There is a script. You have heard it five hundred times by now.
A company announces a layoff. Sometimes thousands of people, sometimes tens of thousands. The press release uses the words restructuring, agility, and the agentic AI era. The CEO posts a 700-word memo with the phrase “today is a hard day” somewhere in the first paragraph. Analysts nod, the stock dips for a week, and three months later the same company posts the highest revenue quarter in its history while announcing a $135 billion AI capex commitment.
And somewhere in the middle of all of this, a security engineer with twelve years of experience opens their laptop to find their badge has been deactivated and a Workday calendar invite labeled “transition discussion” has appeared on their calendar.
We are told this is what AI displacement looks like. It is not. Or, to be more precise: it is partly what AI displacement looks like, and partly something else entirely, and the something-else part is the part nobody wants to put in a press release because the political optics of that part are bad.
Let me walk you through what is actually happening.
The Headline Numbers
In Q1 2026 alone, roughly 80,000 tech workers were laid off across 95 companies, with researchers attributing about 50% of those cuts directly to AI-related explanations. The full-year 2026 projection is expected to exceed the 124,000 cut across all of 2025. By late April, Meta and Microsoft together announced over 20,000 job cuts in a single week. Block cut close to 40% of its workforce. Atlassian cut 10%. Cloudflare cut 20% — 1,100 people — on May 7 while reporting its best revenue quarter ever.
Every one of those announcements named AI as a contributing factor. Sometimes the headline factor. Sometimes the only factor.
Now here is the number nobody is putting in the headlines.
Approximately 92% of companies that announced AI-driven layoffs in 2024 and 2025 increased their total headcount over the same period. Salesforce announced major layoffs and grew from 72,682 employees at end of 2024 to 76,453 at end of 2025. Amazon laid off thousands while growing from 1,556,000 to 1,576,000. Bloomberg data suggests roughly half of AI-attributed layoffs result in the same roles being rehired offshore or at lower salaries.
The heads are not disappearing. They are moving.
What “AI Did It” Actually Translates To
If you take the public statements at face value, AI agents are doing the work of customer support, junior engineering, content moderation, project management, sales development, and an expanding list of structured workflow roles. Salesforce CEO Marc Benioff stated openly that AI agents reduced his support headcount from 9,000 to 5,000. Coinbase is restructuring into AI-native pods. Block leadership has said AI tools and flatter org structures are fundamentally changing what it means to build and run a company.
Some of this is real. AI tooling has genuinely compressed certain workflows. A senior engineer with Claude Code or Cursor is meaningfully more productive than the same engineer was eighteen months ago. A customer support tier-1 ticket queue that used to need 40 people can now run on 12 plus an LLM. These are not lies. They are not even particularly contested.
But “AI compressed the workflow” is a different claim than “AI replaced the worker.” And when you separate the two claims out, the picture changes.
A 2026 survey of tech and banking professionals found that 52% of firms plan to increase India hiring this year, with 38% saying it directly replaces U.S. roles. At the largest firms, the figure goes higher — up to 93% of top firms are planning India expansion. Google, Amazon, Microsoft, and Meta have all been aggressively scaling Indian engineering teams while announcing U.S. layoffs. Microsoft has been expanding AI research centers in India. Google pledged $10 billion for digital expansion in the region. Meta has been hiring core engineering teams there.
If your team gets cut and you check LinkedIn six months later, do not be surprised when the same job titles are open in Hyderabad and Bangalore. That is not AI displacement. That is offshoring with an AI-shaped fig leaf.
The H-1B Angle Nobody Wants to Touch
Here is the part where everyone gets uncomfortable.
H-1B utilization at major tech firms has remained high through the layoff cycles. The political environment around the H-1B program is unstable — there is real expectation of restrictions or significant program changes in the medium term — and large employers know this. The rational corporate response is to move work to where the workers already are, before the visa pipeline narrows.
That is happening in parallel with AI adoption, not because of it. But “we are repositioning ahead of probable H-1B changes” is not a sentence that can appear in a press release without lighting twelve different political fires. “AI is making your job obsolete” can. So that is the sentence that gets used.
For U.S. workers — including a lot of U.S. citizens in tech and cybersecurity — this creates a perverse outcome. Losing your job to AI feels, in the cultural conversation, like losing your job to the inevitable. The technology is bigger than any one person, the future was always going to look like this, learn to prompt engineer or get left behind. The narrative is structured to make resignation feel reasonable.
Losing your job because your employer decided to move it to a lower-cost market while telling the press AI did it is a different kind of loss, and it generates a different kind of anger, and that is precisely why the public framing avoids it.
The Capex Math Nobody Disputes
Underneath all of this, there is a real financial pressure that makes the layoffs structurally necessary regardless of which euphemism gets attached.
Alphabet, Microsoft, Meta, and Amazon are projected to spend approximately $725 billion combined on AI capex in 2026 — up 77% from 2025. This is real money, front-loaded into data centers, accelerators, power infrastructure, and Nvidia inventory. It has to come from somewhere. Payroll is one of the largest controllable line items on a tech company’s income statement.
In other words: the AI capex bill is being partially financed by laying off the workforce, and the workforce is being told the AI itself is why they are being laid off. The same AI that is being built with the savings from their firing. It is a closed narrative loop with a certain dark elegance to it.
The challenge for laid-off workers is that even when the AI productivity gains do materialize — and some genuinely will — the roles being eliminated are not the roles being created. There are roughly 275,000 AI-specific job openings currently sitting unfilled because the skill profile required is narrow and the laid-off workforce cannot pivot into them at speed. A senior infrastructure engineer or compliance analyst with fifteen years of experience does not become a machine learning research scientist over a weekend. The labor market is bifurcating, not balancing.
And Then There Is the Zscaler Problem
If you want to see the AI narrative reach its most cynical form, look at what happened with Zscaler in August 2025.
CEO Jay Chaudhry made public statements during a Cloud Security Alliance summit and on subsequent earnings calls about Zscaler using “over 500 billion transactions per day and hundreds of trillions of signals every day” to train its AI models. Some of those references included specific mentions of “proprietary logs,” “full URLs,” and “complete logs” being used as training material.
Customers, security researchers, and privacy administrators reacted as you would expect them to react. Our customer data — including potentially URL-level traffic data, which can be deeply identifying — is being used to train your AI models? In a zero-trust product whose entire value proposition rests on data containment?
Zscaler issued a clarifying response through its CISO Sam Curry, stating the company does not use customer data to train AI models, that each tenant boundary is maintained, and that only aggregated metadata without personal or proprietary content is used for training. The legal team did its job. The corporate communications team did its job. The clarification is, on its face, the right answer.
But the underlying question is the one that should follow you out of the room: when a zero-trust security vendor processes hundreds of trillions of signals through its platform and then uses the derived signals to train AI models that the vendor monetizes, is the customer’s role in that arrangement really one of pure data processor and pure data controller? Or has the relationship structurally shifted into something that looks more like a data co-monetization arrangement that the original data processing agreement never contemplated?
GDPR-focused commentators raised exactly this question, noting that Zscaler’s data processing agreements reportedly did not include provisions for AI model training. If a processor uses customer data — even aggregated, even de-identified — for a new purpose that was not part of the original contracted purpose, the processor may have become a controller for that processing, which triggers separate legal basis requirements, separate transparency requirements, and separate customer rights.
The Zscaler situation does not fit cleanly into the layoff story, but it fits perfectly into the broader pattern: AI is the narrative that lets companies do things they otherwise could not do — fire people, repurpose data, restructure obligations — by attaching the friction of those actions to a technology that everyone already accepts as inevitable.
What This Means If You Are Updating Your Resume
If you are a security professional reading this because you were just affected by one of these cuts, or you are trying to read the room before the next round, a few things are worth holding in your head at once.
The AI narrative is partly real and partly cover. Treat it as both. The skills are genuinely shifting — agentic AI, MCP server security, AI governance, model risk, data lineage for AI training pipelines — and the people who develop credible depth in those areas are going to be in a meaningfully better position over the next 24 months. But do not over-index on the AI narrative as the only reason your role might be at risk. The offshoring math, the H-1B repositioning, and the post-pandemic correction tail are all live variables.
The companies cutting hardest are not always the ones in trouble. Cloudflare’s layoff happened during its best revenue quarter ever. Meta’s was announced alongside an 87% increase in AI capex. Strong financials and large layoffs are now coexisting because the layoffs are funding the next strategic posture, not responding to current weakness. This makes layoff risk less about company health and more about company strategy, which means the diligence questions you should be asking before accepting your next role are different than they were five years ago.
Geographic concentration is a risk factor now. If you are in a U.S.-based role that is structurally portable — meaning it does not require physical presence, in-person customer interaction, or U.S.-citizenship-gated clearance work — assume your role can be moved. The roles that are sticking domestically are the ones that cannot be moved: federal-cleared work, in-person client services, regulated functions requiring on-soil presence, sales roles with U.S. enterprise accounts, and incident response functions that benefit from time-zone proximity to the customer.
The cybersecurity field is structurally different from generalized tech. Most of the layoff coverage focuses on engineering, customer support, and corporate functions. Security has its own dynamics. Pure defensive operations, GRC, compliance, and audit work tied to U.S. regulatory frameworks — CMMC, SLCGP, state-level requirements, federal contracting — are less portable. Offensive security, threat intel, and detection engineering have specialization premiums that the AI narrative cannot easily compress. Lean into the work that has structural moats: clearance-gated work, regulatory-anchored work, hands-on-keyboard work that requires institutional context.
The “AI took my job” framing is not yours to accept. If you were laid off in 2026, the actual story of why is some combination of: real AI compression of your workflow, offshoring economics, H-1B-program-uncertainty hedging, capex funding pressure, post-pandemic overhiring correction, and strategic repositioning around the next investor narrative cycle. The proportion varies by company. The framing that you alone are responsible for becoming obsolete because you did not adapt fast enough to AI is a framing that serves your former employer’s PR strategy, not your understanding of what happened. Reject the framing. The story is more complicated than that and you are allowed to say so out loud.
The Honest Version
There is a version of all this that is honest. It sounds something like this:
“We are restructuring because our AI capex bill is enormous and we need to fund it. We are moving roles offshore because the labor economics are better and because the H-1B program may not be reliable in two years. We are using AI agents where they perform adequately, which is in narrower areas than our press release implies. Some of the people we are letting go are people we hired during 2021 through 2022 and probably should not have hired. We do not actually know whether the agentic AI era will work out the way our deck slides suggest. Today is hard for the people leaving and we mean that, but we also understand that ‘AI made this necessary’ is a cleaner story for our shareholders than the multi-variable truth.”
No company is going to say that out loud. It is not really a criticism to point that out — it is just structurally what corporate communications is for. But you, reading this, are not bound by their communications strategy. You are allowed to look at the layoff announcement, look at the offshoring data, look at the capex line item, look at the headcount trajectory, look at the customer-data-training controversies, and assemble your own picture of what is going on.
That picture is the one you should be making career decisions against. Not the press release.
