The Cuts Are the Proof

Oracle's SEC filing for fiscal 2026 runs to hundreds of pages, but the pattern that now governs the tech industry fits in two numbers on the same document: 21,000 jobs eliminated, explicitly attributed to AI adoption, and negative free cash flow of $23.7 billion against record revenue of $67.4 billion ^[1]. The company spent $70 billion on AI data centers for OpenAI and Meta while paying $1.84 billion in severance, up from $374 million the year before ^[1]. The filing does not hide the contradiction. It states both figures plainly, as if they belong together. And in the logic that now drives the industry's largest companies, they do. The loop is straightforward. Bridgewater estimates major tech companies will invest $650 billion in AI infrastructure during 2026 alone ^[2]. In June, investors erased more than $500 billion from the largest tech firms in a single sell-off; analysts attributed it to the absence of clear returns on AI spending ^[3]. Michael Burry and Paul Tudor Jones each warned, independently, that the AI rally mirrors the late-stage dot-com bubble, with valuations driven by momentum rather than business fundamentals ^[4]. Ray Dalio put the same point more clinically.

All great technology changes produce bubbles. — Ray Dalio

The pressure to show returns is not theoretical. It lands on quarterly earnings calls, and companies that have spent hundreds of billions on AI infrastructure cannot point to hundreds of billions in AI revenue. So they point to savings instead. The fastest way to produce savings is to eliminate the workforce. The cuts are large and accelerating. Over 165,000 tech jobs were eliminated in early 2026, with projections exceeding 300,000 for the year ^[5]. Oracle led with more than 25,000. Amazon cut 16,000, Microsoft 9,000 ^[5]. Meta eliminated roughly 8,000 roles in May, reassigned 7,000 to new AI-focused organizations, and left 6,000 open positions unfilled ^[6]. Snap cut 1,000 employees, 16% of its workforce, and closed more than 300 open roles ^[7]. Block shed nearly 40% of its headcount ^[8]. ClickUp laid off 22% of staff with the stated goal of becoming an organization where AI agents outnumber human employees three to one ^[9]. Snowflake replaced its entire technical writing department with an AI platform ^[8]. And in the most telling cut of all, Amazon eliminated staff within its own Artificial General Intelligence group in July — the researchers who built its frontier AI capabilities, let go to sharpen focus on initiatives that matter most for customers ^[10]. The CEOs who ordered these cuts speak with striking certainty. Block's Jack Dorsey said a significantly smaller team using the tools his company is building can do more and do it better.

Intelligence tools have changed what it means to build and run a company… A significantly smaller team, using the tools we're building, can do more and do it better. — Jack Dorsey

Amazon's Andy Jassy told investors that as generative AI agents roll out, the company expects them to reduce its total workforce in the next few years ^[11]. ClickUp's Zeb Evans put it bluntly.

the best engineers are not writing code any more. — Zeb Evans

Meta went further than words: it mandated non-optional transfers of its top software engineers into an Applied AI unit where AI agents will autonomously write code while human staff serve as monitors. Maher Saba, the executive overseeing the move, told them the transfers were not optional ^[12]. The evidence beneath their claims is thinner and more contested than their certainty implies. Princeton computer scientist Sayash Kapoor examined cases where companies cited AI as the reason for layoffs and found that in nine out of ten, they do not have an AI application ready to fill those jobs ^[13].

I think a lot of companies when they laid off workers have reached out to this convenient excuse that generative AI and advances in AI systems have been the leading cause for these layoffs. — Sayash Kapoor

Oxford Economics analyzed U.S. job losses and found AI accounted for only 4.5% of the total. Market-condition cuts were four times larger. The firm concluded that companies are not replacing workers with AI on a significant scale and are instead trying to dress up layoffs as a good news story rather than bad news, such as past over-hiring ^[14]. Greyhound Research found fewer than 20% of enterprises have seen measurable profitability impact from AI ^[15]. And at Amazon, employees are gaming the internal AI-adoption metrics the company uses to justify both its infrastructure spending and its workforce cuts: automating non-essential tasks to inflate token consumption scores, a practice employees call tokenmaxxing ^[16]. The counter-evidence is not all on one side. Morgan Stanley found that high-AI-exposure industries contributed 1.7 of 2.4 percentage points of U.S. productivity growth through 2025 while employment remained stable — genuine, measurable gains at the macro level ^[17]. Stanford's Digital Economy Lab documented a 16% decline in entry-level jobs within AI-exposed occupations like software development, a real and concentrated displacement ^[18]. Anthropic CEO Dario Amodei, whose company builds the AI driving the contraction, warned the impact on jobs would be unusually painful and far larger than most governments are prepared for, and said he can already see within his own company that on the junior and intermediate end they need fewer people, not more ^[19]. Google DeepMind's Demis Hassabis confirmed the beginnings of an impact at the junior level ^[20]. The displacement is not imaginary. It is concentrated, uneven, and partly real. But partly real is not the same as proven. The industry has built a feedback loop it cannot audit. Companies spend hundreds of billions on AI infrastructure. Investors demand returns. Companies cut workers and book the savings as evidence that the spending is working. Whether AI has actually absorbed the work those workers did — whether the productivity gain is genuine or an artifact of headcount reduction — is a question the loop has no mechanism to answer. The only way to find out is to fire the workers and see what breaks. By the time anything breaks, the workers are gone and the savings are already presented to investors as the return. The contraction is happening. Some of it reflects real AI capability, some of it reflects post-pandemic over-hiring correction dressed in AI language, and some of it is a bet on capabilities that do not yet exist. The industry cannot tell you which portion is which, and it is not waiting to find out.