The AI Boom Is Real. The Layoffs Are Real. They're Not the Same Story.
Tech firms are spending $750B on AI while cutting six-figure layoffs since 2023. A complete, sourced breakdown of who's actually hurt, why, and what's next.
The AI Boom Is Real. The Layoffs Are Real. They're Not the Same Story.
A complete, sourced breakdown of what AI is actually doing to individual workers, companies, and the economy in 2026, and three honest scenarios for where it goes from here.
In 2026, a laid-off tech worker joined a small experimental program called AI Commons, which pays people displaced by AI up to $1,000 a month while it retrains them. He described what job hunting had become: hundreds of applications, most producing nothing at all. "It was exciting to get a rejection," he said, "because a lot of companies were just ghosting me." (Business Insider, via B17 News, May 2026)
A rejection, as an emotional high point. That's where the AI economy actually lives for a lot of tech workers right now, not in the trillion-dollar headlines.
Except the trillion-dollar headlines are also true. That's what makes 2026 genuinely confusing, and it's worth stating plainly before any of the data: two things are happening at the same time, both verified in the same companies' own financial filings, and neither one proves the other. AI investment is real and still accelerating. Tech layoffs are real and now officially blamed on AI more than any other single cause. But the honest link between the two is far messier than "robots took the jobs," the damage isn't spread evenly across tech workers, and almost nobody agrees on what happens next, including the AI companies themselves.
This is the long version. Every number below is traceable to a primary filing, a government or research institution, or original reporting, linked as you go and grouped again at the end.
Part One: The trillion-dollar bet
Start with the money, because it's the easiest part to verify. These aren't estimates: they come straight from the companies' own SEC filings.
| Company | Latest reported spend | Change | 2026 guidance |
|---|---|---|---|
| Microsoft | $115.9B (FY2026 cash capex) | up ~80% YoY | ~$190B |
| Amazon | $54.2B in Q2 2026 alone; $169.0B trailing 12 months | TTM up 64% YoY | ~$220B |
| Alphabet (Google) | $44.9B in Q2 2026 | roughly doubled YoY from $22.4B | up to $205B (raised again after Q2) |
| Meta | $31.1B in Q2 2026 | -- | $130–145B (narrowed upward) |
Microsoft's FY2026 10-K · Amazon Q2 2026 earnings · Alphabet Q2 2026 earnings · Meta Q2 2026 earnings
Add the four companies' guidance together and you land around $750 billion for 2026, a figure independent financial press arrived at separately from the same public numbers. Goldman Sachs puts total global AI investment for 2026 near $1 trillion, arguing the commonly cited $800B figure understates it by omitting non-US and private-equity-backed spending. Morgan Stanley projects the five major cloud providers reaching roughly $1.2 trillion in 2027 and $1.4 trillion by 2028, with returns on that AI capital in the 25 to 50 percent range.
Think of capital expenditure, "capex," as the receipt for the bet: land, chips, buildings, power. It only means something if it pays for itself, so the next question is whether real revenue is showing up. It is. Google Cloud grew 82 percent last quarter. AWS grew 37 percent, its fastest pace in over four years, with a reported $496 billion order backlog. Microsoft's Azure crossed $100 billion in annual revenue for the first time. Nvidia, which makes the chips everyone else is buying, posted $81.6 billion in quarterly revenue, up 85 percent, with its data-center business alone up 92 percent; it simultaneously authorized $80 billion in additional stock buybacks and raised its dividend twenty-five-fold (Nvidia Q1 FY2027 earnings). This is the detail that separates 2026 from a pure speculation story: money is landing on top of the spending, not just going out the door.
The private side is even more dramatic. OpenAI raised $122 billion at an $852 billion valuation (one outlet, The Information, separately reported $730 billion for the same round, and no public document resolves the gap). Anthropic raised $65 billion at a $965 billion valuation, briefly overtaking OpenAI as the world's most valuable AI startup. Both companies have filed confidentially for IPOs. Anthropic's revenue run rate climbed from roughly $20 billion in early 2026 to a reported $47 billion by August, corroborated by four separate outlets. Two of Anthropic's own investors backed that surge up in their own audited numbers: Microsoft booked a $3.2 billion mark-to-market gain, and Amazon booked a $53.4 billion pre-tax gain, which it attributed "primarily" to its Anthropic stake.
But the same filings that prove the boom also show the strain
Building all of this is so expensive that Alphabet posted its first-ever quarter of negative free cash flow: $44.9 billion going out, only $39.1 billion coming in from operations. Amazon's free cash flow over the past year swung to negative $7.6 billion, and Amazon says so itself, attributing it directly to a $66.1 billion year-over-year jump in AI infrastructure spending. Meta's revenue climbed 28 percent to $60.8 billion, and its operating margin compressed from 43 percent to 31 percent at the same time, with net income falling 14 percent and free cash flow reduced to a mere $784 million. Alphabet has started raising outside money for this, $49.6 billion through a combined stock and convertible-preferred issuance, explicitly earmarked in part for AI infrastructure.
Institutional warnings are stacking up alongside the growth, not instead of it. Moody's said the "unprecedented" spending pace threatens the credit quality of Amazon, Meta, and Alphabet. The Bank for International Settlements, the closest thing central banks have to a shared early-warning system, named an AI-spending bust as one of the top three threats to global financial stability in its June 2026 annual report, and found that hyperscalers "have shifted from funding capital expenditure with operating cash flows to issuing debt" (BIS Annual Economic Report 2026). Fortune reported AI-related "hidden borrowing" reaching roughly $1.65 trillion after a near-1,000 percent surge, a figure independently repeated by two other outlets, though nobody discloses exactly what counts as "hidden."
Then there's circularity. Nvidia was reported in talks to guarantee up to $250 billion in financing for a single OpenAI data-center project, on top of already owning equity stakes in both OpenAI and Anthropic. Commentary has coalesced around Nvidia's roughly $750 billion total web of AI deals as the center of that risk. CNBC's Jim Cramer summed up the historical echo bluntly: "what we learned in 2000 is that you don't lend to companies who buy your goods."
One more wrinkle worth sitting with: a meaningful share of the "record profits" driving hyperscaler stock prices right now are unrealized. Alphabet's net income rose 298 percent last quarter largely on a $98 billion "other income" gain, mostly unrealized mark-to-market gains on its private AI stakes. If the private valuations of OpenAI or Anthropic correct downward on the secondary market, a large slice of that reported earnings strength reverses with them. None of these companies are in trouble today. All of them are spending, and now earning, in a way that used to be a warning sign before AI made it the industry's entire personality.
Part Two: How many layoffs, actually?
Now the other half of the story, and here's where the neat narrative starts fraying immediately: nobody agrees on the actual number.
Tech layoffs have topped six figures every single year since 2023: roughly 262,000 in 2023, more than 150,000 across 549 companies in 2024, and 123,941 to 124,201 at 269 companies in 2025, according to the widely-cited tracker Layoffs.fyi. For 2026, estimates range from about 114,000 through the first five months to nearly 154,000 by early July, depending on which tracker you read.
That spread is the finding, not a rounding error. Rival trackers put all of 2025 at roughly 244,000 to 246,000, double Layoffs.fyi's number, because they count different universes: VC-backed and public tech companies only, versus all industries, versus a global count, versus a US-only count. There is no single authoritative "how many tech layoffs" figure. Every headline number is tracker-dependent, and even Layoffs.fyi's own live running total can't be read directly by an automated check; its homepage renders the count client-side from a spreadsheet.
What's not in dispute is the direction inside tech specifically. Through July 2026, tech led every sector with 149,023 job cuts year-to-date, up 67 percent from a year earlier, accounting for 31 percent of all US job cuts announced in 2026 (Challenger, Gray & Christmas, August 2026). AI has been the single most-cited reason for US job cuts, across every industry, for five straight months, named in over 112,000 job-cut announcements this year, nearly a quarter of all cuts. AI's share of monthly cut announcements rose from about 7 percent in January to nearly 40 percent in May, when the tech sector alone shed 38,242 jobs, its worst single month since 2024.
There's a genuine silver lining buried in the same data set. US layoff announcements fell to 33,429 in July 2026, down 27 percent from June and 46 percent from a year earlier, a two-year low, while announced hiring plans rose 25 percent and hit the highest July level on record. Unemployment overall sat at a healthy 4.2 percent. Andy Challenger, whose firm supplies most of the AI-layoff headlines, put it plainly: "while AI is shifting the labor market, it is not dismantling it."
Is AI actually doing this, or is it a convenient excuse?
Here's where the doubt comes from inside the industry, not outside it. Sam Altman, who runs OpenAI, has publicly said there's "some AI washing where people are blaming AI for layoffs that they would otherwise do, and then there's some real displacement by AI." Nearly six in ten companies, in their own survey responses, admit they frame layoffs or hiring slowdowns as AI-driven when the actual reason is financial.
Even the raw attribution estimates disagree by more than double for the same quarter: Nikkei Asia attributed 47.9 percent of Q1 2026 tech layoffs to AI and automation, while a different tracker, RationalFX, put explicit AI attribution at just 20.4 percent for the same period. Gartner studied companies that cut customer-service staff and blamed AI, found no measurable link between those cuts and any improvement in business results, and now predicts that half of them will quietly rehire for the same roles under different titles by 2027. A separate Gartner survey of 321 customer-service leaders found only 20 percent had actually reduced staffing because of AI at all; Gartner's own analyst concluded, "most recent workforce reductions were influenced by broader economic conditions rather than automation alone." Even Challenger's own firm flags that its classification is getting noisier over time, as companies grow legally and reputationally cautious about naming AI in their announcements at all, in either direction.
The "youth unemployment is an AI story," one of the most widely repeated claims in this entire debate, has a serious internal challenger too. Economists at the Federal Reserve Bank of New York found that the huge, four-fold rise in remote work since the pandemic, not generative AI, explains roughly 64 percent of the increase in youth unemployment, and that the trend actually predates AI's rapid diffusion. Critically, they found the young-versus-older employment gap persists even in jobs that can't be done remotely, holding AI exposure constant (Liberty Street Economics, June 2026). Stanford's own researchers, whose data is some of the strongest evidence that AI is hurting young workers specifically, are careful to say the same thing themselves: correlated, not proven, and worth continued monitoring rather than a verdict.
None of this means AI isn't doing real damage. It means "AI took the jobs" is a headline doing the work a much messier sentence should be doing: some of this is AI, some of it is companies using AI as convenient cover for cuts they'd have made anyway, and cleanly separating the two cannot yet be done with the data that exists in public.
Part Three: The individual, and the fault line nobody expected
If you want the one finding from all of this that holds up cleanest, here it is: the damage isn't spread across tech workers. It's concentrated almost entirely on people at the start of their careers, and it cuts by function as much as by age.
Recruiting firm SignalFire's 2026 talent data shows the shape clearly:
| Function | Change at large "Tech Major" companies | Change at early-stage startups |
|---|---|---|
| Overall engineering hiring | −11% | +7% |
| New-grad / entry-level hiring | −65% | −76% |
| Front-end engineering | −25% (steepest of any specialty) | -- |
| Design | −48% | −22% |
| Product management | −39% | -- |
| Marketing | −36% | −18% |
| AI / ML engineering | +39% since ChatGPT's launch | -- |
| Forward-Deployed Engineer roles | +30% since ChatGPT's launch | -- |
(All figures versus a 2019 baseline; SignalFire, State of Tech Talent Report 2026.) It's not that companies stopped hiring. It's that they stopped hiring people to do well-defined, repeatable work, the kind that used to be how you got your foot in the door, and that's exactly the kind of work AI got good at first.
The age gradient underneath that table is stark. Stanford's Digital Economy Lab, using payroll data covering roughly one in six US workers, found employment for 22-to-25-year-olds in the most AI-exposed occupations shrinking 3.8 percent a year as of April 2026, accelerating from a 2.8 percent decline the year before. Workers 35 to 40 in those exact same occupations are growing about 2 percent a year. Workers 31 to 34 are also contracting, at 1.7 percent. Same job title, same industry, opposite trend line, and the only variable that explains it is how many years of experience someone had before AI got good (Fortune, June 2026). The original version of that study found entry-level employment in AI-exposed occupations had declined roughly 16 percent in relative terms by October 2025, up from 13 percent through the previous July, and the effect only became statistically significant starting in 2024, well after AI itself existed but right as capability caught up to routine knowledge work.
That shows up downstream in unemployment for the people who trained specifically for this industry. Recent computer-science and computer-engineering graduates, ages 22 to 27, now have unemployment rates of 7.0 and 7.8 percent, among the highest of any college major tracked, according to New York Fed data (Yale Insights, May 2026). In a single year, the share of companies actively cutting junior and entry-level roles specifically because of automation jumped from 17 percent to 43 percent, per a global CEO survey from consultancy Oliver Wyman.
For anyone who does lose a job, at any age, the landing is measurably harder than in past downturns. Goldman Sachs found displaced workers from tech-disrupted occupations "take approximately one month longer to find a new job and suffer real earnings losses of more than 3 percent upon reemployment," usually because they're pushed into more routine, lower-skill work than they had before, a pattern researchers call occupational downgrading (Yahoo Finance, April 2026).
The lived version of those statistics is blunter. In Seattle, one laid-off systems administrator applied to 180 jobs and landed a single interview in three months. He'd been laid off before, in 2018, and back then a four-month search meant "interviews all the time." Another engineer searched for seven months, gave up, and took a job driving a Microsoft employee shuttle bus. Roughly 9,000 Seattle-area tech jobs disappeared in the year to March 2026, and workers who are still employed increasingly describe "job hugging," staying in roles they'd otherwise leave, simply because there's nowhere safer to go (OPB/KUOW, March 2026).
Pay for the people who kept their jobs has gone almost flat. Tech salaries rose just 1.6 percent in the most recent tracking, the lowest in fifteen years of data from staffing firm Robert Half, down from 2.9 percent in 2024 and 3.5 percent in 2023. Sixty-six percent of employers now cite economic-stability concerns for shrinking salary budgets, versus just 17 percent a year earlier (via IEEE-USA InSight). At the same time, workers with genuine AI skills are commanding a 56 percent wage premium, up from 25 percent the prior year, according to PwC's tracking of nearly a billion job postings globally (PwC AI Jobs Barometer). Flat at the bottom, richer at the top, and the gap widening every quarter.
Sentiment has caught up with the numbers. A Mercer survey of 12,000 C-suite executives, HR leaders, employees, and investors found 99 percent of CEOs expect AI to cause at least some headcount reduction within two years, while the share of employees who say they "feel good at work" fell from 66 percent in 2024 to 44 percent. Only 27 percent of CEOs said AI's return on investment met or exceeded their expectations, down from 38 percent the year before, and roughly a quarter saw no revenue impact from their AI spending at all. A separate Stanford survey found 64 percent of Americans expect AI to mean fewer jobs over the next 20 years.
Not every company is playing this the same way. ServiceNow's CEO Bill McDermott publicly committed to zero layoffs from the company's own AI rollout, retraining affected staff into AI-agent managers through an internal training program. Klarna's much-publicized story is more complicated than the headlines suggested: the company did redeploy engineers and marketers back into customer support after shrinking its human team too far, but Business Insider issued a formal correction stating that CEO Sebastian Siemiatkowski's "went too far" comment was about broad cost-cutting, not AI implementation specifically. The redeployment happened. The "AI quality backlash" causal story is the part the outlet retracted.
Part Four: What CEOs actually say when the cameras are rolling
If the data leaves you unsure whether AI is really behind the cuts, the people signing the layoff notices have started saying so themselves, on the record, in ways they weren't willing to two years ago.
| Company | What happened | Stated reason |
|---|---|---|
| Block | ~40% of global headcount cut, Feb 2026 (10,000+ → under 6,000) | Jack Dorsey: "a significantly smaller team... can do more and do it better." Explicitly not financial; gross profit still growing. |
| Salesforce | Support staff cut 9,000 → ~5,000 | Marc Benioff: "I need less heads." ~50% of support handled by AI agents; support costs down 17%. |
| Amazon | ~30,000 corporate roles cut (two waves, Oct 2025 + Jan 2026), largest reduction in company history | Andy Jassy: AI "will reduce our total corporate workforce." Amazon still added 20,000 more total employees in 2025 than 2024 (warehouse hiring continued). |
| Meta | ~10,200–10,400 cut across three 2026 waves | Mark Zuckerberg: "one or two people are building something in a week that would have previously taken dozens of people months." 2026 capex ($130–145B) runs 4.5–5.4x its compensation bill (~$27B). |
| Oracle | ~21,000 cut (~13% of global headcount), $1.84B restructuring cost | Told the SEC directly: "the adoption and deployment of AI technologies across our operations have resulted... in reductions to our workforce." |
| Cisco | ~4,000 cut (~5%) in May 2026, despite record Q3 revenue of $15.8B (+12% YoY) | CFO: explicitly "not a savings-driven restructure," but "realigning resources around silicon, optics, security and AI." |
| IBM | "Low single-digit percent" of ~270,000 staff, Q4 2025 | Cut in the same quarter CEO Arvind Krishna raised full-year revenue and free-cash-flow guidance, citing business strength. |
| Microsoft | ~4,800 cut July 2026 (~2.1%, mainly sales/Xbox), after ~9,000 in July 2025 | Chief People Officer, in writing: "the roles eliminated today are not being replaced by AI," even as the CFO confirmed headcount would keep shrinking as resources shift to AI infrastructure. |
| Intel | Down to 75,000 by end of 2025 (~22% cut, after 15,000 in 2024) | CEO Lip-Bu Tan's "no more blank checks" cost discipline; predecessor's memo cited high costs and low margins, no AI narrative at all. |
| Dell | ~10% smaller for three straight fiscal years (108,000 → 97,000, $569M severance) | Classic cost-cutting, characterized the same way throughout, no AI framing. |
| Cognizant | Up to 15,000 cut under "Project Leap" | Explicit move to a "leaner, AI-supported delivery model," with India (250,000+ of its 350,000+ staff) bearing the largest share. |
That split, some CEOs leaning hard into the AI story and others actively distancing from it, is itself informative. AI framing sounds forward-looking if you're cutting from strength (Cisco, Meta, Block), and sounds like discipline if you're cutting from weakness (Intel, Dell), but "we overhired and now we're fixing it" sounds bad either way. Visa is the cleanest example of this tension in action: leaked memos about its 2,600 job cuts (about 7 percent of staff) circulated widely as an AI-replaces-humans story, but the actual memo, per the reporting that leaked it, framed AI as one piece of a much broader cost strategy focused on stablecoin and cross-border payments, not a head-for-head swap (IBTimes UK, July 2026).
The economic logic founders point to is a revenue-per-employee gap. AI-native startups are reported at $2 million to $4 million-plus in revenue per employee, versus roughly $300,000 for the average public SaaS company; the design tool Lovable is cited at roughly $2.7 million per employee, and Midjourney at roughly $18 million per employee, both allegedly outpacing Microsoft's and Meta's own per-employee revenue (Forbes, March 2026). That comparison should be read with a flag attached: it rests on self-reported, unaudited revenue and headcount figures at small, fast-moving companies, and the underlying numbers couldn't be independently re-confirmed. Plausible, not proven.
And workers aren't accepting the official framing quietly. More than 4,500 Google employees petitioned CEO Sundar Pichai in July 2026 for guaranteed severance and layoff protections. The Alphabet Workers Union's president, Parul Koul, didn't mince words: "Make no mistake: this is a company that is enjoying massive, unprecedented success... these layoffs and cuts are not difficult decisions, but simply profit being put over the people that make this company run" (Livemint, July 2026). The base-rate numbers back up some of that suspicion: the five biggest tech companies have announced more than 100,000 combined layoffs since 2022, and all five have roughly the same total headcount today that they did back then, because they'd added nearly a million net employees during the pandemic hiring boom first. Only Meta ended up meaningfully smaller than its 2022 peak, and even Meta has grown again since its own 2023 downsizing.
Part Five: Is this actually a bubble?
None of the spending in Part One is fake, and none of the strain underneath it is either. The strongest argument for "this is fine" and the strongest argument for "this is a bubble" use the exact same underlying numbers and reach opposite conclusions. That's worth sitting with before picking a side.
The case for "this is fine" rests on three pillars, laid out by J.P. Morgan Asset Management in October 2025. First, most of the spending had, until recently, come from the companies' own operating cash rather than borrowed money, a real structural difference from the dot-com crash, when telecom firms built fiber-optic networks on debt they couldn't repay. Second, the AI business is monetizing while it builds, not years afterward: OpenAI's revenue went from near-zero to roughly $20 billion within about two years. Third, and most convincing of all, actual usage of the new computing capacity is running near record highs: North American data-center vacancy sat at just 1.6 percent, against roughly 7 percent vacancy in fiber-optic networks at the dot-com peak. People aren't just building AI infrastructure. They're using nearly all of it.
The case for "this is a bubble" doesn't dispute any of that. It disputes what happens next, and it has gotten measurably stronger since that October 2025 rebuttal was written. Four things carry it.
First, cash flow has already inverted at two of the four largest spenders, as detailed in Part One: Alphabet's negative free cash flow, Amazon's negative $7.6 billion trailing-year figure, Meta's shrinking margins. Second, institutional alarm has moved from commentary to something closer to an official warning: the BIS's finding that hyperscalers are shifting toward debt is a direct contradiction of pillar one of the bull case above. Third, the circular-financing problem detailed earlier, Nvidia's $750 billion web of AI deals and its reported willingness to guarantee financing for the very companies buying its chips, is precisely the structure that broke in 2000, per Cramer's blunt comparison.
Fourth, and most fundamental, the return-on-investment evidence is genuinely weak so far. MIT researchers found that 95 percent of companies running generative-AI pilots report no measurable financial return at all (Fortune, August 2025). Gartner found no correlation between AI-driven workforce cuts and improved business results, and separately predicts that more than 40 percent of "agentic AI" projects will be canceled by the end of 2027 over cost, unclear value, or inadequate risk controls. Economist Torsten Sløk at Apollo argues the ten largest S&P 500 companies are now more overvalued than their 1990s counterparts were, pointing out that the "Magnificent Seven" tech giants' profit margins rose from roughly 15 percent to 25 percent between early 2023 and early 2026, while the other 493 companies in the index stayed flat near 10 percent, a gap he warns could trigger a "painful repricing." Investor Michael Burry, who famously called the 2008 housing crash, had roughly 80 percent of his portfolio in bets against Nvidia and Palantir by May 2026, explicitly invoking "the final months of the 1999 to 2000 bubble," and by August 2026 was warning of a possible 1987-style crash. Even Sam Altman himself conceded, back in August 2025, that AI valuations look "insane" and investors are behaving irrationally, while committing to keep spending aggressively anyway.
One more data point favors the bubble case specifically: AI-era capital spending as a share of sales reached 34 percent in 2026, projected to hit 37 percent by 2028, already above the 32 percent peak reached during the dot-com bubble. The one thing separating this from 2000 is that there's real revenue behind today's spending that most companies lacked back then. Whether that's enough revenue is the entire argument.
The part almost nobody talks about: the payoff hasn't shown up yet, anywhere
Underneath both cases sits one more inconvenient, boring fact: none of this AI spending has actually moved the official statistics that measure whether the economy as a whole is getting more productive. US labor productivity grew just 1.4 percent last quarter, a thoroughly ordinary number, with the share of income going to workers falling to its lowest level since 1947 (Bureau of Labor Statistics, August 2026). One widely-cited economic model, run by MIT's Daron Acemoglu, puts AI's total productivity gain over an entire decade at no more than 0.66 percent, revised down to under 0.53 percent once you account for tasks that are hard for AI to learn well (NBER Working Paper 32487). A separate survey of nearly 6,000 executives found that nine in ten report no measurable impact on either employment or productivity over the past three years (NBER Working Paper 34836).
San Francisco Fed president Mary Daly has a useful historical answer for why that isn't necessarily a contradiction. It took nearly a hundred years after electricity was first generated for it to show up as a measurable productivity boost in the economy, because factories had to be entirely redesigned around it, not just plugged in where the old machines used to be. In the mid-1990s, official data still showed no measurable productivity payoff from computers and the internet, even as then-Fed-Chair Alan Greenspan correctly bet on scattered firm-level evidence that a surge was already underway beneath the surface (FRBSF Economic Letter, February 2026). AI could be exactly that kind of technology: real, eventually transformative, and simply too early to see in the statistics. Or the payoff might never arrive at anywhere near the scale the spending currently assumes. Both stories remain alive today, and the same quarterly earnings keep feeding both of them at once.
Where that leaves an honest reader: the boom case rests almost entirely on audited, primary-source numbers, capex, cloud revenue, real utilization rates. The bubble case rests on a mix of equally hard primary numbers (negative free cash flow, the BIS report, Meta's shrinking margins) and softer, genuinely contested ones ($1.65 trillion in "hidden" borrowing, valuation-multiple comparisons, short-seller conviction). Not in dispute: total investment is still rising, every hyperscaler raised guidance rather than cut it this earnings season, and profitability strain, credit warnings, and circular-financing concerns are all building at the exact same time as the growth. That's not a contradiction. It's just what a genuinely two-sided bet looks like while it's still being placed.
Part Six: The next five years, in three versions
Every serious forecaster admits the same thing before making a prediction: nobody agrees on how to even count this. The World Economic Forum counts jobs "displaced." Goldman Sachs counts jobs "exposed." McKinsey counts "hours automated." The IMF counts jobs "impacted," a category that explicitly includes jobs that get upgraded, not only eliminated. Stack those percentages together and you get nonsense, because none of them are measuring the same thing, and no source reconciles the difference. With that caveat locked in place, here's the actual range of credible forecasts, broken out by time horizon.
Year 1–2 (through roughly mid-2028): the disruption stays narrow, not broad
The single hardest data point available, measured rather than projected, is Stanford's finding of a roughly 16 percent relative decline in entry-level employment in the most AI-exposed occupations by October 2025, a trend with no reversal yet visible. That is the near-term base case: a narrowing career on-ramp rather than a broad-based collapse. It's disputed by the same evidence cited above, the New York Fed's remote-work finding, Gartner's rehire prediction, and PwC's own data showing headcount growth at the most AI-exposed companies actually outpacing the least-exposed ones, prompting PwC to float the idea that AI might currently be "a job expander" rather than a destroyer. The single biggest wildcard for this window is financial, not technological: if the BIS's debt-shift concern or Sløk's "painful repricing" materializes here, the AI capex that currently accounts for roughly two-thirds of headline US GDP growth contracts, and none of the jobs forecasts below model that path at all.
Year 3–4 (roughly 2028–2029): where the real disagreement lives
This is the window where institutional forecasts diverge most sharply, and the disagreement is about direction, not just size.
The optimistic case: the World Economic Forum's Future of Jobs Report 2025 projects 170 million new jobs created and 92 million displaced globally by 2030, a net gain of 78 million, with 41 percent of employers planning workforce reductions as AI automates specific tasks, but 77 percent planning to upskill their existing staff and nearly half planning to move workers from AI-exposed roles into other parts of the business (summarized via Poozle). Goldman Sachs projects generative AI could raise global GDP by roughly 7 percent, nearly $7 trillion, over a decade, grounding its optimism partly in the observation that 60 percent of today's US workers hold jobs that didn't exist in 1940. One important caveat: that WEF report is from January 2025 and hasn't been updated since, despite 18 months of further AI capability gains and the entire 2026 layoff wave. Whether the net-positive 78 million figure still holds is genuinely unknown.
The large-scale-transition case: McKinsey Global Institute projects up to 30 percent of hours currently worked across the US economy could be automated by 2030, requiring an additional 12 million occupational transitions, with lower-wage workers up to 14 times more likely than higher earners to need to switch occupations entirely, and women 1.5 times more likely than men. McKinsey separately estimates generative AI could add $2.6 to $4.4 trillion in value annually across the dozens of business use cases it analyzed. The IMF frames it starkly: 40 percent of jobs globally will be "impacted," upgraded, eliminated, or transformed, rising to 60 percent in advanced economies, with Managing Director Kristalina Georgieva calling it "like a tsunami hitting the labour market" (The Guardian, January 2026).
The distributional case, which cuts across both: Georgieva separately describes an "accordion of opportunities that is open to some and not others," widening the gap between winners and losers while squeezing the middle. One silver lining she points to: local spending by AI-enhanced high earners could spill over into low-wage service work, citing the pattern that every new local tech job has historically generated roughly 4.4 additional local service jobs (Fortune, January 2026). The base rate underneath all of this is sobering, though: economists Daron Acemoglu and Pascual Restrepo find that 50 to 70 percent of the change in the entire US wage structure over the past four decades traces back to relative wage declines for routine-task workers in rapidly automating industries, and a 2026 MIT study attributes 52 percent of the growth in US income inequality between 1980 and 2016 specifically to automation, with about 10 percentage points of that coming from firms using automation to displace workers who had been earning a wage premium (MIT News, May 2026). The most plausible outcome for 2028 to 2029 isn't mass unemployment. It's a deepening of a wage-inequality trend that was already decades old before AI arrived.
Also plausible in this window: a partial deflation of the "AI did it" narrative itself. Gartner's 2027 rehiring prediction lands here, as does its forecast that 40 percent of agentic AI projects get canceled. The attribution problems documented in Part Two, the Nikkei-versus-RationalFX gap, the six-in-ten companies admitting to AI-framing, Altman's own "AI washing" quote, all point the same direction. A world where 2028's layoff data reads as less AI-driven than 2026's, simply because companies get more careful about the label rather than because the underlying automation stopped, is entirely consistent with everything measured so far.
Year 5 (roughly 2030–2031): the AI industry now plans for its own worst case
The most notable shift in the forecasting landscape isn't a number. It's that the AI industry itself now scenario-plans for outcomes it dismissed outright two years earlier. Anthropic's June 2026 "Policy on the AI Exponential" framework explicitly sets out recommendations for three levels of labor disruption: "a world of roughly 5 percent unemployment, one of 10 percent unemployment, and one of unprecedented unemployment," backing the most severe scenario with $350 million in commitments toward wage insurance and worker-transition research (Anthropic, June 2026). Anthropic's own CEO, Dario Amodei, has warned that AI could eliminate up to half of all entry-level white-collar jobs within five years and push unemployment to 10 to 20 percent, a materially more alarmed posture than the "AI augments, it doesn't replace" consensus that held across the industry as recently as 2023 to 2025.
The counterweight: no independent economic institution has endorsed numbers anywhere near that severe. The WEF's 2030 endpoint is still net job creation. Goldman's story is GDP expansion alongside historically-precedented churn. The IMF's 40/60 percent figures explicitly include upgraded and transformed roles, not only eliminated ones. Amodei's 10-to-20-percent projection sits outside the range any of those institutions has published, and it comes from a company with a direct financial incentive to keep advancing the technology regardless. Read it as a frontier-lab scenario, not a consensus forecast.
Three futures are genuinely live at the five-year mark, and nothing in the evidence currently distinguishes between them:
| Scenario | What it looks like | What supports it |
|---|---|---|
| A: the lag resolves favorably | Productivity statistics finally catch up, the way they eventually did with electricity and with computers in the 1990s. The entry-level slump turns out to be a hiring-freeze artifact of a slow-moving job market, not a permanent structural shift. | Audited revenue growth is real and accelerating. Data-center usage is near record highs. Morgan Stanley projects 25–50% returns on AI capital. |
| B: bifurcation without collapse | Total employment holds roughly steady, but the mix shifts permanently. AI/ML engineering keeps growing; design, marketing, product management, and front-end engineering keep shrinking. Management layers flatten further (already 12 engineers per manager at large tech firms versus 10 in 2019), and the entry-level bottleneck becomes permanent rather than temporary. | This is the straight-line extension of everything measured today. It requires the fewest new assumptions of the three. |
| C: the financial correction arrives first | A repricing in AI valuations hits before the productivity payoff does. The immediate labor-market effect is more tech layoffs, not fewer, but concentrated in AI infrastructure and its supply chain rather than the white-collar occupations everyone is currently worried about. | AI-era capex-to-sales already exceeds the dot-com peak. Two of the four largest AI spenders already post negative free cash flow. |
None of the institutions modeling Scenarios A or B has built Scenario C into their forecast. That, on its own, is worth remembering the next time a headline treats any five-year jobs number as settled.
One last data point deserves weight precisely because it comes from inside the industry rather than outside it. Anthropic's own Economic Index surveyed roughly 9,700 active Claude users in mid-2026 and validated their self-reported estimates against actual usage logs. Roughly half of respondents believe AI can already handle more than 50 percent of their job's tasks. Four percent said AI could already do their entire job, today. More than 35 percent expect AI to handle most or nearly all of their tasks within twelve months, a pattern the researchers called "strikingly uniform" across experience levels and professions. If that self-assessment is even roughly calibrated, the five-year question stops being about AI's raw capability and becomes about how fast organizations can actually restructure around capability they already have, which is precisely the redesign lag that Daly's electrification comparison predicts will take far longer than anyone currently expects.
Part Seven: Nobody's actually in charge of fixing this
Here's the strangest detail in the entire picture, and it gets the least attention of anything above: the United States, where almost all of this disruption is concentrated, currently has no federal policy for it at all.
The last major US program built for exactly this kind of worker displacement, Trade Adjustment Assistance, entered phased termination on July 1, 2022, and has never been reauthorized. It was itself modest relative to the disruption it addressed, covering only about 160,000 workers a year against roughly 500,000 annual manufacturing job losses in the 2000s, with only about 40,000 workers a year actually entering retraining. The current AI shock is hitting a workforce safety net that was built for an earlier, narrower kind of disruption, and then dismantled right before this one arrived.
Congress hasn't closed the gap. A bipartisan Senate bill that would merely require companies to report AI-related layoffs to the federal government has sat without a single committee vote since November 2025. There is no official federal statistic for AI-driven layoffs at all; every number in this article ultimately traces back to a private outplacement firm or, increasingly, to an AI company's own internal usage data.
The White House's energy has gone toward preempting state action rather than replacing it. An executive order signed in December 2025, titled "Eliminating State Law Obstruction of National Artificial Intelligence Policy," enables federal lawsuits against state AI laws and threatens funding cuts to non-compliant states, framed around competing with China. As of mid-2026 reporting, no such lawsuit had actually been filed, and states introduced more AI bills in 2026 than in 2025, including from Republicans.
States are legislating anyway, and one is measuring. California's governor signed an executive order creating the country's first state-level tracker for AI-related job losses, launched in June 2026 and updated monthly; its early data shows no statewide AI-driven unemployment spike, but does show localized risk concentrated among Bay Area tech workers in highly AI-exposed, degree-holding roles, with the state's own researchers cautioning that the measure "captures task exposure, not necessarily adoption or displacement" (Yahoo Finance, July 2026). Illinois signed an AI Safety Measures Act, effective January 2027, requiring annual third-party audits of large AI developers (JD Supra, July 2026). California, Colorado, Connecticut, Illinois, and New York are all advancing similar measures.
Into the federal vacuum has stepped a coalition funded largely by the same companies doing the cutting. RAISE US launched in June 2026 with over $500 million secured toward a $1 billion goal to retrain displaced American workers, anchored by Amazon, Anthropic, Microsoft, and the OpenAI Foundation, alongside Bank of America, IBM, Cisco, and others, with the AFL-CIO's president on its board. Whether a billion-dollar private fund can outperform the enrollment-and-placement gap that made Trade Adjustment Assistance a cautionary tale in the first place is untested. For contrast, the Labor Department's own headline 2026 AI-literacy offering was reportedly a text-message course delivered via SMS in partnership with a private startup.
A cluster of competing redistribution ideas emerged within about ten weeks of each other, and the most aggressive proposals came from the AI companies themselves. OpenAI published a policy blueprint proposing robot taxes on automated labor, a public wealth fund modeled on Alaska's oil-revenue dividend, and government-backed pilots of a 32-hour workweek at full pay. Anthropic's framework proposes a graduated path from tax incentives up to a universal basic income funded by a 3 percent "token tax" on AI companies' own revenue, a policy Amodei has acknowledged "would work against his own financial interest" but still calls reasonable. On the legislative side, Senator Bernie Sanders introduced a bill proposing a 50 percent federal equity stake in major AI companies; Senator Elizabeth Warren proposed a wealth tax on AI billionaires plus an energy-usage excise tax on data centers.
Universal basic income has moved from a fringe idea to mainstream discourse without a single government actually enacting it. Elon Musk continues advocating what he calls "universal high income." Economist Nouriel Roubini argues UBI has become effectively "inevitable" given how much of the workforce AI and robotics are likely to displace over the next two decades, framing the real choice as post-hoc redistribution versus governments taking direct ownership stakes in AI companies upfront, "some form of socialism," in his words (Financial Express, July 2026).
The world is not responding the same way
The regional differences aren't matters of degree. They're different in kind. The European Union has a fixed statutory calendar and just relaxed the part that matters most to workers: its AI Act's "Digital Omnibus" delayed compliance requirements for high-risk AI systems, a category that explicitly includes employment-related AI, from August 2026 to as late as August 2028, even as narrower chatbot-disclosure rules did take effect on schedule. The practical result is that EU workers facing AI-driven employment decisions have transparency rules today but no binding compliance regime until at least late 2027.
China is handling this through labor courts rather than new statute. At least three precedent-setting rulings have sided with workers whose jobs were eliminated by AI, with one Hangzhou court finding that AI-driven cost-cutting alone doesn't justify a termination if the employer failed to properly accommodate the affected worker, a ruling given special designation to guide future cases (The Star, May 2026). This unfolds against roughly 17 percent youth unemployment and a national five-year plan explicitly calling for aggressive AI adoption throughout the economy, a genuine live tension between promoting the technology and preventing the unemployment it might cause.
The United Kingdom changed prime ministers in the middle of this debate, and the new government announced a youth technical-education overhaul starting at age 14, explicitly framed as a response to AI reshaping the job market, against a backdrop of the lowest mid-year graduate hiring since 2020 and record demand for AI skills in job postings (Reuters/Indeed, August 2026). Singapore is reframing rather than reacting: a government minister argued publicly that the "centre of gravity" of skills policy must shift from schools to continuous in-employment reskilling, since job-hopping and role changes are becoming the norm rather than the exception (Channel NewsAsia, August 2026).
India is treating this primarily as opportunity rather than threat: its IT sector added roughly 135,000 net jobs in the most recent fiscal year despite AI-related fears, remaining a net hirer overall, with roughly 350,000 AI-related job openings logged in a single 90-day window (The Hans India, July 2026).
One caution runs through the entire policy debate, and it's worth ending on: multiple sources, including a major gaming-industry CEO and California's own state researchers, flag that the AI-attribution figures driving policy urgency in the first place may themselves be inflated. Policy built on an inflated number risks solving the wrong problem. But policy that waits for a clean number, one that Challenger's own analysts warn may never arrive as companies grow more careful about their language, risks solving nothing at all.
What we genuinely still don't know
This isn't a small list, and pretending otherwise would undersell how unsettled this entire topic is, even with every figure above checked against a primary source.
- How many total layoffs, really? It depends entirely on which tracker you trust, and they disagree by more than 100,000 jobs for the same year.
- Is AI actually the cause? Not cleanly. Two credible attribution estimates disagree by more than double for the same quarter, and the AI industry's own leadership uses the phrase "AI washing" to describe part of the phenomenon.
- Is the entry-level slump permanent or temporary? Genuinely unknown, and it's the single biggest factor deciding whether Scenario A or Scenario B above turns out to be right.
- Is this being financed with cash or debt? It was mostly cash a year ago. The Bank for International Settlements says the mix has since shifted toward debt. How far, nobody has fully measured yet.
- What is OpenAI actually worth? Even reporters covering the same funding round disagree by $122 billion.
- Does the WEF's headline "net +78 million jobs" figure still hold? It's from January 2025. There has been no update since, despite 18 months of further AI capability gains and an entire layoff wave.
- Will a financial correction reorder all of this before the labor-market effects even finish playing out? None of the major jobs forecasts, WEF, McKinsey, or Goldman, model that path at all. It's the largest variable nobody has priced in.
So what do you actually do with all of this
Go back to the man getting excited about a rejection email. His experience and Jack Dorsey's shareholder letter are both accurate descriptions of 2026. They're just true for different people, standing on different rungs of the same ladder, and the entire AI economy right now is really a story about which rung you happen to be standing on.
If you're already established in a well-defined, AI-adjacent skill, this is probably the best labor market you will ever see: wages up 56 percent, demand outstripping supply, companies publicly competing for your attention. If you're trying to get your first foothold, or you're a generalist whose job used to be doing a lot of routine things reasonably well, this is close to the worst market in over a decade, and the usual advice, work harder, apply to more places, doesn't fix a market where the entry point itself has narrowed. Neither the trillion-dollar spending numbers nor the layoff numbers tell you which of those two people you're going to be. Only where you're already standing does.
The technology isn't going to slow down, and for now, neither is the spending behind it. What's still genuinely undecided, the part nobody, not the economists, not the CEOs, not even the AI companies writing their own worst-case scenarios, has settled yet, is whether the ladder grows a new bottom rung once the current one finishes disappearing. That's the number worth watching over the next five years. Not the one with a dollar sign in front of it.
Sources
Company filings and earnings Microsoft FY2026 Form 10-K · Alphabet Q2 2026 earnings · Amazon Q2 2026 earnings · Meta Q2 2026 earnings · Nvidia Q1 FY2027 earnings
Official data and institutions U.S. Bureau of Labor Statistics, Productivity and Costs · Bank for International Settlements, Annual Economic Report 2026 · Federal Reserve Bank of New York, Liberty Street Economics · Federal Reserve Bank of San Francisco, Economic Letter · California AI job-loss tracker, via Yahoo Finance · Illinois AI Safety Measures Act, via JD Supra
Research Stanford Digital Economy Lab · SignalFire State of Tech Talent 2026 · J.P. Morgan Asset Management on AI-deal circularity · Gartner: half of AI-driven cuts will reverse by 2027 · PwC Global AI Jobs Barometer · Anthropic, Policy on the AI Exponential · NBER Working Paper 32487 (Acemoglu, AI macroeconomics) · NBER Working Paper 34836 (Yotzov et al., executive survey) · MIT News on automation and wage inequality · Epoch AI on AI's share of US GDP · Yale Insights on entry-level job destruction
Reporting Challenger, Gray & Christmas, August 2026 · Fortune: Stanford entry-level data · Fortune: MIT NANDA report · Fortune: Georgieva on AI's spillover effects · The Guardian: IMF on the AI "tsunami" · Yahoo Finance: Goldman Sachs on displaced workers · OPB/KUOW: Seattle's laid-off tech workforce · Business Insider / B17 News: AI Commons UBI pilot · Livemint: Google employee petition · IBTimes UK: Visa memo · Forbes: AI-native firms and revenue per employee · The Star (Malaysia): China's AI labor tension · Reuters/Indeed via Global Banking & Finance: UK hiring and AI skills · Channel NewsAsia: Singapore's reskilling shift · The Hans India: India's IT sector hiring · Financial Express: Roubini on UBI · IEEE-USA InSight: tech salary trends · World Economic Forum's 78-million-jobs figure, summarized
Also drawn on: McKinsey & Company / McKinsey Global Institute ("The economic potential of generative AI"; "Generative AI and the future of work in America") and The Hill's coverage of OpenAI's policy proposals, both cited by name; their pages could not be independently re-fetched at publication time (both are known to block automated access), so no direct link is given for either.
This article draws on a multi-source research pass completed August 2026, cross-checking every statistic against a primary filing or original reporting where available, with contested or unverified figures explicitly flagged rather than presented as settled.