs: 5,332 workers, 33 WARN notices from Dec 2016 to Jul 2026”, 2026. https://www.warntracker.com/company/oracle-america
Tim Green UK-based Systems Theorist & Independent Technology Writer
Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.
His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.
ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk
Listen to the free weekly SmarterArticles Podcast
